Beruflich Dokumente
Kultur Dokumente
Day: 1
Problem 1. Let ABC be a triangle with circumcentre O. The points D, E and F lie in the interiors of the
sides BC , CA and AB respectively, such that DE is perpendicular to CO and DF is perpendicular to BO. (By
interior we mean, for example, that the point D lies on the line BC and D is between B and C on that line.)
Let K be the circumcentre of triangle AF E . Prove that the lines DK and BC are perpendicular.
Problem 2. Let n be a positive integer. Find the greatest possible integer m, in terms of n, with the following
property: a table with m rows and n columns can be lled with real numbers in such a manner that for any
two dierent rows [a1 , a2 , . . . , an ] and [b1 , b2 , . . . , bn ] the following holds:
max(|a1 − b1 |, |a2 − b2 |, . . . , |an − bn |) = 1.
for all x, y ∈ R.
Problem 4. A set A of integers is called sum-full if A ⊆ A + A, i.e. each element a ∈ A is the sum of some
pair of (not necessarily dierent) elements b, c ∈ A. A set A of integers is said to be zero-sum-free if 0 is the
only integer that cannot be expressed as the sum of the elements of a nite nonempty subset of A.
Does there exist a sum-full zero-sum-free set of integers?
Problem 6. There are innitely many people registered on the social network Mugbook. Some pairs of
(dierent) users are registered as friends, but each person has only nitely many friends. Every user has at
least one friend. (Friendship is symmetric; that is, if A is a friend of B , then B is a friend of A.)
Each person is required to designate one of their friends as their best friend. If A designates B as her best
friend, then (unfortunately) it does not follow that B necessarily designates A as her best friend. Someone
designated as a best friend is called a 1-best friend. More generally, if n > 1 is a positive integer, then a user
is an n-best friend provided that they have been designated the best friend of someone who is an (n − 1)-best
friend. Someone who is a k -best friend for every positive integer k is called popular.
(a) Prove that every popular person is the best friend of a popular person.
(b) Show that if people can have innitely many friends, then it is possible that a popular person is not the
best friend of a popular person.
Problem 7. Let ABC be an acute-angled triangle with circumcircle Γ and orthocentre H . Let K be a point
of Γ on the other side of BC from A. Let L be the reection of K in the line AB , and let M be the reection
of K in the line BC . Let E be the second point of intersection of Γ with the circumcircle of triangle BLM .
Show that the lines KH , EM and BC are concurrent. (The orthocentre of a triangle is the point on all three
of its altitudes.)
Problem 8. A word is a nite sequence of letters from some alphabet. A word is repetitive if it is a con-
catenation of at least two identical subwords (for example, ababab and abcabc are repetitive, but ababa and
aabb are not). Prove that if a word has the property that swapping any two adjacent letters makes the word
repetitive, then all its letters are identical. (Note that one may swap two adjacent identical letters, leaving a
word unchanged.)
EGMO 2013
European Girls' Mathematical Olympiad
Problem 1. The side BC of the triangle ABC is extended beyond C to D so that CD = BC. The
side CA is extended beyond A to E so that AE = 2CA.
Prove that, if AD = BE, then the triangle ABC is right-angled.
Problem 2. Determine all integers m for which the m×m square can be dissected into five rectangles,
the side lengths of which are the integers 1, 2, 3, . . . , 10 in some order.
(a) Prove that there exists a set S of 6n pairwise different positive integers, such that the least common
multiple of any two elements of S is no larger than 32n2 .
(b) Prove that every set T of 6n pairwise different positive integers contains two elements the least
common multiple of which is larger than 9n2 .
EGMO 2013
European Girls' Mathematical Olympiad
Problem 4. Find all positive integers a and b for which there are three consecutive integers at which
the polynomial
n5 + a
P (n) =
b
takes integer values.
Problem 5. Let Ω be the circumcircle of the triangle ABC. The circle ω is tangent to the sides
AC and BC, and it is internally tangent to the circle Ω at the point P . A line parallel to AB and
intersecting the interior of triangle ABC is tangent to ω at Q.
Prove that ∠ACP = ∠QCB.
Problem 6. Snow White and the Seven Dwarves are living in their house in the forest. On each of
16 consecutive days, some of the dwarves worked in the diamond mine while the remaining dwarves
collected berries in the forest. No dwarf performed both types of work on the same day. On any two
different (not necessarily consecutive) days, at least three dwarves each performed both types of work.
Further, on the first day, all seven dwarves worked in the diamond mine.
Prove that, on one of these 16 days, all seven dwarves were collecting berries.
Problem 1. Determine all real constants t such that whenever a, b, c are the lengths of the sides of
a triangle, then so are a2 + bct, b2 + cat, c2 + abt.
Problem 2. Let D and E be points in the interiors of sides AB and AC, respectively, of a triangle
ABC, such that DB = BC = CE. Let the lines CD and BE meet at F . Prove that the incentre
I of triangle ABC, the orthocentre H of triangle DEF and the midpoint M of the arc BAC of the
circumcircle of triangle ABC are collinear.
Problem 3. We denote the number of positive divisors of a positive integer m by d(m) and the
number of distinct prime divisors of m by ω(m). Let k be a positive integer. Prove that there exist
infinitely many positive integers n such that ω(n) = k and d(n) does not divide d(a2 + b2 ) for any
positive integers a, b satisfying a + b = n.
Problem 4. Determine all integers n ≥ 2 for which there exist integers x1 , x2 , . . . , xn−1 satisfying
the condition that if 0 < i < n, 0 < j < n, i =
6 j and n divides 2i + j, then xi < xj .
Problem 5. Let n be a positive integer. We have n boxes where each box contains a non-negative
number of pebbles. In each move we are allowed to take two pebbles from a box we choose, throw
away one of the pebbles and put the other pebble in another box we choose. An initial configuration
of pebbles is called solvable if it is possible to reach a configuration with no empty box, in a finite
(possibly zero) number of moves. Determine all initial configurations of pebbles which are not solvable,
but become solvable when an additional pebble is added to a box, no matter which box is chosen.
Day: 1
Problem 1. Let 4ABC be an acute-angled triangle, and let D be the foot of the altitude from C.
The angle bisector of ∠ABC intersects CD at E and meets the circumcircle ω of triangle 4ADE
again at F. If ∠ADF = 45◦ , show that CF is tangent to ω.
Problem 3. Let n, m be integers greater than 1, and let a1 , a2 , . . . , am be positive integers not
greater than nm . Prove that there exist positive integers b1 , b2 , . . . , bm not greater than n, such that
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) < n,
Day: 2
Problem 5. Let m, n be positive integers with m > 1. Anastasia partitions the integers 1, 2, . . . , 2m
into m pairs. Boris then chooses one integer from each pair and finds the sum of these chosen integers.
Prove that Anastasia can select the pairs so that Boris cannot make his sum equal to n.
Problem 6. Let H be the orthocentre and G be the centroid of acute-angled triangle 4ABC with
AB 6= AC. The line AG intersects the circumcircle of 4ABC at A and P. Let P 0 be the reflection of
P in the line BC. Prove that ∠CAB = 60◦ if and only if HG = GP 0 .
Day: 1
Problem 1. Let n be an odd positive integer, and let x1 , x2 , . . ., xn be non-negative real numbers.
Show that
min (x2i + x2i+1 ) ≤ max (2xj xj+1 ),
i=1,...,n j=1,...,n
where xn+1 = x1 .
Problem 2. Let ABCD be a cyclic quadrilateral, and let diagonals AC and BD intersect at X. Let
C1 , D1 and M be the midpoints of segments CX, DX and CD, respectively. Lines AD1 and BC1
intersect at Y , and line M Y intersects diagonals AC and BD at different points E and F , respectively.
Prove that line XY is tangent to the circle through E, F and X.
Problem 3. Let m be a positive integer. Consider a 4m × 4m array of square unit cells. Two
different cells are related to each other if they are in either the same row or in the same column. No
cell is related to itself. Some cells are coloured blue, such that every cell is related to at least two blue
cells. Determine the minimum number of blue cells.
Day: 2
Problem 4. Two circles, ω1 and ω2 , of equal radius intersect at different points X1 and X2 . Consider
a circle ω externally tangent to ω1 at a point T1 , and internally tangent to ω2 at a point T2 . Prove
that lines X1 T1 and X2 T2 intersect at a point lying on ω.
Problem 5. Let k and n be integers such that k ≥ 2 and k ≤ n ≤ 2k − 1. Place rectangular tiles,
each of size 1 × k or k × 1, on an n × n chessboard so that each tile covers exactly k cells, and no two
tiles overlap. Do this until no further tile can be placed in this way. For each such k and n, determine
the minimum number of tiles that such an arrangement may contain.
Problem 6. Let S be the set of all positive integers n such that n4 has a divisor in the range n2 + 1,
n2 + 2, . . ., n2 + 2n. Prove that there are infinitely many elements of S of each of the forms 7m,
7m + 1, 7m + 2, 7m + 5, 7m + 6 and no elements of S of the form 7m + 3 or 7m + 4, where m is an
integer.
2017 Day: 1
Zürich
Problem 1. Let ABCD be a convex quadrilateral with ∠DAB = ∠BCD = 90◦ and ∠ABC >
∠CDA. Let Q and R be points on segments BC and CD, respectively, such that line QR intersects
lines AB and AD at points P and S, respectively. It is given that P Q = RS. Let the midpoint of
BD be M and the midpoint of QR be N . Prove that the points M , N , A and C lie on a circle.
Problem 2. Find the smallest positive integer k for which there exist a colouring of the positive
integers Z>0 with k colours and a function f : Z>0 → Z>0 with the following two properties:
(i) For all positive integers m, n of the same colour, f (m + n) = f (m) + f (n).
In a colouring of Z>0 with k colours, every integer is coloured in exactly one of the k colours. In both
(i) and (ii) the positive integers m, n are not necessarily different.
Problem 3. There are 2017 lines in the plane such that no three of them go through the same
point. Turbo the snail sits on a point on exactly one of the lines and starts sliding along the lines in
the following fashion: she moves on a given line until she reaches an intersection of two lines. At the
intersection, she follows her journey on the other line turning left or right, alternating her choice at
each intersection point she reaches. She can only change direction at an intersection point. Can there
exist a line segment through which she passes in both directions during her journey?
2017 Day: 2
Zürich
Problem 4. Let n ≥ 1 be an integer and let t1 < t2 < . . . < tn be positive integers. In a group of
tn + 1 people, some games of chess are played. Two people can play each other at most once. Prove
that it is possible for the following two conditions to hold at the same time:
(ii) For every i with 1 ≤ i ≤ n, there is someone who has played exactly ti games of chess.
b) Prove that for every odd positive integer m there exists an integer n ≥ 2 such that m belongs
to an expensive n-tuple.
There are exactly n factors in the product on the left hand side.
Problem 6. Let ABC be an acute-angled triangle in which no two sides have the same length. The
reflections of the centroid G and the circumcentre O of ABC in its sides BC, CA, AB are denoted by
G1 , G2 , G3 , and O1 , O2 , O3 , respectively. Show that the circumcircles of the triangles G1 G2 C, G1 G3 B,
G2 G3 A, O1 O2 C, O1 O3 B, O2 O3 A and ABC have a common point.
The centroid of a triangle is the intersection point of the three medians. A median is a line connecting
a vertex of the triangle to the midpoint of the opposite side.
Problem 1. Let ABC be a triangle with CA = CB and ∠ACB = 120◦ , and let M be the midpoint
of AB. Let P be a variable point on the circumcircle of ABC, and let Q be the point on the segment CP
such that QP = 2QC. It is given that the line through P and perpendicular to AB intersects the
line M Q at a unique point N .
Prove that there exists a fixed circle such that N lies on this circle for all possible positions of P .
(a) Prove that every integer x ≥ 2 can be written as the product of one or more elements of A, which
are not necessarily different.
(b) For every integer x ≥ 2, let f (x) denote the minimum integer such that x can be written as the
product of f (x) elements of A, which are not necessarily different.
Prove that there exist infinitely many pairs (x, y) of integers with x ≥ 2, y ≥ 2, and
Problem 3. The n contestants of an EGMO are named C1 , . . . , Cn . After the competition they
queue in front of the restaurant according to the following rules.
• The Jury chooses the initial order of the contestants in the queue.
– If contestant Ci has at least i other contestants in front of her, she pays one euro to the
Jury and moves forward in the queue by exactly i positions.
– If contestant Ci has fewer than i other contestants in front of her, the restaurant opens and
the process ends.
(a) Prove that the process cannot continue indefinitely, regardless of the Jury’s choices.
(b) Determine for every n the maximum number of euros that the Jury can collect by cunningly
choosing the initial order and the sequence of moves.
Problem 5. Let Γ be the circumcircle of triangle ABC. A circle Ω is tangent to the line segment AB
and is tangent to Γ at a point lying on the same side of the line AB as C. The angle bisector of ∠BCA
intersects Ω at two different points P and Q.
Prove that ∠ABP = ∠QBC.
Problem 6.
(a) Prove that for every real number t such that 0 < t < 21 there exists a positive integer n with the
following property: for every set S of n positive integers there exist two different elements x and
y of S, and a non-negative integer m (i.e. m ≥ 0), such that
|x − my| ≤ ty.
1
(b) Determine whether for every real number t such that 0 < t < 2 there exists an infinite set S of
positive integers such that
|x − my| > ty
for every pair of different elements x and y of S and every positive integer m (i.e. m > 0).
Problem 1. Find all triples (a, b, c) of real numbers such that ab + bc + ca = 1 and
a2 b + c = b2 c + a = c2 a + b.
Problem 2. Let n be a positive integer. Dominoes are placed on a 2n × 2n board in such a way
that every cell of the board is adjacent to exactly one cell covered by a domino. For each n, determine
the largest number of dominoes that can be placed in this way.
(A domino is a tile of size 2 × 1 or 1 × 2. Dominoes are placed on the board in such a way that
each domino covers exactly two cells of the board, and dominoes do not overlap. Two cells are said
to be adjacent if they are different and share a common side.)
Problem 3. Let ABC be a triangle such that ∠CAB > ∠ABC, and let I be its incentre. Let D
be the point on segment BC such that ∠CAD = ∠ABC. Let ω be the circle tangent to AC at A
and passing through I. Let X be the second point of intersection of ω and the circumcircle of ABC.
Prove that the angle bisectors of ∠DAB and ∠CXB intersect at a point on line BC.
Problem 4. Let ABC be a triangle with incentre I. The circle through B tangent to AI at I meets
side AB again at P . The circle through C tangent to AI at I meets side AC again at Q. Prove that
P Q is tangent to the incircle of ABC.
Problem 5. Let n ≥ 2 be an integer, and let a1 , a2 , . . . , an be positive integers. Show that there
exist positive integers b1 , b2 , . . . , bn satisfying the following three conditions:
(A) ai ≤ bi for i = 1, 2, . . . , n;
(Here, bxc denotes the integer part of real number x, that is, the largest integer that does not ex-
ceed x.)
Problem 6. On a circle, Alina draws 2019 chords, the endpoints of which are all different. A point
is considered marked if it is either
Alina labels each marked point. Of the 4038 points meeting criterion (i), Alina labels 2019 points with
a 0 and the other 2019 points with a 1. She labels each point meeting criterion (ii) with an arbitrary
integer (not necessarily positive).
Along each chord, Alina considers the segments connecting two consecutive marked points. (A
chord with k marked points has k − 1 such segments.) She labels each such segment in yellow with
the sum of the labels of its two endpoints and in blue with the absolute value of their difference.
Alina finds that the N + 1 yellow labels take each value 0, 1, . . . , N exactly once. Show that at
least one blue label is a multiple of 3.
(A chord is a line segment joining two different points on a circle.)
Problem 1. Let ABC be a triangle with circumcentre O. The points D, E and F lie in the interiors of the
sides BC, CA and AB respectively, such that DE is perpendicular to CO and DF is perpendicular to BO.
(By interior we mean, for example, that the point D lies on the line BC and D is between B and C on that
line.)
Let K be the circumcentre of triangle AF E. Prove that the lines DK and BC are perpendicular.
O
E
F
B C
D
Solution 1 (submitter). Let `C be the tangent at C to the circumcircle of 4ABC. As CO ⊥ `C , the lines
DE and `C are parallel. Now we find that
hence the quadrilateral BDEA is cyclic. Analogously, we find that the quadrilateral CDF A is cyclic. As we
now have ∠CDE = ∠A = ∠F DB, we conclude that the line BC is the external angle bisector of ∠EDF .
Furthermore, ∠EDF = 180◦ − 2∠A. Since K is the circumcentre of 4AEF , ∠F KE = 2∠F AE = 2∠A. So
∠F KE + ∠EDF = 180◦ , hence K lies on the circumcircle of 4DEF . As |KE| = |KF |, we have that K is the
midpoint of the arc EF of this circumcircle. It is well known that this point lies on the internal angle bisector
of ∠EDF . We conclude that DK is the internal angle bisector of ∠EDF . Together with the fact that BC is
the external angle bisector of ∠EDF , this yields that DK ⊥ BC, as desired.
Solution 2 (submitter). As in the previous solution, we show that the quadrilaterals BDEA and CDF A
are both cyclic. Denote by M and L respectively the circumcentres of these quadrilaterals. We will show that
the quadrilateral KLOM is a parallelogram. The lines KL and M O are the perpendicular bisectors of the line
segments AF and AB, respectively. Hence both KL and M O are perpendicular to AB, which yields KL k M O.
In the same way we can show that the lines KM and LO are both perpendicular to AC and hence parallel
as well. We conclude that KLOM is indeed a parallelogram. Now, let K 0 , L0 , O0 and M 0 be the respective
projections of K, L, O and M to BC. We have to show that K 0 = D. As L lies on the perpendicular bisector
of CD, we have that L0 is the midpoint of CD. Similarly, M 0 is the midpoint of BD and O0 is the midpoint
of BC. Now we are going to use directed lengths. Since KLOM is a parallelogram, M 0 K 0 = O0 L0 . As
O0 L0 = O0 C − L0 C = 1
2 · (BC − DC) = 1
2 · BD = M 0 D,
1
Solution 3 (submitter). Denote by `A , `B and `C the tangents at A, B and C to the circumcircle of 4ABC.
Let A0 be the point of intersection of `B and `C and define B 0 and C 0 analogously. As in the first solution,
we find that DE k `C and DF k `B . Now, let Q be the point of intersection of DE and `A and let R be
the point of intersection of DF and `A . We easily find 4AQE ∼ 4AB 0 C. As |B 0 A| = |B 0 C|, we must have
|QA| = |QE|, hence 4AQE is isosceles. Therefore the perpendicular bisector of AE is the internal angle bisector
of ∠EQA = ∠DQR. Analogously, the perpendicular bisector of AF is the internal angle bisector of ∠DRQ.
We conclude that K is the incentre of 4DQR, thus DK is the angle bisector of ∠QDR. Because the sides of
the triangles 4QDR and 4B 0 A0 C 0 are pairwise parallel, the angle bisector DK of ∠QDR is parallel to the
angle bisector of ∠B 0 A0 C 0 . Finally, as the angle bisector of ∠B 0 A0 C 0 is easily seen to be perpendicular to BC
(as it is the perpendicular bisector of this segment), we find that DK ⊥ BC, as desired.
Remark (submitter). The fact that the quadrilateral BDEA is cyclic (which is an essential part of the first
two solutions) can be proven in various ways. Another possibility is as follows. Let P be the midpoint of BC.
Then, as ∠CP O = 90◦ , we have ∠P OC = 90◦ −∠OCP . Let X be the point of intersection of DE and CO, then
we have that ∠CDE = ∠CDX = 90◦ − ∠XCD = 90◦ − ∠OCP . Hence ∠CDE = ∠P OC = 21 ∠BOC = ∠BAC.
From this we can conclude that BDEA is cyclic.
Solution 4 (PSC). This is a simplified variant of Solution 1. ∠COB = 2∠A (angle at centre of circle ABC)
and OB = OC so ∠OBC = ∠BCO = 90◦ − ∠A. Likewise ∠EKF = 2∠A and ∠KF E = ∠F EK = 90◦ − ∠A.
Now because DE ⊥ CO, ∠EDC = 90◦ − ∠DCO = 90◦ − ∠BCO = ∠A and similarly ∠BDF = ∠A, so
∠F DE = 180◦ − 2∠A. So quadrilateral KF DE is cyclic (opposite angles), so (same segment) ∠KDE =
∠KF E = 90◦ − ∠A, so ∠KDC = 90◦ and DK is perpendicular to BC.
Problem 2. Let n be a positive integer. Find the greatest possible integer m, in terms of n, with the following
property: a table with m rows and n columns can be filled with real numbers in such a manner that for any
two different rows [a1 , a2 , . . . , an ] and [b1 , b2 , . . . , bn ] the following holds:
(1) Each graph Gk is bipartite. Indeed, if it contained a cycle of odd length, then the sum of ±1 along this
cycle would need to be equal to 0, which contradicts the length of the cycle being odd.
(2) For every i 6= j, ij ∈ E(Gk ) for some k. This follows directly from the problem statement.
For every graph Gk fix some bipartition (Ak , Bk ) of {1, 2, . . . , m}, i.e., a partition of {1, 2, . . . , m} into two
disjoint sets Ak , Bk such that the edges of Gk traverse only between Ak and Bk . If m > 2n , then there are two
distinct indices i, j such that they belong to exactly the same parts Ak , Bk , that is, i ∈ Ak if and only if j ∈ Ak
for all k = 1, 2, . . . , n. However, this means that the edge ij cannot be present in any of the graphs G1 , G2 ,
. . . , Gn , which contradicts (2). Therefore, m ≤ 2n .
Solution 2 (PSC). In any table with the given property, the least and greatest values in a column cannot
differ by more than 1. Thus, if each value that is neither least nor greatest in its column is changed to be equal
to either the least or the greatest value in its column (arbitrarily), this does not affect any |ai − bi | = 1, nor
does it increase any difference above 1, so the table still has that given property. But after such a change, for
any choice of what the least and greatest values in each column are, there are only two possible choices for each
entry in the table (either the least or the greatest value in its column); that is, only 2n possible distinct rows,
and the given property implies that all rows must be distinct. As in the previous solution, we see that this
number can be achieved.
2
Solution 3 (Coordinators). We prove by induction on n that m ≤ 2n .
First suppose n = 1. If real numbers x and y have |x − y| = 1 then bxc and byc have opposite parities and
hence it is impossible to find three real numbers with all differences 1. Thus m ≤ 2.
Suppose instead n > 1. Let a be the smallest number appearing in the first column of the table; then every
entry in the first column of the table lies in the interval [a, a + 1]. Let A be the collection of rows with first
entry a and B be the collection of rows with first entry in (a, a + 1]. No two rows in A differ by 1 in their first
entries, so if we list the rows in A and delete their first entries we obtain a table satisfying the conditions of
the problem with n replaced by n − 1; thus, by the induction hypothesis, there are at most 2n−1 rows in A.
Similarly, there are at most 2n−1 rows in B. Hence m ≤ 2n−1 + 2n−1 = 2n . As before, this number can be
achieved.
Solution 4 (Coordinators). Consider the rows of the table as points of Rn . As the values in each column
differ by at most 1, these points must lie in some n-dimensional unit cube C. Consider the unit cubes centred
on each of the m points. The conditions of the problem imply that the interiors of these unit cubes are pairwise
disjoint. But now C has volume 1, and each of these cubes intersects C in volume at least 2−n : indeed, if the
unit cube centred on a point of C is divided into 2n cubes of equal size then one of these cubes must lie entirely
within C. Hence m ≤ 2n . As before, this number can be achieved.
Solution 5 (Coordinators). Again consider the rows of the table as points of Rn . The conditions of the
problem imply that these points must all lie in some n-dimensional unit cube C, but no two of the points lie in
any smaller cube. Thus if C is divided into 2n equally-sized subcubes, each of these subcubes contains at most
one row of the table, giving m ≤ 2n . As before, this number can be achieved.
for all x, y ∈ R.
hence f (0) = 0. Now set x = 0 and y = 1 in the given equation and use (1) again:
4 = f (f (1)) = 2f (1),
so f (1) = 2 and therefore also f (2) = f (f (1)) = 4. Finally substitute y = 1 − x in the equation:
As f is injective, from this it follows that f (x) = 2 − 2(1 − x) = 2x. It is easy to see that this function satisfies
the original equation. Hence the only solution is the function defined by f (x) = 2x for all x ∈ R.
f (f (x)) = 4x
As the right hand side is independent of x and f is injective, 2κ − 2x + f (x) is constant, i.e. f (x) = 2x + α.
Substituting this into the original equation, we see that 2x + α is a solution to the original equation if and
only if 4(y 2 + xy + x) + (3 + 2y)α = 4(y 2 + xy + x) + 2yα for all x, y, i.e. if and only if α = 0. Thus the unique
solution to the equation is f (x) = 2x.
3
Problem 4. A set A of integers is called sum-full if A ⊆ A + A, i.e. each element a ∈ A is the sum of some
pair of (not necessarily different) elements b, c ∈ A. A set A of integers is said to be zero-sum-free if 0 is the
only integer that cannot be expressed as the sum of the elements of a finite nonempty subset of A.
Does there exist a sum-full zero-sum-free set of integers?
Remark. The original formulation of this problem had a weaker definition of zero-sum-free that did not
require all nonzero integers to be sums of finite nonempty subsets of A.
Remark (James Aaronson and Adam P Goucher). Let n be a positive integer, and write u = 2n ; we
claim that the set
is a sum-full zero-sum-free set. The proof is similar to that used for the standard examples.
4
European Girls’ Mathematical Olympiad 2012—Day 2 Solutions
Solution 1 (submitter). Rearranging the equation, 2qn(p + 1) = (n + 2)(2pq + p + q + 1). The left hand
side is even, so either n + 2 or p + q + 1 is even, so either p = 2 or q = 2 since p and q are prime, or n is even.
If p = 2, 6qn = (n + 2)(5q + 3), so (q − 3)(n − 10) = 36. Considering the divisors of 36 for which q is
prime, we find the possible solutions (p, q, n) in this case are (2, 5, 28) and (2, 7, 19) (both of which satisfy the
equation).
If q = 2, 4n(p + 1) = (n + 2)(5p + 3), so n = pn + 10p + 6, a contradiction since n < pn, so there is no
solution with q = 2.
Finally, suppose that n = 2k is even. We may suppose also that p and q are odd primes. The equation
becomes 2kq(p + 1) = (k + 1)(2pq + p + q + 1). The left hand side is even and 2pq + p + q + 1 is odd, so k + 1
is even, so k = 2` + 1 is odd. We now have
or equivalently
`q(p + 1) = (` + 1)(pq + p + 1).
Note that q | pq + p + 1 if and only if q | p + 1. Furthermore, because (p, p + 1) = 1 and q is prime,
(p + 1, pq + p + 1) = (p + 1, pq) = (p + 1, q) > 1 if and only if q | p + 1.
Since (`, ` + 1), we see that, if q - p + 1, then ` = pq + p + 1 and ` + 1 = q(p + 1), so q = p + 2
(and (p, p + 2, 2(2p2 + 6p + 3)) satisfies the original equation). In the contrary case, suppose p + 1 = rq, so
`(p + 1) = (` + 1)(p + r), a contradiction since ` < ` + 1 and p + 1 ≤ p + r.
Thus the possible values of q − p are 2, 3 and 5.
1
Solution 3 (Coordinators). Subtract 2 from both sides to get
1 1 4
− = .
p+1 q n+2
From this, since n is positive, we have that q > p + 1. Therefore q and p + 1 are coprime, since q is prime.
Group the terms on the LHS to get
q−p−1 4
= .
q(p + 1) n+2
Now (q, q − p − 1) = (q, p + 1) = 1 and (p + 1, q − p − 1) = (p + 1, q) = 1 so the fraction on the left is in lowest
terms. Therefore the numerator must divide the numerator on the right, which is 4. Since q − p − 1 is positive,
it must be 1, 2 or 4, so that q − p must be 2, 3 or 5. All of these can be attained, by (p, q, n) = (3, 5, 78),
(2, 5, 28) and (2, 7, 19) respectively.
Problem 6. There are infinitely many people registered on the social network Mugbook. Some pairs of
(different) users are registered as friends, but each person has only finitely many friends. Every user has at
least one friend. (Friendship is symmetric; that is, if A is a friend of B, then B is a friend of A.)
Each person is required to designate one of their friends as their best friend. If A designates B as her best
friend, then (unfortunately) it does not follow that B necessarily designates A as her best friend. Someone
designated as a best friend is called a 1-best friend. More generally, if n > 1 is a positive integer, then a user
is an n-best friend provided that they have been designated the best friend of someone who is an (n − 1)-best
friend. Someone who is a k-best friend for every positive integer k is called popular.
(a) Prove that every popular person is the best friend of a popular person.
(b) Show that if people can have infinitely many friends, then it is possible that a popular person is not the
best friend of a popular person.
Solution 1 (submitter, adapted). For any person A, let f 0 (x) = x, let f (A) be A’s best friend, and define
f k+1 (A) = f (f k (A)), so any person who is a k-best friend is f k (A) for some person A; clearly a k-best friend
is also an `-best friend for all ` < k. Let X be a popular person. For each positive integer k, let xk be a person
with f k (xk ) = X. Because X only has finitely many friends, infinitely many of the f k−1 (xk ) (all of whom
designated X as best friend) must be the same person, who must be popular.
If people can have infinitely many friends, consider people Xi for positive integers i and Pi,j for i < j positive
integers. Xi designates Xi+1 as her best friend; Pi,i designates X1 as her best friend; Pi,j designates Pi+1,j as
her best friend if i < j. Then all Xi are popular, but X1 is not the best friend of a popular person.
Solution 2 (submitter, adapted). For any set S of people, let f −1 (S) be the set of people who designated
someone in S as their best friend. Since each person has only finitely many friends, if S is finite then f −1 (S) is
finite.
Let X be a popular person and put V0 = {X} and Vk = f −1 (Vk−1 ). All Vi are finite and (since X is popular)
nonempty.
If any two sets Vi , Vj , with 0 ≤ i < j are not disjoint, define f i (x) for positive integers i as in Solution 1.
It follows ∅ 6= f i (Vi ∩ Vj ) ⊆ f i (Vi ) ∩ f i (Vj ) ⊆ V0 ∩ Vj−i , thus X ∈ Vj−i . But this means that f j−i (X) = X,
therefore f n(j−i) (X) = X. Furthermore, if Y = f j−i−1 (X), then f (Y ) = X and f n(j−i) (Y ) = Y , so X is the
best friend of Y , who is popular.
If all sets Vn are disjoint, by König’s infinity lemma there exists an infinite sequence of (distinct) xi , i ≥ 0,
with xi ∈ Vi and xi = f (xi+1 ) for all i. Now x1 is popular and her best friend is x0 = X.
If people can have infinitely many friends, proceed as in Solution 1.
2
Problem 7. Let ABC be an acute-angled triangle with circumcircle Γ and orthocentre H. Let K be a point
of Γ on the other side of BC from A. Let L be the reflection of K in the line AB, and let M be the reflection
of K in the line BC. Let E be the second point of intersection of Γ with the circumcircle of triangle BLM .
Show that the lines KH, EM and BC are concurrent. (The orthocentre of a triangle is the point on all three
of its altitudes.)
Solution 1 (submitter). Since the quadrilateral BM EL is cyclic, we have ∠BEM = ∠BLM . By construc-
tion, |BK| = |BL| = |BM |, and so (using directed angles)
We see also that ∠BEM = ∠BAH, and so the point N of intersection of EM and AH lies on Γ.
Let X be the point of intersection of KH and BC, and let N 0 be the point of intersection of M X and AH.
Since BC bisects the segment KM by construction, the triangle KXM is isosceles; as AHkM K, HXN 0 is
isosceles. Since AH ⊥ BC, N 0 is the reflection of H in the line BC. It is well known that this reflection
lies on Γ, and so N 0 = N . Thus E, M , N and M , X, N 0 all lie on the same line M N ; that is, EM passes
through X.
E Γ
H
M
X
C
L B
K N = N0
Remark (submitter). The condition that K lies on the circumcircle of ABC is not necessary; indeed, the
solution above does not use it. However, together with the fact that the triangle ABC is acute-angled, this
condition implies that M is in the interior of Γ, which is necessary to avoid dealing with different configurations
including coincident points or the point of concurrence being at infinity.
Solution 2 (PSC). We work with directed angles. Let HK meet BC at X. Let M X meet AH at HA on Γ
(where HA is the reflection of H in BC). Define E 0 to be where HA M meets Γ (again). Our task is to show
that ∠M E 0 B = ∠M LB.
Observe that
Now
3
Problem 8. A word is a finite sequence of letters from some alphabet. A word is repetitive if it is a con-
catenation of at least two identical subwords (for example, ababab and abcabc are repetitive, but ababa and
aabb are not). Prove that if a word has the property that swapping any two adjacent letters makes the word
repetitive, then all its letters are identical. (Note that one may swap two adjacent identical letters, leaving a
word unchanged.)
Solution 1 (submitter). In this and the subsequent solutions we refer to a word with all letters identical as
constant.
Let us consider a nonconstant word W , of length |W | = w, and reach a contradiction. Since the word
W must contain two distinct adjacent letters, be it W = AabB with a 6= b, we may assume B = cC to be
non-empty, and so W = AabcC. By the proper transpositions we get the repetitive words W 0 = AbacC = P w/p ,
of a period P of length p | w, 1 < p < w, and W 00 = AacbC = Qw/q , of a period Q of length q | w, 1 < q < w.
However, if a word U V is repetitive, then the word V U is also repetitive, of a same period length; therefore we
can work in the sequel with the repetitive words W00 = CAbac, of a period P 0 of length p, and W000 = CAacb,
of a period Q0 of length q. The main idea now is that the common prefix of two repetitive words
cannot be too long.
Now, if a word a1 a2 . . . aw = T w/t is repetitive, of a period T of length t | w, 1 ≤ t < w, then the word (and
any subword of it) is t-periodic, i.e. ak = ak+t , for all 1 ≤ k ≤ w − t. Therefore the word CA is both p-periodic
and q-periodic.
We now use the following classical result:
Wilf-Fine Theorem. Let p, q be positive integers, and let N be a word of length n, which is both p-periodic
and q-periodic. If n ≥ p + q − gcd(p, q) then the word N is gcd(p, q)-periodic (but this need not be the case if
instead n ≤ p + q − gcd(p, q) − 1).
By this we need |CA| ≤ p + q − gcd(p, q) − 1 ≤ p + q − 2, hence w ≤ p + q + 1, otherwise W00 and W000 would
be identical, absurd. Since p | w and 1 < p < w, we have 2p ≤ w ≤ p + q + 1, and so p ≤ q + 1; similarly we
have q ≤ p + 1.
If p = q, then |CA| ≤ p + p − gcd(p, p) − 1 = p − 1, so 2p ≤ w ≤ p + 2, implying p ≤ 2. But the three-letter
suffix acb is not periodic (not even for c = a or c = b), thus must be contained in Q0 , forcing q ≥ 3, contradiction.
If p 6= q, then max(p, q) = min(p, q) + 1, so 3 min(p, q) ≤ w ≤ 2 min(p, q) + 2, hence min(p, q) ≤ 2, forcing
min(p, q) = 2 and max(p, q) = 3; by an above observation, we may even say q = 3 and p = 2, leading to c = b. It
follows 6 = 3 min(p, q) ≤ w ≤ 2 min(p, q) + 2 = 6, forcing w = 6. This leads to CA = aba = abb, contradiction.
Solution 2 (submitter). We will take over from the solution above, just before invoking the Wilf-Fine
Theorem, by replacing it with a weaker lemma, also built upon a seminal result of combinatorics on words.
Lemma. Let p, q be positive integers, and let N be a word of length n, which is both p-periodic and q-
periodic. If n ≥ p + q then the word N is gcd(p, q)-periodic.
Proof. Let us first prove that two not-null words U , V commute, i.e. U V = V U , if and only if there exists
a word W with |W | = gcd(|U |, |V |), such that U = W |U |/|W | , V = W |V |/|W | . The “if” part being trivial, we
will prove the “only if” part, by strong induction on |U | + |V |. Indeed, for the base step |U | + |V | = 2 we
have |U | = |V | = 1, and so clearly we can take W = U = V . Now, for |U | + |V | > 2, if |U | = |V | it follows
U = V , and so we can again take W = U = V . If not, assume without loss of generality |U | < |V |; then
V = U V 0 , so U U V 0 = U V 0 U , whence U V 0 = V 0 U . Since |V 0 | < |V |, it follows 2 ≤ |U | + |V 0 | < |U | + |V |,
0
so by the induction hypothesis there exists a suitable word W such that U = W |U |/|W | , V 0 = W |V |/|W | , so
0 0
V = U V 0 = W |U |/|W | W |V |/|W | = W (|U |+|V |)/|W | = W |V |/|W | .
Now, assuming without loss of generality p ≤ q, q = kp+r, we have N = QP S, with |Q| = q, |P | = p. If r = 0
all is clear; otherwise it follows we can write P = U V , Q = V (U V )k , with |V | = r, whence U V = V U , implying
P Q = QP , and so by the above result there will exist a word W of length gcd(p, q) such that P = W p/ gcd(p,q) ,
Q = W q/ gcd(p,q) , therefore N is gcd(p, q)-periodic.
By this we need |CA| ≤ p + q − 1, hence w ≤ p + q + 2, otherwise by the previous lemma W00 and W000 would
be identical, absurd. Since p | w and 1 < p < w, we have 2p ≤ w ≤ p + q + 2, and so p ≤ q + 2; similarly we have
q ≤ p + 2. That implies max(p, q) ≤ min(p, q) + 2. Now, from k max(p, q) = w ≤ p + q + 2 ≤ 2 max(p, q) + 2
we will have (k − 2) max(p, q) ≤ 2; but max(p, q) ≤ 2 is impossible, since the three-letter suffix acb is not
periodic (not even for c = a or c = b), thus must be contained in Q0 , forcing q ≥ 3. Therefore k = 2, and so
w = 2 max(p, q).
4
If max(p, q) = min(p, q), then w = 2p = 2q, for a quick contradiction.
If max(p, q) = min(p, q) + 1, it follows 3 min(p, q) ≤ w = 2 max(p, q) = 2 min(p, q) + 2, hence min(p, q) ≤ 2,
forcing min(p, q) = 2 and max(p, q) = 3; by an above observation, we may even say q = 3 and p = 2, leading to
c = b. It follows w = 2 max(p, q) = 6, leading to CA = aba = abb, contradiction.
If max(p, q) = min(p, q) + 2, it follows 3 min(p, q) ≤ w = 2 max(p, q) = 2 min(p, q) + 4, hence min(p, q) ≤ 4.
From min(p, q) | w = 2 max(p, q) then follows either min(p, q) = 2 and max(p, q) = 4, thus w = 8, clearly
contradictory, or else min(p, q) = 4 and max(p, q) = 6, thus w = 12, which also leads to contradiction, by just
a little deeper analysis.
Solution 3 (PSC). We define the distance between two words of the same length to be the number of positions
in which those two words have different letters. Any two words related by a transposition have distance 0 or 2;
any two words related by a sequence of two transpositions have distance 0, 2, 3 or 4.
Say the period of a repetitive word is the least k such that the word is the concatenation of two or more
identical subwords of length k. We use the following lemma on distances between repetitive words.
Lemma. Consider a pair of distinct, nonconstant repetitive words with periods ga and gb, where (a, b) = 1
and a, b > 1, the first word is made up of kb repetitions of the subword of length ga and the second word is
made up of ka repetitions of the subword of length gb. These two words have distance at least max(ka, kb).
Proof. We may assume k = 1, since the distance between the words is k times the distance between their
initial subwords of length gab. Without loss of generality suppose b > a.
For each positive integer m, look at the subsequence in each word of letters in positions congruent to m
(mod g). Those subsequences (of length ab) have periods dividing a and b respectively. If they are equal, then
they are constant (since each letter is equal to those a and b before and after it, mod ab, and (a, b) = 1).
Because a > 1, there is some m for which the first subsequence is not constant, and so is unequal to the second
subsequence. Restrict attention to those subsequences.
We now have two distinct repetitive words, one (nonconstant) made up of b repetitions of a subword of
length a and one made up of a repetitions of a subword of length b. Looking at the first of those words, for any
1 ≤ t ≤ b consider the letters in positions t, t + b, . . . , t + (a − 1)b. These letters cover every position (mod a);
since the first word is not constant, the letters are not all equal, but the letters in the corresponding positions
in the second word are all equal. At least one of these letters in the first word must change to make them all
equal to those in the corresponding positions in the second word; repeating for each t, at least b letters must
change, so the words have distance at least b.
In the original problem, consider all the words (which we suppose to be repetitive) obtained by a transposition
of two adjacent letters from the original nonconstant word; say that word has length n. Suppose those words
include two distinct words with periods n/a and n/b; those words have distance at most 4. If a > 4 or b > 4,
we have a contradiction unless a | b or b | a. If a > 4 is the greatest number of repetitions in any of the words
(n/a is the smallest period), then unless all the numbers of repetitions divide each other there must be words
with 2 or 4 repetitions, words with 3 repetitions and all larger numbers of repetitions must divide each other
and be divisible by 6.
We now divide into three cases: all the numbers of repetitions may divide either other; or there may be
words with (multiples of) 2, 3 and 6 repetitions; or all words may have at most 4 repetitions, with at least one
word having 3 repetitions and at least one having 2 or 4 repetitions.
Case 1. Suppose all the numbers of repetitions divide each other. Let k be the least number of repetitions.
Consider the word as being divided into k blocks, each of ` letters; any transposition of two adjacent letters
leaves those blocks identical. If any two adjacent letters within a block are the same, then this means all the
blocks are already identical; since the word is not constant, the letters in the first block are not all identical,
so there are two distinct adjacent letters in the first block, and transposing them leaves it distinct from the
other blocks, a contradiction. Otherwise, all pairs of adjacent letters within each block are distinct; transposing
any adjacent pair within the first block leaves it identical to the second block. If the first block has more than
two letters, this is impossible since transposing the first two letters has a different result from transposing the
second two. So the blocks all have length 2; similarly, there are just two blocks, the arrangement is abba but
transposing the adjacent letters bb does not leave the word repetitive.
Case 2. Suppose some word resulting from a transposition is made of (a multiple of) 6 repetitions, some
of 3 repetitions and some of 2 repetitions (or 4 repetitions, counted as 2). Consider it as a sequence of 6 blocks,
each of length `. If the six blocks are already identical, then as the word is not constant, there are some
two distinct adjacent letters within the first block; transposing them leaves a result where the blocks form a
5
pattern BAAAAA, which cannot have two, three or six repetitions. So the six blocks are not already identical.
If a transposition within a block results in them being identical, the blocks form a pattern (without loss of
generality) BAAAAA, ABAAAA or AABAAA. In any of these cases, apply the same transposition (that
converts between A and B) to an A block adjacent to the B block, and the result cannot have two, three or six
repetitions. Finally, consider the case where some transposition between two adjacent blocks results in all six
blocks being identical. The patterns are BCAAAA, ABCAAA and AABCAA (and considering the letters at
the start and end of each block shows B 6= C). In all cases, transposing two adjacent distinct letters within an
A block produces a result that cannot have two, three or six repetitions.
Case 3. In the remaining case, all words have at most 4 repetitions, at least one has 3 repetitions and
at least one has 2 or 4 repetitions. For the purposes of this case we will think of 4-repetition words as being
2-repetition words. The number of each letter is a multiple of 6, so n ≥ 12; consider the word as made of six
blocks of length `.
If the word is already repetitive with 2 repetitions, pattern ABCABC, any transposition between two
distinct letters leaves it no longer repetitive with two repetitions, so it must instead have three repetitions
after the transposition. If AB is not all one letter, transposing two adjacent letters within AB implies that
CA = BC, so A = B = C, the word has pattern AAAAAA but transposing within the initial AA means it no
longer has 3 repetitions. This implies that AB is all one letter, but similarly BC must also be all one letter and
so the word is constant, a contradiction.
If the word is already repetitive with 3 repetitions, it has pattern ABABAB and any transposition leaves it
no longer having 3 repetitions, so having 2 repetitions instead. ABA is not made all of one letter (since the word
is not constant) and any transposition between two adjacent distinct letters therein turns it into BAB; such a
transposition affects at most two of the blocks, so A = B, the word has pattern AAAAAA and transposing two
adjacent distinct letters within the first half cannot leave it with two repetitions.
So the word is not already repetitive, and so no two adjacent letters are the same; all transpositions give
distinct strings. Consider transpositions of adjacent letters within the first four letters; three different words
result, of which at most one is periodic with two repetitions (it must be made of two copies of the second half
of the word) and at most one is periodic with three repetitions, a contradiction.
6
EGMO 2013
Problems with Solutions
B C
D
A
Q
B C
D
Solution 2: Notice that A is the centroid of triangle BDE, since C is the midpoint of
[BD] and AE = 2CA. Let M be the midpoint of [BE]. Then M , A, D lie on a line, and
2
further, AM = 21 AD = 12 BE. This implies that ∠EAB = 90◦ .
B C
D
Solution 3: Let P be the midpoint [AE]. Since C is the midpoint of [BD], and,
moreover, AC = EP , we have
But ∠CAD < ∠CED and ∠BEC + ∠CED < 180◦ , so we must be in the first case,
i.e. ∠BEP = ∠CAD. It follows that triangles BEP and DAC are congruent, and thus
∠BP A = ∠ACB. But AP = AC, so BA is a median of the isosceles triangle BCP . Thus
AB ⊥ P C, completing the proof.
B
C D
3
Hence 4z 2 = 12x2 + 4y 2 or z 2 − y 2 = 3x2 . Let H be the foot of the perpendicular through
B to AC, and write h = BH. Then
y 2 − h2 = CH 2 , z 2 − h2 = EH 2 .
Thus
which gives a2 = b2 + c2 . Therefore ∠BAC is a right angle by the converse of the theorem
of Pythagoras.
−→ −→ −−→ −−→
Solution 6: Let AC = ~x and AB = ~y . Now AD = 2~x − ~y and EB = 2~x + ~y . Then
−−→ −−→ −−→ −−→
BE · BE = AD · AD ⇐⇒ (2~x + ~y )2 = (2~x − ~y )2 ⇐⇒ ~x · ~y = 0
−→ −−→
Remark. It is perhaps more natural to introduce CA = ~a and CB = ~b. Then we have
the equality
2 2
3~a − ~b = ~a + ~b =⇒ ~a · ~a − ~b = 0.
4
Solution 7: Let a, b, c, d, e denote the complex co-ordinates of the points A, B, C, D,
E and take the unit circle to be the circumcircle of ABC. We have
BE = AD ⇐⇒ (b − e)(b − e) = (d − a)(d − a)
⇐⇒ 2(d − a)(b − a) + 2(d − a)(b − a) + 4(b − a)(b − a) = 0
⇐⇒ 2(d − a)(a − b) + 2(d − a)(b − a)ab + 4(b − a)(a − b) = 0
⇐⇒ (d − a) − (d − a)ab + 2(b − a) = 0
⇐⇒ 2c − b − a − 2cab + a + b + 2(b − a) = 0
⇐⇒ c2 − ab + bc − ac = 0
⇐⇒ (b + c)(c − a) = 0,
Solution 8: We use areal co-ordinates with reference to the triangle ABC. Recall that if
(x1 , y1 , z1 ) and (x2 , y2 , z2 ) are points in the plane, then the square of the distance between
these two points is −a2 vw − b2 wu − c2 uv, where (u, v, w) = (x1 − x2 , y1 − y2 , z1 − z2 ).
In our case A = (1, 0, 0), B = (0, 1, 0), C = (0, 0, 1), so E = (3, 0, 2) and, introducing
point F as in the first solution, F = (−1, 0, 2). Then
and thus a2 = b2 + c2 . Therefore ∠BAC is a right angle by the converse of the theorem
of Pythagoras.
5
Problem 2. (Proposed by Matti Lehtinen, Finland)
Determine all integers m for which the m × m square can be dissected into
five rectangles, the side lengths of which are the integers 1, 2, 3, . . . , 10 in some
order.
Solution: The solution naturally divides into three different parts: we first obtain some
bounds on m. We then describe the structure of possible dissections, and finally, we deal
with the few remaining cases.
In the first part of the solution, we get rid of the cases with m 6 10 or m > 14.
Let `1 , . . . , `5 and w1 , . . . , w5 be the lengths and widths of the five rectangles. Then the
rearrangement inequality yields the lower bound
`1 w1 + `2 w2 + `3 w3 + `4 w4 + `5 w5
1
= `1 w1 + `2 w2 + `3 w3 + `4 w4 + `5 w5 + w1 `1 + w2 `2 + w3 `3 + w3 `4 + w5 `5
2
1
> 1 · 10 + 2 · 9 + 3 · 8 + · · · + 8 · 3 + 9 · 2 + 10 · 1 = 110,
2
and the upper bound
`1 w1 + `2 w2 + `3 w3 + `4 w4 + `5 w5
1
= `1 w1 + `2 w2 + `3 w3 + `4 w4 + `5 w5 + w1 `1 + w2 `2 + w3 `3 + w3 `4 + w5 `5
2
1
6 1 · 1 + 2 · 2 + 3 · 3 + · · · + 8 · 8 + 9 · 9 + 10 · 10 = 192.5,
2
As the area of the square is sandwiched between 110 and 192.5, the only possible candi-
dates for m are 11, 12, and 13.
In the second part of the solution, we show that a dissection of the square into five
rectangles must consist of a single inner rectangle and four outer rectangles that each
cover one of the four corners of the square. Indeed, if one of the sides the square had
three rectangles adjacent to it, removing these three rectangles would leave a polygon
with eight vertices, which is clearly not the union of two rectangles. Moreover, since
m > 10, each side of the square has at least two adjacent rectangles. Hence each side of
the square has precisely two adjacent rectangles, and thus the only way of partitionning
the square into five rectangles is to have a single inner rectangle and four outer rectangles
each covering of the four corners of the square, as claimed.
Let us now show that a square of size 12 × 12 cannot be dissected in the desired
way. Let R1 , R2 , R3 and R4 be the outer rectangles (in clockwise orientation along the
6
boundary of the square). If an outer rectangle has a side of length s, then some adjacent
outer rectangle must have a side of length 12 − s. Therefore, neither of s = 1 or s = 6
can be sidelengths of an outer rectangle, so the inner rectangle must have dimensions
1 × 6. One of the outer rectangles (say R1 ) must have dimensions 10 × x, and an adjacent
rectangle (say R2 ) must thus have dimensions 2 × y. Rectangle R3 then has dimensions
(12 − y) × z, and rectangle R4 has dimensions (12 − z) × (12 − x). Note that exactly one
of the three numbers x, y, z is even (and equals 4 or 8), while the other two numbers are
odd. Now, the total area of all five rectangles is
Finally, we handle the cases m = 11 and m = 13, which indeed are solutions. The
corresponding rectangle sets are 10 × 5, 1 × 9, 8 × 2, 7 × 4 and 3 × 6 for m = 11, and
10 × 5, 9 × 8, 4 × 6, 3 × 7 and 1 × 2 for m = 13. These sets can be found by trial and
error. The corresponding partitions are shown in the figure below.
3×6
9×8
10×5
10×5
7×4
8×2
1×2
4×6
3×7
1×9
Remark. The configurations for m = 11 and m = 13 given above are not unique.
A Variant for Obtaining Bounds. We first exclude the cases m 6 9 by the observa-
tion that one of the small rectangles has a side of length 10 and must fit into the square;
hence m > 10.
To exclude the cases m > 14, we work via the perimeter: as every rectangle has at
most two sides on the boundary of the m × m square, the perimeter 4m of the square is
bounded by 1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 + 10 = 55; hence m 6 13.
7
We are left to deal with the case m = 10: clearly, the rectangle with side length 10
must have one its sides of length 10 along the boundary of the square. The remaining
rectangle R of dimensions 10 × s, say, would have to be divided into four rectangles with
different sidelengths strictly less than 10. If there were at least two rectangles adjacent
to one of the sides of length s of R, removing these two rectangles from R would leave
a polygon with at least six vertices (since the sidelengths of the rectangles partitioning
R are strictly less than 10). It is clearly impossible to partition such a polygon into no
more than two rectangles with different sidelengths. Hence, given a side of length s of R,
there is only one rectangle adjacent to that side, so the rectangles adjacent to the sides
of length s of R would have to have the same length s, a contradiction.
Remark. Note that the argument of the second part of the main solution cannot be
directly applied to the case m = 10.
A Variant for Dealing with m = 12. As in the previous solution, we show that the
inner rectangle must have dimensions 1 × 6. Since the area of the square and the area of
the inner rectangle are even, the areas of the four outer rectangles must sum to an even
number. Now the four sides of the square are divided into segments of lengths 2 and 10,
3 and 9, 4 and 8, and 5 and 7. Hence the sides with adjacent segments of lengths 3 and
9, and 5 and 7 must be opposite sides of the square (otherwise, exactly one of the outer
rectangles would have odd area). However, the difference of two rectangle side lengths on
opposite sides of the square must be 1 or 6 (in order to accomodate the inner rectangle).
This is not the case, so there is no solution with m = 12.
Remark. In the case m = 12, having shown that the inner rectangle must have dimen-
sions 1 × 6, this case can also be dealt with by listing the remaining configurations one
by one.
8
Problem 3. (Proposed by Dan Schwarz, Romania)
Let n be a positive integer.
(a) Prove that there exists a set S of 6n pairwise different positive integers,
such that the least common multiple of any two elements of S is no larger
than 32n2 .
(b) Prove that every set T of 6n pairwise different positive integers contains
two elements the least common multiple of which is larger than 9n2 .
Solution: (a) Let the set A consist of the 4n integers 1, 2, . . . , 4n and let the set B
consist of the 2n even integers 4n + 2, 4n + 4, . . . , 8n. We claim that the 6n-element set
S = A ∪ B has the desired property.
Indeed, the least common multiple of two (even) elements of B is no larger than
8n · (8n/2) = 32n2 , and the least common multiple of some element of A and some
element of A ∪ B is at most their product, which is at most 4n · 8n = 32n2 .
(b) We prove the following lemma: “If a set U contains m + 1 integers, where m > 2,
that are all not less than m, then some two of its elements have least common multiple
strictly larger than m2 .”
Let the elements of U be u1 > u2 > · · · > um+1 > m. Note that 1/u1 6 1/ui 6 1/m
for 1 6 i 6 m + 1. We partition the interval [1/u1 ; 1/m] into m subintervals of equal
length. By the pigeonhole principle, there exist indices i, j with 1 6 i < j 6 m + 1 such
that 1/ui and 1/uj belong to the same subinterval. Hence
1 1 1 1 1 1
0 < − 6 − < .
uj ui m m u1 m2
Now 1/uj −1/ui is a positive fraction with denominator lcm(ui , uj ). The above thus yields
the lower bound lcm(ui , uj ) > m2 , completing the proof of the lemma.
Applying the lemma with m = 3n to the 3n + 1 largest elements of T , which are all
not less than 3n, we arrive at the desired statement.
A Variant. Alternatively, for part (b), we prove the following lemma: “If a set U
contains m > 2 integers that all are greater than m, then some two of its elements have
least common multiple strictly larger than m2 .”
Let u1 > u2 > · · · > um be the elements of U . Since um > m = m2 /m, there exists
a smallest index k such that uk > m2 /k. If k = 1, then u1 > m2 , and the least common
multiple of u1 and u2 is strictly larger than m2 . So let us suppose k > 1 from now on, so
that we have uk > m2 /k and uk−1 6 m2 /(k − 1). The greatest common divisor d of uk−1
and uk satisfies
m2 m2 m2
d 6 uk−1 − uk < − = .
k−1 k (k − 1)k
9
This implies m2 /(dk) > k − 1 and uk /d > k − 1, and hence uk /d > k. But then the least
common multiple of uk−1 and uk equals
uk−1 uk uk m2
> uk · > · k = m2 ,
d d k
and the proof of the lemma is complete.
If we remove the 3n smallest elements from set T and apply the lemma with m = 3n
to the remaining elements, we arrive at the desired statement.
10
Problem 4. (Proposed by Vesna Iršič, Slovenia)
Find all positive integers a and b for which there are three consecutive integers
at which the polynomial
n5 + a
P (n) =
b
takes integer values.
As the positive integer b divides 22, we are left with the four cases b = 1, b = 2, b = 11
and b = 22.
If b is even (i.e. b = 2 or b = 22), then we get a contradiction from (3), because the
integer B = 2(5x3 + 5x2 ) + 5(x4 + x) + 1 is odd, and hence not divisible by any even
integer.
For b = 1, it is trivial to see that a polynomial of the form P (n) = n5 + a, with a any
positive integer, has the desired property.
For b = 11, we note that
Hence a polynomial of the form P (n) = (n5 + a)/11 has the desired property if and only
if a ≡ ±1 (mod 11). This completes the proof.
11
A Variant. We start by following the first solution up to equation (4). We note that
b = 1 is a trivial solution, and assume from now on that b > 2. As (x − 1)5 + a and x5 + a
have different parity, b must be odd. As B in (3) is a multiple of b, we conclude that (i)
b is not divisible by 5 and that (ii) b and x are relatively prime. As C = 10x(2x2 + 1) in
(4) is divisible by b, we altogether derive
Hence b = 11 is the only remaining candidate, and it is handled as in the first solution.
Solution 2: Let p be a prime such that p divides b. For some integer x, we have
Let r = tk and s = t` . In particular, the above yields r, s 6≡ 1 (mod p), and thus
It follows that
Now t5 ≡ 1 (mod p), so r and s must be congruent, modulo p, to some of the non-trivial
fifth roots of unity t, t2 , t3 , t4 . Observe that, for any pair of these non-trivial roots of unity,
12
either one is the other’s inverse, or one is the other’s square. In the first case, we have
r + r−1 ≡ 2 (mod p), implying r ≡ 1 (mod p), a contradiction. Hence
r + r2 ≡ 2 (mod p) =⇒ (r − 1)(r + 2) ≡ 0 (mod p),
or
s2 + s ≡ 2 (mod p) =⇒ (s − 1)(s + 2) ≡ 0 (mod p).
Thus, since r, s 6≡ 1 (mod p), we have r ≡ −2 (mod p) or s ≡ −2 (mod p), and thus
1 ≡ r5 ≡ −32 (mod p) or an analogous equation obtained from s. Hence p | 33. Since
p ≡ 1 (mod 5), it follows that p = 11, i.e. b is a power of 11.
Examining the fifth powers modulo 11, we see that b = 11 is indeed a solution with
a ≡ ±1 (mod 11) and, correspondingly, x ≡ ±4 (mod 11). Now suppose, for the sake of
contradiction, that 112 divides b. Then, for some integer m, we must have
x − 1, x, x + 1 ≡ ± 3 + 11m, 4 + 11m, 5 + 11m (mod 121),
and thus, substituting into the condition of the problem,
35 + 55 · 34 m ≡ 45 + 55 · 44 m ≡ 55 + 55 · 54 m (mod 121)
=⇒ 1 − 22m ≡ 56 + 44m ≡ −21 + 11m (mod 121).
Hence 33m ≡ 22 (mod 121) and 33m ≡ 44 (mod 121), so 22 ≡ 0 (mod 121), a contra-
diction. It follows that b | 11.
Finally, we conclude that the positive integers satisfying the original condition are
b = 11, with a ≡ ±1 (mod 11), and b = 1, for any positive integer a.
Solution 3: Denote the three consecutive integers by x−1, x, and x+1 as in Solution 1.
By computing the differences in (1), we find
F := (x + 1)5 − x5 = 5x4 + 10x3 + 10x2 + 5x + 1 ≡ 0 (mod b),
G := x5 − (x − 1)5 = 5x4 − 10x3 + 10x2 − 5x + 1 ≡ 0 (mod b).
By determining the polynomial greatest divisor of F (x) and G(x) using the Euclidean
algorithm, we find that
p(x)F (x) + q(x)G(x) = 22, (7)
where
p(x) = −15x3 + 30x2 − 28x + 11,
q(x) = 15x3 + 30x2 + 28x + 11.
Since b | F (x) and b | G(x), it follows from (7) that b | 22. We now finish off the problem
as in Solution 1.
13
Problem 5. (Proposed by Waldemar Pompe, Poland)
Let Ω be the circumcircle of the triangle ABC. The circle ω is tangent to the
sides AC and BC, and it is internally tangent to Ω at the point P . A line
parallel to AB and intersecting the interior of triangle ABC is tangent to ω
at Q.
Prove that ∠ACP = ∠QCB.
BM = M A =⇒ BL + LM = M K + KA =⇒ LC + LM = M K + CK
=⇒ 2LM + M C = M C + 2CK =⇒ LM = CK.
Thus arcs F Q and DE of ω are equal, too, where D is the intersection of CP with ω. Since
CE and CF are tangents to ω, this implies that ∠DEC = ∠CF Q. Further, CE = CF ,
and thus triangles CED and CF Q are congruent. In particular, ∠ECD = ∠QCF , as
required.
M
C
K Q
D
ω F
E
I
A B
14
A Variant. As above, we show that arcs F Q and DE of ω are equal, which implies
that DEF Q is an isoceles trapezoid, and so we have ∠F ED = ∠QF E. Together with
|F Q| = |DE|, this implies that, since E and F are images of each other under reflection
in the angle bisector CI of ∠C, so are the segments [EQ] and [F D], and, in particular,
D and Q. In turn, this yields ∠ECD = ∠QCF , as required.
Remark. Let J denote the incentre of ABC. By Sawayama’s theorem, J is the midpoint
of [EF ], i.e. P J is a median of P F E. Since C is the intersection of the tangents AC and
BC to the circumcircle of P F E at E and F , respectively, P C is a symmedian of P F E.
Thus ∠CP E = ∠F P J. But, since the arcs F Q and DE of ω are equal, ∠CP E = ∠F P Q.
This shows that J lies on the line P Q.
Another Variant. We show that arcs QE and F D are equal, and then finish as in the
main solution. Let BP meet ω again at Z. Consider the homothety H that maps ω onto
Ω. Under H , D 7→ C and Z 7→ B, so DZ k CB. (This also follows by considering the
common tangent to ω and Ω, and tangential angles.) Now, by power of a point,
BF 2 = BZ · BP, CF 2 = CD · CP.
Now DZ k CB implies BZ/BP = CD/CP , and so, dividing the two previous equa-
tions by each other, and taking square roots, BF/CF = BP/CP . Hence P F bissects
angle ∠BP C. Now let ∠BP F = ∠F P C = β. By tangential angles, it follows that
∠CF D = β. Further, ∠BAC = ∠BP C = 2β. Let the tangent to ω through Q and
parallel to AB meet AC at X. Then ∠QXC = 2β, so, since XQ = XE by tangency,
∠QEX = β. By tangential angles, it follows that arcs F D and QE are equal, as claimed.
X D Q
ω F
A B
Z
15
Solution 2: Let I and O denote the respective centres of ω and Ω. Observe that
CI is the angle bisector of angle ∠C, because ω is tangent to AC and BC. Consider the
homethety H that maps ω onto Ω. Let M be the image of Q under H . By construction,
IQ ⊥ AB, so OM ⊥ AB. Thus the diameter OM of Ω passes through the midpoint of
the arc AB of Ω, which also lies on the angle bisector CI. This implies that ∠ICM = 90◦ .
We next show that P, I, Q, C lie on a circle. Notice that
∠P QI = 90◦ − 12 ∠QIP = 90◦ − 12 ∠M OP = 90◦ − 180◦ − ∠P CM
= ∠P CI + ∠ICM − 90◦ = ∠P CI.
Hence P, I, Q, C lie on a circle. But P I = IQ, so CI is the angle bisector of ∠P CQ. Since
CI is also the angle bisector of angle ∠C, it follows that ∠ACP = ∠QCB, as required.
M
C
ω O
A B
T
16
But QI ⊥ AB by construction, and thus
∠OIQ = 90◦ − ∠BT O = 180◦ − α − β − 2γ
=⇒ ∠QIP = 180◦ − ∠OIQ = α + β + 2γ
=⇒ ∠P QI = 90◦ − 12 α − 12 β − γ.
Hence ∠ICQ = ∠P QI, and thus P IQC is cyclic. Since P I = QI, it follows that CI is
the angle bisector of ∠P CQ, which completes the proof.
Solution 3: Let I and O denote the respective centres of ω and Ω. Let D be the second
point of intersection of CP with ω, and let ` denote the tangent to ω at D, which meets
AC at S. Hence ID ⊥ `. By construction, P , I, O lie one a line, and hence the isosceles
triangles P ID and P OC are similar. In particular, it follows that OC ⊥ `, so C is the
midpoint of the arc of Ω defined by the points of intersection of ` with Ω. It is easy to
see that this implies that
∠DSC = ∠ABC.
Under reflection in the angle bisector CI of ∠C, ` is thus mapped to a tangent to ω
parallel to AB and intersecting the interior of ABC, since ω is mapped to itself under
this reflection. In particular, D is mapped to Q, and thus ∠QCB = ∠ACD, as required.
D
S
ω O
` I
A B
17
Solution 4: Let the tangent to ω at Q meet AC and BC at X and Y , respectively. Then
AC/XC = BC/Y C, and thus there is a radius r such that r2 = AC · Y C = BC · XC.
Let Γ denote the circle with centre C and radius r, and consider the inversion I in the
circle Γ . Under I ,
C Γ
B0
X Y
P0 Q
A0
ω
A B
Solution 5: Let r be the radius such that r2 = AC ·BC. Let J denote the composition
of the inversion I in the circle of centre C and radius r, followed by the reflection in the
18
angle bisector of ∠C. Under J ,
A 7−→ B, B 7→ A;
Ω 7−→ the line AB;
ω 7−→ ω 0 , the excircle of ABC opposite the vertex C;
P 7−→ Q0 , the point where ω 0 touches AB;
In particular, note that the image ω 0 of ω under J is a circle tangent to AC, BC and
AB, so it is either the incircle of ABC, or the excircle opposite vertex C. Observe that
r > min {AC, BC}, so the image of the points of tangency of ω must lie outside ABC,
and thus ω 0 cannot be the incircle. Thus ω 0 is the excircle opposite vertex C as claimed.
Further, the point of tangency P is mapped to Q0 .
Now, since the line CP is mapped to itself under the inversion I , and mapped onto
CQ0 under J , CP and CQ0 are images of each other under reflection in the angle bisector
of ∠C. But C, Q, Q0 lie on a line for there is a homothety with centre C that maps ω
onto the excircle ω 0 . This completes the proof.
Q0
A B
P
ω0
19
Now, a point Z on the circle Ω satisfies
The triangle ABC is defined by the points E and. F on ω, the intersection C of the
corresponding tangents lying on Ω. Thus c = 2ef (e + f ), and further
and this is the equality defining o. The points A and B are the second intersection points
of Ω with the tangents to ω at E and F respectively. A point Z on the tangent through
E is given by z = 2e − e2 z ∗ , and thus A and C satisfy
2e − e2 z ∗ z ∗ − o 2e − e2 z ∗ + z ∗ − 2o − 1 = 0
⇐⇒ −e2 z ∗ 2 + 2e + oe2 − o z ∗ − 2eo + 2o + 1 = 0
⇐⇒ z ∗ 2 − 2e∗ + o − oe∗ 2 z ∗ + 2e∗ o + 2oe∗ 2 + e∗ 2 = 0,
2e∗ f
a∗ + c∗ = 2e∗ + o − oe∗ 2 =⇒ a∗ = + o 1 − e∗ 2 ,
e+f
and similarly
2f ∗ e
∗
b = + o 1 − f ∗2 .
f +e
Then
2(ef ∗ − e∗ f )
b ∗ − a∗ = + o e∗ 2 − f ∗ 2
e+f
2ef f ∗ 2 − e∗ 2
= + o e∗ 2 − f ∗ 2
e+f
!
∗2 ∗2
2ef
= f −e − o = f ∗ 2 − e∗ 2 (c − o).
e+f
z = 2q − q 2 z ∗ ⇐⇒ z − q = q − q 2 z ∗ = −q 2 (z ∗ − q ∗ ),
b−a z−q −q 2 (z ∗ − q ∗ )
= = = −q 2 .
b ∗ − a∗ z∗ − q∗ z∗ − q∗
20
It follows that
b−a f 2 − e2 (c∗ − o) ∗
2 2c − o
q2 = − = − = e f
b ∗ − a∗ f ∗ 2 − e∗ 2 (c − o) c−o
(c∗ − o)2 ∗
2 2 (c − o)
2
= e2 f 2 = e f ,
|c − o|2 (1 + o)2
c∗ − o
q = ef ,
1+o
where the choice of sign is to be justified a posteriori. Further, the point D satisfies
d−p c−p c−p c+1
−dp = ∗ ∗
= ∗ =⇒ d=− = ∗ ,
d −p c − p∗ c∗ p
−1 c +1
using p = −1 to obtain the final equality.
Now, it suffices to show that (i) DQ k EF ⊥ CI and (ii) the midpoint of [DQ] is on
CI. The desired equality then follows by symmetry with respect to the angle bisector of
the angle ∠ACB. Notice that (i) is equivalent with
d−q e−f
∗ ∗
= ∗ ⇐⇒ dq = ef.
d −q e − f∗
c + 1 c∗ − o
dq = ef ⇐⇒ ef = ef ⇐⇒ (c + 1)(c∗ − o) = (c∗ + 1)(1 + o)
c∗ + 1 1 + o
⇐⇒ cc∗ − o(c + c∗ ) − 2o − 1 = 0.
The last equality is precisely the defining relation for o, (1). This proves (i). Further, the
midpoint of [DQ] is 21 (d + q), so it remains to check that
d+q c
dq = ∗ ∗
= ∗ = ef,
d +q c
where the first equality expresses that [DQ] is a chord of ω (obviously) containing its
midpoint, the second equality expresses the alignment of the midpoint of [DQ], C and
I, and the third equality follows from the expression for c. But we have just shown that
dq = ef . This proves (ii), justifies the choice of sign for q a posteriori, and thus completes
the solution of the problem.
21
Problem 6. (Proposed by Emil Kolev, Bulgaria)
Snow White and the Seven Dwarves are living in their house in the forest.
On each of 16 consecutive days, some of the dwarves worked in the diamond
mine while the remaining dwarves collected berries in the forest. No dwarf
performed both types of work on the same day. On any two different (not
necessarily consecutive) days, at least three dwarves each performed both
types of work. Further, on the first day, all seven dwarves worked in the
diamond mine.
Prove that, on one of these 16 days, all seven dwarves were collecting
berries.
Solution 1: We define V as the set of all 128 vectors of length 7 with entries in {0, 1}.
Every such vector encodes the work schedule of a single day: if the i-th entry is 0 then
the i-th dwarf works in the mine, and if this entry is 1 then the i-th dwarf collects berries.
The 16 working days correspond to 16 vectors d1 , . . . , d16 in V , which we will call day-
vectors. The condition imposed on any pair of distinct days means that any two distinct
day-vectors di and dj differ in at least three positions.
We say that a vector x ∈ V covers some vector y ∈ V , if x and y differ in at most
one position; note that every vector in V covers exactly eight vectors. For each of the 16
day-vectors di we define Bi ⊂ V as the set of the eight vectors that are covered by di . As,
for i 6= j, the day-vectors di and dj differ in at least three positions, their corresponding
sets Bi and Bj are disjoint. As the sets B1 , . . . , B16 together contain 16 · 8 = 128 = |V |
distinct elements, they form a partition of V ; in other words, every vector in V is covered
by precisely one day-vector.
The weight of a vector v ∈ V is defined as the number of 1-entries in v. For
7
k = 0, 1, . . . , 7, the set V contains k vectors of weight k. Let us analyse the 16 day-
vectors d1 , . . . , d16 by their weights, and let us discuss how the vectors in V are covered
by them.
1. As all seven dwarves work in the diamond mine on the first day, the first day-vector is
d1 = (0000000). This day-vector covers all vectors in V with weight 0 or 1.
22
To summarize, one day-vector has weight 0, seven have weight 3, and seven have weight 4.
None of these 15 day-vectors covers any vector of weight 6 or 7, so that the eight heavy-
weight vectors in V must be covered by the only remaining day-vector; and this remaining
vector must be (1111111). On the day corresponding to (1111111) all seven dwarves are
collecting berries, and that is what we wanted to show.
As every dwarf collected berries exactly eight times, we get that, further,
23
which simplifies to
which simplifies to
Multiplying (1) by −40, multiplying (2) by 10, multiplying (3) by −1, multiplying (4) by
4, and then adding up the four resulting equations yields 5d(3) = 30 and hence d(3) = 6.
Then (4) yields d(4) = 11. As d(3) + d(4) = 17, the total number of days cannot be 16.
We have reached the desired contradiction.
As d(5) and d(6) are positive integers, (5) implies 0 6 d(6) 6 2. Only the case d(6) = 1
yields an integral value d(5) = 4. The equations (1) and (2) then yield d(3) = 10 and
d(4) = 0.
Now let us look at the d(3) = 10 special days on which exactly three dwarves were
collecting berries. One of the dwarves collected berries on at least five special days (if every
dwarf collected berries on at most four special days, this would allow at most 7 · 4/3 < 10
special days); we call this dwarf X. On at least two out of these five special days, some
dwarf Y must have collected berries together with X. Then these two days contradict
the problem statement. We have reached the desired contradiction.
24
Problems and Solutions
Day 1
The EGMO 2014 Problem Committee thanks the following countries for submitting
problem proposals:
• Bulgaria
• Iran
• Japan
• Luxembourg
• Netherlands
• Poland
• Romania
• Ukraine
• United Kingdom
Okan Tekman
Selim Bahadr
ahin Emrah
Solution 1.
Ift < 2/3, take a triangle with sides c = b = 1 and a = 2 − . Then b2 + cat + c2 +
abt − a2 − bct = 3t − 2 + (4 − 2t − ) ≤ 0 for small positive ; for instance, for any
0 < < (2 − 3t)/(4 − 2t).
On the other hand, if t > 2, then take a triangle with sides b = c = 1 and a = .
Then b2 + cat + c2 + abt − a2 − bct = 2 − t + (2t − ) ≤ 0 for small positive ; for
instance, for any 0 < < (t − 2)/(2t).
Now assume that 2/3 ≤ t ≤ 2 and b + c > a. Then using (b + c)2 ≥ 4bc we obtain
Solution 2.
2 2 2 2 2
(x +y −z +xy +xz +yz)+2(z 2 +xz +yz −xy) = ((x−y)2 +4(x+y)z +2z 2 ) > 0
3 3
and at t=2 where
As the rst three terms are non-negative and the last term is positive, the result
follows.
Solution 4.
H of the triangle DEF and the midpoint M of the arc BAC of the circumcircle of
the triangle ABC are collinear.
Solution 1.
be the point of intersection of the lines CI and BE . Then we have the power relation
Let ω1 and ω2 be the circles with diameters BD and CE , respectively. From the
power relations above we conclude that IH is the radical axis of the circles ω1 and
ω2 .
Let O1 and O2 be centers of ω1 and ω2 , respectively. Then M B = M C , BO1 = CO2
and ∠M BO1 = ∠M CO2 , and the triangles M BO1 and M CO2 are congruent. Hence
M O1 = M O2 . Since radii of ω1 and ω2 are equal, this implies that M lies on the
radical axis of ω1 and ω2 and M , I , H are collinear.
Solution 2.
IQ · ID.
Let R be the second point intersection of the circumcircle of triangle HEP and the
lineHI . As IH · IR = IP · IE = IQ · ID, the points D, Q, H , R are also concyclic.
We have ∠DQH = ∠EP H = ∠DF E = ∠BF C = 180 − ∠BIC = 90 − ∠BAC/2.
◦ ◦
∠ERH = ∠180◦ −(90◦ −∠BAC/2) = 90◦ +∠BAC/2. Hence R is inside the triangle
DEH and ∠DRE = 360◦ − ∠DRH − ∠ERH = 180◦ − ∠BAC and it follows that
the points A, D , R, E are concyclic.
Suppose that we have a coordinate system and (bx , by ), (cx , cy ), (dx , dy ), (ex , ey ) are
−→ −−→ −→ −−→
the coordinates of the points B , C , D , E , respectively. From BI · CD = 0, CI · BE =
−−→ −−→ −−→ −−→ −→ → − → − → − → −
0, EH · CD = 0, DH · BE = 0 we obtain IH · ( B − C − E + D ) = 0. Hence the
slope of the line IH is (cx + ex − bx − dx )/(by + dy − cy − ey ).
Assume that the x-axis lies along the line BC , and let α = ∠BAC , β = ∠ABC ,
θ = ∠ACB . Since DB = BC = CE , we have cx − bx = BC , ex − dx = BC −
BC cos β − BC cos θ, by = cy = 0, dy − ey = BC sin β − BC sin θ. Therefore the
slope of IH is (2 − cos β − cos θ)/(sin β − sin θ).
Now we will show that the slope of the line M I is the same. Let r andR be
the inradius and circumradius of the triangle ABC , respectively. As ∠BM C =
∠BAC = α and BM = M C , we have
BC α
Å ã
AC − AB
my − iy = cot −r and mx − ix =
2 2 2
where (mx , my ) and (ix , iy ) are the coordinates of M and I, respectively. Therefore
BC AC AB
= = = 2R ,
sin α sin β sin θ
hence
α Å ã
2 α
Å ã
BC cot = 4R cos = 2R(1 + cos α)
2 2
and
r
= cos α + cos β + cos θ − 1
R
as
BC cot(α/2) − 2r 2R(1 + cos α) − 2r 2 − cos β − cos θ
= =
AC − AB 2R(sin β − sin θ) sin β − sin θ
giving the collinearity of the points I, H, M .
Solution 4.
Let the bisectors BI and CI meet the circumcircle of ABC again at P and Q,
respectively. Let the altitude of DEF belonging to D meet BI at R and the one
belonging to E meet CI at S.
Since BI is angle bisector of the CBD, BI and CD are
iscosceles triangle
∠M P B = ∠M QC = ∠M QI , and ∠P IQ = (P ¯ A + CB
¯ + AQ)/2
¯ = (P
¯ C + CB
¯ +
BQ)/2
¯ = ∠P M Q. Therefore M P IQ is a parallelogram.
Since CI is angle bisector of the iscosceles triangle BCE , the triangle BSE is
also isosceles. Hence ∠F BS = ∠EBS = ∠SEB = ∠HEF = ∠HDF = ∠RDF =
M , H, I are collinear.
3. We denote the number of positive divisors of a positive integer m by d(m) and
Prove that there exist innitely many positive integers n such that ω(n) = k and
d(n) does not divide d(a2 + b2 ) for any positive integers a, b satisfying a + b = n.
Proposed by JPN
Solution.
We will show that any number of the form n = 2p−1 m where m is a positive integer
that has exactly k − 1 prime factors all of which are greater than 3 and p is a prime
(p−1)/2
number such that (5/4) > m satises the given condition.
Suppose that a and b are positive integers such that a + b = n and d(n) | d(a2 + b2 ).
2 2 2 2 cp−1
Then p | d(a + b ). Hence a + b = q r where q is a prime, c is a positive integer
and r is a positive integer not divisible by q . If q ≥ 5, then
gives a contradiction. So q is 2 or 3.
Ifq = 3, then a2 + b2 is divisible by 3 and this implies that both a and b are divisible
by 3. This means n = a + b is divisible by 3, a contradiction. Hence q = 2.
1. But then t < p − 1 gives (c/2)p = t + 1 < p, which is not possible either.
Problems and Solutions
Day 2
The EGMO 2014 Problem Committee thanks the following countries for submitting
problem proposals:
• Bulgaria
• Iran
• Japan
• Luxembourg
• Netherlands
• Poland
• Romania
• Ukraine
• United Kingdom
Okan Tekman
Selim Bahadr
ahin Emrah
Solution 1.
Suppose that n has one of these forms. For an integer i, let xi be the largest integer
xi
such that 2 divides i. Now assume that 0 < i < n, 0 < j < n, i 6= j , n divides
Solution 2.
nite set, the integers x1 , x2 , . . . , xn−1 exist if and only if for each i in D there exists
k(i)
a positive integer k(i) such that f (i) belongs to E . This can be seen as follows:
• If f k (i) does not belong to E for any k > 0 for some i, then there exists
k2 > k1 > 0 such that f k1 (i) = f k2 (i), leading to the contradiction xf k1 (i) <
xf k2 (i) = xf k1 (i) .
• On the other hand, if such k(i) exists for each i in D, and if k0 (i) denotes
the smallest such, then the condition of the problem is satised by letting
In other words, the integers x1 , x2 , . . . , xn−1 exist if and only if for each i in D there
k(i)
exists a positive integer k(i) such that (−2) i ≡ n/3, n/2 or 2n/3 (mod n). For
i = 1, this implies that n = 2 with k ≥ 1 or n = 3 · 2k with k ≥ 0. On the other
k
hand, if n has one of these forms, letting k(i) = k does the trick for all i in D .
Solution 3.
Suppose that x1 , x2 , . . . , xk−1 satisfy the condition of the problem for n = k . Let
y2i = xi for 1 ≤ i ≤ k − 1 and choose y2i−1 for 1 ≤ i ≤ k to be less than
min{x1 , x2 , . . . , xk−1 }. Now suppose that for n = 2k we have 0 < i < n, 0 < j < n,
i 6= j , n divides 2i+j . Then j is even. If i is also even, then 0 < i/2 < k , 0 < j/2 < k
and k divides 2(i/2) + (j/2); hence yi = xi/2 < xj/2 = yj . On the other hand, if i is
and let bi+1 be the remainder of the division of (−2)bi by n for i ≥ 0. No terms of
this sequence is 0 and no two consecutive terms are both equal to b1 as m > 3. On
φ(m)
the other hand, as (−2) ≡ 1 (mod m), we have bφ(m) ≡ (−2)φ(m) 2k ≡ 2k ≡ b0
(mod n), and hence bφ(m) = b0 . Since 2bi + bi+1 is divisible by n for all i ≥ 0, we
have xb0 < xb1 < · · · < xb = xb0 , a contradiction.
φ(m)
5. Let n be a positive integer. We have n boxes where each box contains a non-
negative number of pebbles. In each move we are allowed to take two pebbles from
a box we choose, throw away one of the pebbles and put the other pebble in another
moves. Determine all initial congurations of pebbles which are not solvable, but
become solvable when an additional pebble is added to a box, no matter which box
is chosen.
The answer is any conguration with 2n − 2 pebbles which has even numbers of
n
X xi − 1
D(x) =
i=1
2
1 1
D(x) = N (x) − n + O(x)
2 2
where N (x) is the total number of pebbles and O(x) is the number of boxes with
odd number of pebbles). As D is nonnegative for any conguration which does not
have any empty boxes, it is also nonnegative for any solvable conguration. On the
As N (x) and O(x) have the same parity, a conguration x is solvable exactly when
O(x) ≥ 2n−N (x), and unsolvable exactly when O(x) ≤ 2n−2−N (x). In particular,
any conguration with 2n − 1 pebbles is solvable, and a conguration with 2n − 2
pebbles is unsolvable if and only if all boxes contain even numbers of pebbles.
Suppose that x0 is obtained from x by adding a pebble in some box. Then O(x0 ) =
O(x) + 1 or O(x0 ) = O(x) − 1. If x is unsolvable and x0 is solvable, then we must
0 0
have O(x) ≤ 2n − 2 − N (x) and O(x ) ≥ 2n − N (x ) = 2n − 1 − N (x), and hence
O(x0 ) = O(x) + 1. That is, the pebble must be added to a box with an even number
of pebbles. This can be the case irrespective of where the pebble is added only if
all boxes contain even numbers of pebbles, and 0 = O(x) ≤ 2n − 2 − N (x) and
0
1 = O(x ) ≥ 2n − 1 − N (x); that is, N (x) = 2n − 2.
Solution 2. Let x be a conguration and x̃ be another conguration obtained from
x by removing two pebbles from a box and depositing them in another box.
Let us call two congurations equivalent if they have the same total number of
pebbles and parities of the number of pebbles in the corresponding boxes are the
same. (It does not matter whether we consider this equivalence for a xed ordering
Let us a call a conguration with all boxes containing two or fewer pebbles scant.
makes it solvable if and only if the conguration has exactly one empty box and
the pebble is added to the empty box or to a box containing two pebbles. Hence,
conguration irrespective of where it is added if and only if all boxes have even
numbers of pebbles and exactly one of them is empty. Therefore, the addition of a
the pebble is added if and only if the conguration has 2n − 2 pebbles and all boxes
have even numbers of pebbles.
Proof of Claim 1 : Suppose that the two pebbles were moved from box B in x to
box B̃ in x̃, and x is solvable. Then we perform exactly the same sequence of moves
for x̃ as we did for x except that instead of the rst move that is made out of B we
make a move from B̃ (into the same box), and if there was no move from B , then
Proof of Claim 2 : Any move from a scant conguration either leaves the number of
empty boxes the same and the resulting conguration is also scant (if it is made into
an empty box), or increases the number of empty boxes by one (if it is made into
a nonempty box). In the second case, if the move was made into a box containing
one pebble, then the resulting conguration is still scant. On the other hand, if
it is made into a box containing two pebbles, then the resulting conguration is
equivalent to the scant conguration which has one pebble in the box the move was
made into and exactly the same number of pebbles in all other boxes as the original
1
The answer is the functions f (x) = x, f (x) = −x and f (x) = − x.
2
Solution.
1
It can be easily checked that the functions f (x) = x, f (x) = −x and f (x) = −x
2
satisfy the given condition. We will show that these are the only functions doing so.
for allx. In particular, 0 is a value of f . Suppose that u and v are such that
f (u) = 0 = f (v). Plugging x = u or v and y = u or v in the original equations
2 2 2 2 2 2
we get f (u ) = u , f (u ) = uv , f (v ) = uv and f (v ) = v . We conclude that
for all x and y . f (x)2 + y 2 + 2xf (y) 6= f (y)2 + x2 + 2yf (x) for some x and y .
Suppose
2 2
Then by the observations above, (x + f (y))(y + f (x)) 6= 0 and f (x) + y + 2xf (y) =
−(f (y)2 + x2 + 2yf (x)). But these conditions are contradictory as the second one
2 2
can be rewritten as (f (x) + y) + (f (y) + x) = 0.
for all x and y . In particular, letting y = 0 we obtain f (x)2 = (f (0) − x)2 for all x.
Let f (x) = s(x)(f (0) − x) where s : R → {1, −1}. Plugging this in (***) gives
then −1 + 2f (0)/x = 1. That is, there can be only one such x, that x is f (0), and
hence f (x) = f (0) − x for all x. Putting this back in the original equation gives
Remark:
The following line of reasoning or a variant of it can be used between (*) and (***):
Suppose that f (x1 ) = f (x2 ) 6= 0 for some x1 and x2 . Then from (*) it follows that
x1 f (−f (x1 )) = x2 f (−f (x2 )) = x2 f (−f (x1 )) and hence either x1 = x2 or f (x1 ) =
f (x2 ) = −a. In the second case, using (*) again we obtain a2 = a/2 and therefore a =
1/2. Now letting x = 1/2 in the original equation gives f (y 2 + f (y)) = y(f (y) + 1/2)
for all y . From this letting y = 0 we obtain f (0) = 1/2, and letting f (y) = −1/2 we
2 2
obtain f (y − 1/2) = 0 and y = 1. To summarize, f (x1 ) = f (x2 ) 6= 0 implies either
for all x. f (x)2 + 2xf (0) 6= f (0)2 + x2 for some x, then by the observation above
If
2 2 2
we must have f (1/2) = 0, f (0) = 1/2 and f (x) + 2xf (0) = −(f (0) + x ). We can
2 2
rewrite this as f (x) + (f (0) + x) = 0 to obtain x = 1/2 and f (0) = −x = −1/2,
2 2 2
which contradicts f (0) = 1/2. So we conclude that f (x) + 2xf (0) = f (0) + x
2 2
for all x. This implies f (x) = (f (0) − x) for all x. In particular, the second case
2 2
considered above is not possible as (f (0) − 1) = f (1) = f (−1) = (f (0) + 1) means
f (0) = 0, contradicting f (0) = 1/2. Therefore f is injective and from (**) now it
follows that
• Ireland
• Japan
• Luxembourg
• Macedonia
• Mexico
• Republic of Moldova
• Netherlands
• Poland
• Romania
• Slovenia
• Turkey
• Ukraine
Yauheni Barabanov
Mikhail Karpuk
Aliaksei Vaidzelevich
Igor Voronovich
PROBLEMS
Day 1
Problem 1. Let ABC be an acute-angled triangle, and let D be the foot of the altitude from C.
The angle bisector of ∠ABC intersects CD at E and meets the circumcircle ω of triangle ADE
again at F. If ∠ADF = 45◦ , show that CF is tangent to ω.
Problem 3. Let n, m be integers greater than 1, and let a1 , a2 , . . . , am be positive integers not
greater than nm . Prove that there exist positive integers b1 , b2 , . . . , bm not greater than n, such that
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) < n,
Day 2
Problem 5. Let m, n be positive integers with m > 1. Anastasia partitions the integers 1, 2, . . . , 2m
into m pairs. Boris then chooses one integer from each pair and finds the sum of these chosen integers.
Prove that Anastasia can select the pairs so that Boris cannot make his sum equal to n.
Problem 6. Let H be the orthocentre and G be the centroid of acute-angled triangle ABC with
AB = AC. The line AG intersects the circumcircle of ABC at A and P. Let P be the reflection of
P in the line BC. Prove that ∠CAB = 60◦ if and only if HG = GP .
4
Problem 1. Let ABC be an acute-angled triangle, and let D be the foot of the altitude from C.
The angle bisector of ∠ABC intersects CD at E and meets the circumcircle ω of triangle ADE
again at F. If ∠ADF = 45◦ , show that CF is tangent to ω.
(Luxembourg)
C Solution 1: Since ∠CDF = 90◦ − 45◦ = 45◦ , the line DF bisects
F ∠CDA, and so F lies on the perpendicular bisector of segment AE,
which meets AB at G. Let ∠ABC = 2β. Since ADEF is cyclic,
∠AF E = 90◦ , and hence ∠F AE = 45◦ . Further, as BF bisects
∠ABC, we have ∠F AB = 90◦ − β, and thus
ω E ∠EAB = ∠AEG = 45◦ − β, and ∠AED = 45◦ + β,
45◦ so ∠GED = 2β. This implies that right-angled triangles EDG
A G D B and BDC are similar, and so we have |GD|/|CD| = |DE|/|DB|.
Thus the right-angled triangles DEB and DGC are similar,
whence ∠GCD = ∠DBE = β. But ∠DF E = ∠DAE = 45◦ −
β, then ∠GF D = 45◦ − ∠DF E = β. Hence GDCF is cyclic, so
∠GF C = 90◦ , whence CF is perpendicular to the radius F G of ω.
It follows that CF is a tangent to ω, as required.
Solution 2: As ∠ADF = 45◦ line DF is an exterior bisector of ∠CDB. Since BF bisects ∠DBC line
CF is an exterior bisector of ∠BCD. Let ∠ABC = 2β, so ∠ECF = (∠DBC + ∠CDB)/2 = 45◦ + β.
Hence ∠CF E = 180◦ − ∠ECF − ∠BCE − ∠EBC = 180◦ − (45◦ + β + 90◦ − 2β + β) = 45◦ . It follows
that ∠F DC = ∠CF E, then CF is tangent to ω.
Solution 3: Note that AE is diameter of circumcircle of ABC since ∠CDF = 90◦ . From
∠AEF = ∠ADF = 45◦ it follows that triangle AF E is right-angled and isosceles. Without loss of
generality, let points A, E and F have coordinates (−1, 0), (1, 0) and (0, 1) respectively. Points F , E,
B are collinear, hence B have coordinates (b, 1 − b) for some b = −1. Let point C be intersection of
line tangent to circumcircle of AF E at F with line ED. Thus C have coordinates (c, 1) and from
C E ⊥ AB we get c = 2b/(b + 1). Now vector BC = (2b/(b + 1) − b, b) = b/(b + 1) · (1 − b, b + 1),
vector BF = (−b, b) = (−1, 1) · b and vector BA = (−(b + 1), −(1 − b)). Its clear that (1 − b, b + 1)
and (−(b + 1), −(1 − b)) are symmetric with respect to F E = (−1, 1), hence BF bisects ∠C BA and
C = C which completes the proof.
Solution 4: Again F lies on the perpendicular bisector of segment
M
AE, so AF E is right-angled and isosceles. Let M be an intersection
of BC and AF. Note that AM B is isosceles since BF is a bisector
and altitude in this triangle. Thus BF is a symmetry line of AM B.
C Then ∠F DA = ∠F EA = ∠M EF = 45◦ , AF = F E = F M and
F
∠DAE = ∠EM C. Let us show that EC = CM. Indeed,
∠CEM = 180◦ − (∠AED + ∠F EA + ∠M EF ) = 90◦ − ∠AED =
ω E = ∠DAE = ∠EM C.
It follows that F M CE is a kite, since EF = F M and M C = CE.
45◦
A Hence ∠EF C = ∠CF M = ∠EDF = 45◦ , so F C is tangent to ω.
D B
Solution 5: Let the tangent to ω at F intersect CD at C . Let ∠ABF =
∠F BC = β. It follows that ∠C F E = 45◦ since C F is tangent. We
have
sin ∠BDC sin ∠DF C sin ∠F BC sin 90◦ sin(90◦ − β) sin β 2 sin β cos β
·
· = · · = = 1.
sin ∠CDF sin ∠C F B sin ∠CBD sin 45◦ sin 45◦ sin 2β sin 2β
So by trig Ceva on triangle BDF, lines F C , DC and BC are concurrent (at C), so C = C . Hence
CF is tangent to ω.
5
Problem 2. A domino is a 2 × 1 or 1 × 2 tile. Determine in how many ways exactly n2 dominoes
can be placed without overlapping on a 2n × 2n chessboard so that every 2 × 2 square contains at least
two uncovered unit squares which lie in the same row or column.
(Turkey)
2
2n
Solution: The answer is .
n
Divide the schessboard into 2 × 2 squares. There are exactly n2 such squares on the chessboard.
Each of these squares can have at most two unit squares covered by the dominos. As the dominos
cover exactly 2n2 squares, each of them must have exactly two unit squares which are covered, and
these squares must lie in the same row or column.
We claim that these two unit squares are covered by the same domino tile. Suppose that this is
not the case for some 2 × 2 square and one of the tiles covering one of its unit squares sticks out to
the left. Then considering one of the leftmost 2 × 2 squares in this division with this property gives a
contradiction.
Now consider this n × n chessboard consisting of 2 × 2 squares of the original board. Define A, B,
C, D as the following configurations on the original chessboard, where the gray squares indicate the
domino tile, and consider the covering this n × n chessboard with the letters A, B, C, D in such a
A= B= C= D=
way that the resulting configuration on the original chessboard satisfies the condition of the question.
Note that then a square below or to the right of one containing an A or B must also contain an
A or B. Therefore the (possibly empty) region consisting of all squares containing an A or B abuts
the lower right corner of the chessboard and is separated from the (possibly empty) region consisting
of all squares containing a C or D by a path which goes from the lower left corner to the upper right
corner of this chessboard and which moves up or right at each step.
A similar reasoning shows that the (possibly empty) region consisting of all squares containing an
A or D abuts the lower left corner of the chessboard and is separated from the (possibly empty) region
consisting of all squares containing a B or C by a path which goes from the upper left corner to the
lower right corner of this chessboard and which moves down or right at each step.
D D C C C C
D D C C C B
D D D B B B
D D D A A B
D D D A A B
D A A A A B
Therefore the n × n chessboard is divided by these two paths into four (possibly empty) regions
that consist respectively of all squares containing A or B or C or D. Conversely, choosing two such
paths and filling the four regions separated by them with As, Bs, Cs and Ds counterclockwise starting
at the bottom results in a placement of the dominos on the original board satisfying the condition of
the question. 2
2n 2n
As each of these paths can be chosen in ways, there are ways the dominos can be
n n
placed.
6
Problem 3. Let n, m be integers greater than 1, and let a1 , a2 , . . . , am be positive integers not
greater than nm . Prove that there exist positive integers b1 , b2 , . . . , bm not greater than n, such that
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) < n,
(USA)
Solution 1: Suppose without loss of generality that a1 is the smallest of the ai . If a1 ≥ nm − 1, then
the problem is simple: either all the ai are equal, or a1 = nm − 1 and aj = nm for some j. In the first
case, we can take (say) b1 = 1, b2 = 2, and the rest of the bi can be arbitrary, and we have
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) ≤ gcd(a1 + b1 , a2 + b2 ) = 1.
In the second case, we can take b1 = 1, bj = 1, and the rest of the bi arbitrary, and again
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) ≤ gcd(a1 + b1 , aj + bj ) = 1.
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) ≥ n.
Also, we have
gcd(a1 + b1 , a2 + b2 , . . . , am + bm ) ≤ a1 + b1 ≤ nm + n − 2.
Thus there are at most nm − 1 possible values for the greatest common divisor. However, there
are nm choices for the m-tuple (b1 , . . . , bm ). Then, by the pigeonhole principle, there are two m-tuples
that yield the same values for the greatest common divisor, say d. But since d ≥ n, for each i there
can be at most one choice of bi ∈ {1, 2, . . . , n} such that ai + bi is divisible by d — and therefore there
can be at most one m-tuple (b1 , b2 , . . . , bm ) yielding d as the greatest common divisor. This is the
desired contradiction.
Solution 2: Similarly to Solution 1 suppose that a1 ≤ nm −2. The gcd of a1 +1, a2 +1, a3 +1, . . . , am +1
is co-prime with the gcd of a1 + 1, a2 + 2, a3 + 1, . . . , am + 1, thus a1 + 1 ≥ n2 . Now change another 1
into 2 and so on. After m − 1 changes we get a1 + 1 ≥ nm which gives us a contradiction.
Solution 3: We will prove stronger version of this problem:
m−1
For m, n > 1, let a1 , . . . , am be positive integers with at least one ai ≤ n2 . Then there are
integers b1 , . . . , bm , each equal to 1 or 2, such that gcd(a1 + b1 , . . . , am + bm ) < n.
Proof: Suppose otherwise. Then the 2m−1 integers gcd(a1 + b1 , . . . , am + bm ) with b1 = 1 and bi = 1
or 2 for i > 1 are all pairwise coprime, since for any two of them, there is some i > 1 with ai + 1
appearing in one and ai + 2 in the other. Since each of these 2m−1 integers divides a1 + 1, and each
m−1 −1 m−1
is ≥ n with at most one equal to n, it follows that a1 + 1 ≥ n(n + 1)2 so a1 ≥ n2 . The same
is true for each ai , i = 1, . . . , n, a contradiction.
m−1
Remark: Clearly the n2 bound can be strengthened as well.
7
Problem 4. Determine whether there exists an infinite sequence a1 , a2 , a3 , . . . of positive integers
which satisfies the equality
√
an+2 = an+1 + an+1 + an
for every positive integer n.
(Japan)
Since each an is a positive integer we see that bn is positive integer for n ≥ 2 and the sequence (bn ) is
strictly increasing for n ≥ 3. Thus bn + bn−1 = (bn+1 − bn )(bn+1 + bn ) ≥ bn+1 + bn , whence bn−1 ≥ bn+1
– a contradiction to increasing of the sequence (bi ).
Thus we conclude that there exists no sequence (an ) of positive integers satisfying the given
condition of the problem.
Solution 2: Suppose that such a sequence exists. We will calculate its members one by one and get
a contradiction.
√ √
From the equality a3 = a2 + √a2 + a1 it follows that a3 > a2 . Denote positive
√ integers a3 + a2
by b and a3 by√a, then we have 2a > b. Since a4 = a + b and a5 = a + b + 2a + b are positive
integers, then 2a + b is positive integer.
√ √ √
Consider a6 = a + b + 2a + b + 2a + 2b + 2a + b. Number c = 2a + 2b + 2a + b must be
√
positive integer, obviously it is greater than 2a + b. But
√ √ √ √
( 2a + b + 1)2 = 2a + b + 2 2a + b + 1 = 2a + 2b + 2a + b + ( 2a + b − b) + 1 > c2 .
√ √
So 2a + b < c < 2a + b + 1 which is impossible.
Solutions 3: We will show that there is no sequence (an ) of positive integers which consists of N > 5
members and satisfies
√
an+2 = an+1 + an+1 + an (1)
for all n = 1, . . . , N − 2. Moreover, we will describe all such sequences with five members.
Since every ai is a positive integer it follows from (1) that there exists such positive integer k
(obviously k depends on n) that
an+1 + an = k2 . (2)
From (1) we have (an+2 − an+1 )2 = an+1 + an , consider this equality as a quadratic equation with
respect to an+1 :
a2n+1 − (2an+2 + 1)an+1 + a2n+2 − an = 0.
√
2an+2 + 1 ± D
Obviously its solutions are an+1 1,2 = , where
2
D = 4(an + an+2 ) + 1. (3)
8
From the last equality, using that an+1 and an+2 are positive integers, we conclude that D is a square
of some odd number i.e. D = (2m + 1)2 for some positive integer m ∈ N, substitute this into (3):
Now adding an to both sides of (1) and using (2) and (4) we get m(m + 1) = k2 + k whence m = k.
So
an + an+1 = k2 ,
(5)
an + an+2 = k2 + k
for some positive integer k (recall that k depends on n).
Write equations (5) for n = 2 and n = 3, then for some positive integers k and we get
⎧
⎪
⎪ a2 + a3 = k2 ,
⎪
⎪
⎨ a + a = k 2 + k,
2 4
2
(6)
⎪
⎪ 3
a + a4 = ,
⎪
⎪
⎩ 2
a3 + a5 = + .
2k2 − 2 + k 2 − k 2 + k 2 + 2 + k
a2 = , a3 = , a4 = , a5 = . (7)
2 2 2 2
From a2 < a4 we obtain k2 < 2 hence k < .
Consider a6 : √
a6 = a5 + a5 + a4 = a5 + 2 + + k .
Since 0 < k < l we have 2 < 2 + + k < ( + 1)2 . So a6 cannot be integer i.e. there is no such
sequence with six or more members.
To find all required sequences with five members we must find positive integers a2 , a3 , a4 and a5
which satisfy (7) for some positive integers k < . Its clear that k and must be of the same parity.
Vise versa, let positive integers k, be of the same parity and satisfy k < then from (7) we get
integers a2 , a3 , a4 and a5 then a1 = (a3 − a2 )2 − a2 and it remains to verify that a1 and a2 are positive
i.e. 2k2 + k > 2 and 2(2 − k2 − k)2 > 2k2 − 2 + k.
Solution 4: It is easy to see that (an ) is increasing for large enough n. Hence
√
an+1 < an + 2an (1)
and
an < an−1 + 2an−1 . (2)
Lets define bn = an + an−1 . Using AM-QM inequality we have
√ √
2an + 2an−1 2an + 2an−1
≤ . (3)
2 2
Adding (1), (2) and using (3):
√
bn+1 < bn + 2an + 2an−1 ≤ bn + 2 bn .
9
Problem 5. Let m, n be positive integers with m > 1. Anastasia partitions the integers 1, 2, . . . , 2m
into m pairs. Boris then chooses one integer from each pair and finds the sum of these chosen integers.
Prove that Anastasia can select the pairs so that Boris cannot make his sum equal to n.
(Netherlands)
For each Pj we will compute the possible values for the expression s = a1 +. . .+am , where ai ∈ Pj,i
are the chosen integers. Here, Pj,i denotes the i-th coordinate of the ordered partition Pj .
We will denote by σ the number m 2
i=1 i = (m + m)/2.
• Consider the partition P1 and a certain choice with corresponding sum s. We find that
m
m
m2 = (2i − 1) ≤ s ≤ 2i = m2 + m.
i=1 i=1
• Consider the partition P2 and a certain choice with corresponding s. We find that
m
s≡ i≡σ (mod m).
i=1
ai ≡ i − di (mod m).
Hence,
m
m
s≡ ai ≡ (i − di ) ≡ σ − d (mod m),
i=1 i=1
which can only be congruent to σ modulo m if all di are equal, which forces s = (m2 + m)/2 or
s = (3m2 + m)/2. Since m > 1, it holds that
m2 + m 3m2 + m
< m2 < m2 + m < .
2 2
Hence if m2 ≤ n ≤ m2 + m and n ≡ σ (mod m), then s cannot be equal to n, so partition P3
suffices for such n.
10
Note that all n are treated in one of the cases above, so we are done.
Common notes for solutions 1B and 1C: Given the analysis of P1 and P2 as in the solution 1A,
we may conclude (noting that σ ≡ m(m + 1)/2 (mod m)) that if m is odd then m2 and m2 + m are
m
the only candidates for counterexamples n, while if m is even then m2 + is the only candidate.
2
There are now various ways to proceed as alternatives to the partition P3 .
Solution 1B: Consider the partition ({1, m+2}, {2, m+3}, . . . , {m−1, 2m}, {m, m+1}). We consider
possible sums mod m + 1. For the first m − 1 pairs, the elements of each pair are congruent mod m + 1,
1
so the sum of one element of each pair is (mod m + 1) congruent to m(m + 1)− m, which is congruent
2
m+1
to 1 if m + 1 is odd and 1 + if m + 1 is even. Now the elements of the last pair are congruent
2
to −1 and 0, so any achievable value of n is congruent to 0 or 1 if m + 1 is odd, and to 0 or 1 plus
m+1 m m
if m + 1 is even. If m is even then m2 + ≡ 1 + , which is not congruent to 0 or 1. If m is
2 2 2
m+1
odd then m2 ≡ 1 and m2 + m ≡ 0, neither of which can equal 0 or 1 plus .
2
Solution 1C: Similarly, consider the partition ({1, m}, {2, m + 1}, . . . , {m − 1, 2m − 2}, {2m − 1, 2m}),
this time considering sums of elements of pairs mod m − 1. If m − 1 is odd, the sum is congruent to
m−1 m m
1 or 2; if m − 1 is even, to 1 or 2 plus . If m is even then m2 + ≡ 1 + , and this can only
2 2 2
be congruent to 1 or 2 when m = 2. If m is odd, m2 and m2 + m are congruent to 1 and 2, and these
m−1
can only be congruent to 1 or 2 plus when m = 3. Now the cases of m = 2 and m = 3 need
2
considering separately (by finding explicit partitions excluding each n).
Solution 2: This solution does not use modulo arguments. Use only P1 from the solution 1A to
conclude that m2 ≤ n ≤ m2 + m. Now consider the partition ({1, 2m}, {2, 3}, {4, 5}, . . . , {2m −
2, 2m − 1}). If 1 is chosen from the first pair, the sum is at most m2 ; if 2m is chosen, the sum
is at least m2 + m. So either n = m2 or n = m2 + m. Now consider the partition ({1, 2m −
1}, {2, 2m}, {3, 4}, {5, 6}, . . . , {2m − 3, 2m − 2}). Sums of one element from each of the last m − 2
pairs are in the range from (m − 2)m = m2 − 2m to (m − 2)(m + 1) = m2 − m − 2 inclusive. Sums
of one element from each of the first two pairs are 3, 2m + 1 and 4m − 1. In the first case we have
n ≤ m2 −m+1 < m2 , in the second m2 +1 ≤ n ≤ m2 +m−1 and in the third n ≥ m2 +2m−1 > m2 +m.
So these three partitions together have eliminated all n.
11
Problem 6. Let H be the orthocenter and G be the centroid of acute-angled triangle ABC with
AB = AC. The line AG intersects the circumcircle of ABC at A and P. Let P be the reflection of
P in the line BC. Prove that ∠CAB = 60◦ if and only if HG = GP .
(Ukraine)
ω ω
P P
M
O O
G
A H
Solution 1: Let ω be the circumcircle of ABC. Reflecting ω in line BC, we obtain circle ω which,
obviously, contains points H and P . Let M be the midpoint of BC. As triangle ABC is acute-angled,
then H and O lie inside this triangle.
Let us assume that ∠CAB = 60◦ . Since
hence O lies on ω . Reflecting O in line BC, we obtain point O which lies on ω and this point is
the center of ω . Then OO = 2OM = 2R cos ∠CAB = AH, so AH = OO = HO = AO = R,
where R is the radius of ω and, naturally, of ω . Then quadrilateral AHO O is a rhombus, so A and
O are symmetric to each other with respect to HO. As H, G and O are collinear (Euler line), then
∠GAH = ∠HO G. Diagonals of quadrilateral GOP O intersects at M. Since ∠BOM = 60◦ , so
MB
OM = M O = ctg 60◦ · M B = √ .
3
12
C
A
P P
T K
ω
O M O
G
A H
follows that quadrilateral P M O A is cyclic, then ∠O M A = ∠O P A = ∠O A P . Let OM and
P A intersect at T. Triangles T O A and A O M are similar, hence O A /O M = O T /O A . In
the other words, O M · O T = O A2 . Using Menelaus’ theorem for triangle HKA and line T O , we
obtain that
A O HO KT KT
· ·
=3· = 1.
O H OK T A T A
It follows that KT /T A = 1/3 and KA = 2KT. Using Menelaus’ theorem for triangle T O A and line
HK we get that
O H A K T O 1 TO TO
1=
· ·
= ·2·
= .
HA KT OO 2 OO OO
It means that T O = OO , so O A2 = O M · O T = OO2 . Hence O A = OO and, consequently,
O ∈ ω . Finally we conclude that 2∠CAB = ∠BOC = 180◦ − ∠CAB, so ∠CAB = 60◦ .
ω H F G O
ε
P P
O D O
δ D
G
G
F
ε
A H H O
Solution 2: Let O and G denote the reflection of O and G, respectively, with respect to the line
BC. We then need to show ∠CAB = 60◦ iff G H = G P. Note that H OP is isosceles and hence
13
G H = G P is equivalent to G lying on the bisector ∠H OP. Let ∠H AP = ε. By the assumption
AB = AC, we have ε = 0. Then ∠H OP = 2∠H AP = 2ε, hence G H = G P iff ∠G OH = ε. But
∠GO H = ∠G OH . Let D be the midpoint of OO . It is known that ∠GDO = ∠GAH = ε. Let F be
the midpoint of HG. Then HG = F O (Euler line). Let ∠GO H = δ. We then have to show δ = ε iff
∠CAB = 60◦ . But by similarity (GDO ∼ F O O) we have ∠F O O = ε. Consider the circumcircles
of the triangles F O O and GO H. By the sine law and since the segments HG and F O are of equal
length we deduce that the circumcircles of the triangles F O O and GO H are symmetric with respect
to the perpendicular bisector of the segment F G iff δ = ε. Obviously, O is the common point of
these two circles. Hence O must be fixed after the symmetry about the perpendicular bisector of the
segment F G iff δ = ε so we have ε = δ iff HOO is isosceles. But HO = H O = R, and so
R 1
ε = δ ⇐⇒ OO = R ⇐⇒ OD = ⇐⇒ cos ∠CAB = ⇐⇒ ∠CAB = 60◦ .
2 2
Solution 3: Let H and G denote the reflection of points H and G with respect to the line BC. It
is known that H belongs to the circumcircle of ABC. The equality HG = GP is equivalent to
H G = G P. As in the Solution 2, it is equivalent to the statement that point G belongs to the
perpendicular bisector of H P, which is equivalent to OG ⊥ H P , where O is the circumcenter of
ABC.
Let points A(a), B(b), and C(c = −b) belong to the unit circle in the complex plane. Point G
have coordinate g = (a + b − b)/3. Since BC is parallel to the real axis point H have coordinate
h = a = 1/a.
p−a
Point P (p) belongs to the unit circle, so p = 1/p. Since a, p, g are collinear we have =
g−a
p−a g−a
. After computation we get p = . Since G (g) is the reflection of G with respect to the
g−a 1 − ga
chord BC, we have g = b + (−b) − b(−b)g = b − b + g. Let b − b = d. We have d = −d. So
since h and p belong to the unit circle (note that H = P because AB = AC). This is equivalent
to g = g h p and from (1), after easy computations, this is equivalent to a2 g2 + a2 + d2 + 1 =
(a2 + 1)(d2 + 1) = 0.
We cannot have a2 + 1 =√0, because then
√ a = ±i, but AB √ = AC. Hence
√ d = b − b = ±i, and the
pair {b, c = −b} is either {− 3/2 + i/2, 3/2 + i/2} or {− 3/2 − i/2, 3/2 − i/2}. Both cases are
equivalent to ∠BAC = 60◦ which completes the proof.
14
EGMO 2016, Day 1 – Solutions
where xn+1 = x1 .
min (x2k + x2k+1 ) ≤ x2j + x2j+1 ≤ 2x2j+1 ≤ 2xj+1 xj+2 ≤ max (2xk xk+1 ).
k=1,...,n k=1,...,n
−b, b, a, b, a, . . . , b, a.
M
C1
D
E
D1
X
F
A B
Solution. We are to prove that ∠EXY = ∠EF X; alternatively, but equi-
valently, ∠AY X + ∠XAY = ∠BY F + ∠XBY .
Since the quadrangle ABCD is cyclic, the triangles XAD and XBC are
similar, and since AD1 and BC1 are corresponding medians in these triangles,
it follows that ∠XAY = ∠XAD1 = ∠XBC1 = ∠XBY .
Finally, ∠AY X = ∠BY F , since X and M are corresponding points in the
similar triangles ABY and C1 D1 Y : indeed, ∠XAB = ∠XDC = ∠M C1 D1 ,
and ∠XBA = ∠XCD = ∠M D1 C1 .
2
and count blue cells to get mr1 + mr2 + mr3 < 6m. Subtraction of the latter
from the former multiplied by 3/2 yields mr1 − mr3 > mr2 /2 ≥ 0, and the
conclusion follows.
•
• •
• • •
• • • • •
• • • • • • • • • • • •
3
• • •
• • •
• • • •
• •
• •
• • • • • •
4
EGMO 2016, Day 2 – Solutions
Solution 1. Let the line Xk Tk and ω meet again at Xk0 , k = 1, 2, and notice
that the tangent tk to ωk at Xk and the tangent t0k to ω at Xk0 are parallel.
Since the ωk have equal radii, the tk are parallel, so the t0k are parallel, and
consequently the points X10 and X20 coincide (they are not antipodal, since
they both lie on the same side of the line T1 T2 ). The conclusion follows.
1 t1 2 X1
X1
T2
2
T1 T2
t2
X2 Y
X2 T1 1
Solution 3. Invert from X1 and use an asterisk to denote images under this
inversion. Notice that ωk∗ is the tangent from X2∗ to ω ∗ at Tk∗ , and the pole
X1 lies on the bisectrix of the angle formed by the ωk∗ , not containing ω ∗ .
Letting X1 T1∗ and ω ∗ meet again at Y , standard angle chase shows that Y
lies on the circle X1 X2∗ T2∗ , and the conclusion follows.
2
and is achieved, for instance, by the maximal arrangement consisting of the
vertical tile [0, 1] × [1, k + 1] along with k − 1 horizontal tiles stacked in
[1, k+1]×[0, k−1], another k−1 horizontal tiles stacked in [1, k+1]×[k+1, 2k],
and two horizontal tiles stacked in [k, 2k] × [k − 1, k + 1]. This example shows
that the corresponding minimum does not exceed n + 1 < 2n − 2k + 2. The
argument in the solution also applies to the case n = 2k to infer that for a
maximal arrangement of minimal cardinality either r = 0 or c = 0, and the
cardinality is at least n. Clearly, we may and will assume r = 0. Suppose,
if possible, such an arrangement contains exactly n tiles. Then each row
contains exactly one tile, and there are no vertical tiles. Since there is no
room left for an additional tile, some tile T must cover a cell of the leftmost
column, so it covers the k leftmost cells along its row, and there is then room
for another tile along that row — a contradiction.
(2) For every pair (r, c) of integers in the range 2k − n, . . ., k − 1, at least
one of which is positive, say c > 0, there exists a maximal arrangement of
cardinality 2n − r − c.
Use again the board [0, n] × [0, n] to stack k − r horizontal tiles in each of
the rectangles [0, k]×[0, k−r] and [k−c, 2k−c]×[k, 2k−r], erect k−c vertical
tiles in each of the rectangles [0, k − c] × [k − r, 2k − r] and [k, 2k − c] × [0, k],
then stack n−2k +r horizontal tiles in the rectangle [k −c, 2k −c]×[2k −r, n],
and erect n − 2k + c vertical tiles in the rectangle [2k − c, n] × [1, k + 1].
Problem 6. Let S be the set of all positive integers n such that n4 has a
divisor in the range n2 + 1, n2 + 2, . . ., n2 + 2n. Prove that there are infinitely
many elements of S of each of the forms 7m, 7m + 1, 7m + 2, 7m + 5, 7m + 6
and no elements of S of the form 7m + 3 or 7m + 4, where m is an integer.
The former is a Pell equation whose solutions are (m1 , n1 ) = (3, 2) and
3
(m3 , n3 ) ≡ (1, 0), it follows that S contains infinitely many integers from
each of the residue classes 0 and ±2 modulo 7.
The other equation is easily transformed into a Pell equation, m02 −
12n02 = 1, by noticing that m and n are both divisible by 3, say m = 3m0
and n = 3n0 . In this case, the solutions are (m1 , n1 ) = (21, 6) and
This time iteration shows that (mk+4 , nk+4 ) ≡ (mk , nk ). Since (m1 , n1 ) ≡
(0, −1), (m2 , n2 ) ≡ (−3, 0), (m3 , n3 ) ≡ (0, 1), and (m4 , n4 ) ≡ (3, 0), it follows
that S contains infinitely many integers from each of the residue classes 0
and ±1 modulo 7.
Finally, since the nk from the two sets of formulae exhaust S, by the
preceding no integer in the residue classes ±3 modulo 7 is a member of S.
We now turn to the proof of the lemma. Let n be a member of S, and
let d = n2 + m be a divisor of n4 in the range n2 + 1, n2 + 2, . . . , n2 + 2n, so
1 ≤ m ≤ 2n. Consideration of the square of n2 = d − m shows m2 divisible
by d, so m2 /d is a positive integer. Since n2 < d < (n + 1)2 , it follows that
d is not a square; in particular, m2 /d 6= 1, so m2 /d ≥ 2. On the other hand,
1 ≤ m ≤ 2n, so m2 /d = m2 /(n2 + m) ≤ 4n2 /(n2 + 1) < 4. Consequently,
m2 /d = 2 or m2 /d = 3; that is, m2 /(n2 + m) = 2 or m2 /(n2 + m) = 3. In
the former case, 2n2 + 1 = (m − 1)2 , and in the latter, 12n2 + 9 = (2m − 3)2 .
Conversely, if 2n2 + 1 = m2 for some positive integer m, then 1 < m2 <
4n , so 1 < m < 2n, and n4 = (n2 + m + 1)(n2 − m + 1), so the first factor
2
4
Problem 1
Let ABCD be a convex quadrilateral with ∠DAB = ∠BCD = 90◦ and ∠ABC > ∠CDA. Let Q and R
be points on the segments BC and CD, respectively, such that the line QR intersects lines AB and AD at
points P and S, respectively. It is given that P Q = RS. Let the midpoint of BD be M and the midpoint of
QR be N . Prove that M, N, A and C lie on a circle.
Solution 1
Note that N is also the midpoint of P S. From right-angled triangles P AS and CQR we obtain ∠AN P =
2∠ASP , ∠CN Q = 2∠CRQ, hence ∠AN C = ∠AN P +∠CN Q = 2(∠ASP +∠CRQ) = 2(∠RSD+∠DRS) =
2∠ADC.
Similarly, using right-angled triangles BAD and BCD, we obtain ∠AM C = 2∠ADC.
Thus ∠AM C = ∠AN C, and the required statement follows.
Solution 2
In this proof we show that we have ∠N CM = ∠N AM instead. From right-angled triangles BCD and QCR
we get ∠DRS = ∠CRQ = ∠RCN and ∠BDC = ∠DCM . Hence ∠N CM = ∠DCM − ∠RCN . From right-
angled triangle AP S we get ∠P SA = ∠SAN . From right-angled triangle BAD we have ∠M AD = ∠BDA.
Moreover, ∠BDA = ∠DRS + ∠RSD − ∠RDB.
Therefore ∠N AM = ∠N AS − ∠M AD = ∠CDB − ∠DRS = ∠N CM , and the required statement follows.
Solution 3
As N is also the midpoint of P S, we can shrink triangle AP S to a triangle A0 QR (where P is sent to Q and
S is sent to R). Then A0 , Q, R and C lie on a circle with center N . According to the shrinking the line A0 R
is parallel to the line AD. Therefore ∠CN A = ∠CN A0 = 2∠CRA0 = 2∠CDA = ∠CM A. The required
statement follows.
1
Problem 2
Find the smallest positive integer k for which there exist a colouring of the positive integers Z>0 with k
colours and a function f : Z>0 → Z>0 with the following two properties:
(i) For all positive integers m, n of the same colour, f (m + n) = f (m) + f (n).
(ii) There are positive integers m, n such that f (m + n) 6= f (m) + f (n).
In a colouring of Z>0 with k colours, every integer is coloured in exactly one of the k colours. In both (i)
and (ii) the positive integers m, n are not necessarily different.
Merlijn Staps, the Netherlands
Solution 1:
The answer is k = 3.
First we show that there is such a function and coloring for k = 3. Consider f : Z>0 → Z>0 given by f (n) = n
for all n ≡ 1 or 2 modulo 3, and f (n) = 2n for n ≡ 0 modulo 3. Moreover, give a positive integer n the i-th
color if n ≡ i (3).
By construction we have f (1 + 2) = 6 6= 3 = f (1) + f (2) and hence f has property (ii).
Now let n, m be positive integers with the same color i. If i = 0, then n + m has color 0, so f (n + m) =
2(n + m) = 2n + 2m = f (n) + f (m). If i = 1, then n + m has color 2, so f (n + m) = n + m = f (n) + f (m).
Finally, if i = 2, then n + m has color 1, so f (n + m) = n + m = f (n) + f (m). Therefore f also satisfies
condition (i).
Proof. Suppose that b < a is an odd integer, and that a and b have the same color. Then on the one hand,
f (a + b) = f (a) + bf (1). On the other hand, we also have f (a + b) = f ( a+b a+b a+b
2 ) + f ( 2 ) = (a + b)f (1), as 2
is a positive integer smaller than a. Hence f (a) = f (a + b) − bf (1) = (a + b)f (1) − bf (1) = af (1), which is
again a contradiction. Therefore all odd integers smaller than a have a color different from that of a.
Lemma 2. Any even integer b < a has the same color as a
Proof. Suppose b < a is an even integer, and that a and b have different colors. Then a − b is an odd integer
smaller than a, so it has the same color as b. Thus f (a) = f (a − b) + f (b) = (a − b)f (1) + bf (1) = af (1), a
contradiction. Hence all even integers smaller than a have the same color as a.
Suppose now a + 1 has the same color as a. As a > 1, we have a+1 a+1
2 < a and therefore f (a + 1) = 2f ( 2 ) =
(a + 1)f (1). As a − 1 is an even integer smaller than a, we have by Lemma 2 that a − 1 also has the same
color as a. Hence 2f (a) = f (2a) = f (a + 1) + f (a − 1) = (a + 1)f (1) + (a − 1)f (1) = 2af (1), which implies
that f (a) = af (1), a contradiction. So a and a + 1 have different colors.
Since a − 2 is an odd integer smaller than a, by Lemma 1 it has a color different from that of a, so a − 2
and a + 1 have the same color. Also, we have seen by Lemma 2 that a − 1 and a have the same color. So
f (a) + f (a − 1) = f (2a − 1) = f (a + 1) + f (a − 2) = (a + 1)f (1) + (a − 2)f (1) = (2a − 1)f (1), from which
it follows that f (a) = (2a − 1)f (1) − f (a − 1) = (2a − 1)f (1) − (a − 1)f (1) = af (1), which contradicts our
choice of a and finishes the proof.
1
Solution 2:
We prove that k ≤ 3 just as in first solution.
Next we show that there is no such function and coloring for k = 2.
Consider any coloring of Z>0 with 2 colors and any function f : Z>0 → Z>0 satisfying conditions (i) and (ii).
We first notice with m = n that f (2n) = 2f (n).
Lemma 3. For every n ∈ Z>0 , f (3n) = 3f (n) holds.
Proof. Define c = f (n), d = f (3n). Then we have the relations
• If n and 2n have the same color, then f (3n) = f (n) + f (2n) = 3c = 3f (n).
• If n and 3n have the same color, then 4c = f (4n) = f (n) + f (3n) = c + f (3n), so f (3n) = 3f (n).
• If 2n and 4n have the same color, then 2d = f (6n) = f (2n) + f (4n) = 2c + 4c = 6c, so f (3n) = d = 3c.
• Otherwise n and 4n have the same color, and 2n and 3n both have the opposite color to n. Therefore
we compute 5c = f (n) + f (4n) = f (5n) = f (2n) + f (3n) = 2c + f (3n) so f (3n) = 3f (n).
Consequently, for k = 2 we necessarily have f (3n) = 3f (n).
Now let a be the smallest integer such that f (a) 6= af (1). In particular a is odd and a > 3. Consider the
three integers a, a−3 a+3
2 , 2 . By pigeonhole principle two of them have the same color.
• If a and a−32 have the same color, then 3 a−1 a−1 3a−3 a−3
2 f (1) = 3f ( 2 ) = f ( 2 ) = f (a) + f ( 2 ) = f (a) +
a−3
2 f (1), so f (a) = af (1).
• If a and a+32 have the same color, then 3 a+1 a+1 3a+3 a+3
2 f (1) = 3f ( 2 ) = f ( 2 ) = f (a) + f ( 2 ) = f (a) +
a+3
2 f (1), so f (a) = af (1).
In the three cases we find a contradiction with f (a) 6= af (1), so it finishes the proof.
Solution 3:
As before we prove that k ≤ 3 and for any such function and colouring we have f (2n) = 2f (n).
Now we show that there is no such function and coloring for k = 2.
Consider any coloring of Z>0 with 2 colors and any function f : Z>0 → Z>0 satisfying conditions (i) and (ii).
Say the two colors are white (W) and black (B). Pick m, n any two integers such that f (m+n) = f (m)+f (n).
Without loss of generality we may assume that m + n, m are black and n is white.
Lemma 4. For all l ∈ Z>0 and every x whose color is black, we have x + lm is black and f (x + lm) =
f (x) + lf (m).
Proof. We proceed by induction. It is clearly true for l = 0. If x + lm is black and satisfies f (x + lm) =
f (x) + lf (m), then f (x + (l + 1)m) = f (x + lm) + f (m) = f (x) + (l + 1)f (m) and f (x + (l + 1)m + n) =
f (x + lm) + f (m + n) = f (x) + lf (m) + f (m + n) 6= f (x) + (l + 1)f (m) + f (n) = f (x + (l + 1)m) + f (n), so
x + (l + 1)m is not the same color of n, therefore x + (l + 1)m is black. Thjs completes the induction.
In particular we then must have that 2l n is white for every l, because otherwise since 2l m is black we would
have 2l f (m + n) = f (2l m + 2l n) = f (2l m) + f (2l n) = 2l (f (m) + f (n)), and consequently f (m + n) =
f (m) + f (n).
Lemma 5. For every l ≥ 1, 2l m + 2l−1 n is black.
2
Proof. On the one hand we have 2l f (m + n) = f (2l m + 2l n) = f (2l−1 (2m + n) + 2l−1 n). On the other hand
we have
2l f (m+n) = 2l−1 ·2f (m+n) 6= 2l−1 (f (m+n)+f (m)+f (n)) = 2l−1 (f (2m+n)+f (n)) = f (2l m+2l−1 n))+f (2l−1 n).
Therefore 2l m + 2l−1 n and 2l−1 n have different color, which means 2l m + 2l−1 n is black.
Combining the two lemmas give jm + 2l−1 n is black for all j ≥ 2l and every l ≥ 1.
t t
Now write m = 2l−1 m0 with m0 odd. Let t be a number such that 2m−1 0 is an integer and j = 2m−1 l
0 n ≥ 2 , i.e.
0 l−1 l−1
t is some multiple of φ(m ). Then we must have that jm + 2 n is black, but by definition jm + 2 n =
(2t − 1)2l−1 n + 2l−1 n = 2t+l−1 n is white. This is a contradiction, so k = 2 is impossible.
3
Problem 3
There are 2017 lines in a plane such that no 3 of them go through the same point. Turbo the snail can slide
along the lines in the following fashion: she initially moves on one of the lines and continues moving on a
given line until she reaches an intersection of 2 lines. At the intersection, she follows her journey on the other
line turning left or right, alternating the direction she chooses at each intersection point she passes. Can it
happen that she slides through a line segment for a second time in her journey but in the opposite direction
as she did for the first time?
Márk Di Giovanni, Hungary
1. Solution
We show that this is not possible.
The lines divide the plane into disjoint regions. We claim that there exists an alternating 2-coloring of these
regions, that is each region can be colored in black or white, such that if two regions share a line segment,
they have a different color. We show this inductively.
If there are no lines, this is obvious. Consider now an arrangement of n lines in the plane and an alternating
2-coloring of the regions. If we add a line g, we can simply switch the colors of all regions in one of the half
planes divided by g from white to black and vice versa. Any line segment not in g will still be between two
regions of different color. Any line segment in g cuts a region determined by the n lines in two, and since we
switched colors on one side of g this segment will also lie between two regions of different color.
Now without loss of generality we may assume, that Turbo starts on a line segment with a white region on
her left and a black one on her right. At any intersection, if she turns right, she will keep the black tile to
her right. If she turns left, she will keep the white tile to her left. Thus wherever she goes, she will always
have a white tile on her left and a black tile on her right. As a consequence, she can cross every line segment
only in the direction where she has a white tile to her left, and never the opposite direction where she would
have a black tile to the left.
2. Solution
Suppose the assumption is true.
Let’s label each segment in the snail’s path with L or R depending on the direction that Turbo chose at the
start point of this segment (one segment can have several labels if it has been visited several times).
Consider the first segment that has been visited twice in different directions, name this segment s1 . Assume
without loss of generality that it is labeled with L. Then next segment must be labeled with R, name this
one s2 .
Let’s look at the label which s1 can get on the second visit. If it gets L then the previous segment in the
path must be s2 . But in this case s1 is not the first segment that has been visited twice in different directions
because s2 has been visited earlier. So the second label of s1 must be R, and Turbo must have come from
the opposite side of s2 .
1
Since Turbo alters directions at each point, labels in her path also alter. And because two labels of s1 are
different, the number of visited segments between these two visits must be even.
Now let’s make the following observation: each segment in the path corresponds to exactly one line, and its
previous and next segments are on opposite sides of this line.
3. Solution
Suppose that the snail always slides slightly to the right of the line segments on her path. When turning to
the right, she does not cross any line, whereas when turning to the left, she crosses exactly two lines. This
means that at any time of her journey, she has crossed an even number of lines.
Assuming that at some point she slides along a segment for the second time, but in the opposite direction,
we argue that she needs to cross an odd number of lines. Let ` be the line on which the revisit happens. In
order to get to the other side of `, the snail has to cross ` an odd number of times. To visit the same segment
of `, she must cross every other line an even number of times.
4. Solution
Let us color in red all intersection points of the given lines and let us choose one of two possible directions
on each segment (draw an arrow on each segment). Consider a red point R where two given lines a and b
meet, and the four segments a1 , a2 , b1 , b2 with endpoint R (so that ai ⊂ a, bj ⊂ b). R is called a saddle if
on a1 , a2 the arrows go out of R while on b1 , b2 the arrows enter R, or visa versa, on b1 , b2 the arrows go
out of R while on a1 , a2 the arrows enter R. The set of arrows (chosen on all segments) is said to be good if
all red points are saddles. It is sufficient to prove that there exists a good set of arrows. Indeed, if initially
Turbo is moving along (or opposite) the arrow, then this condition holds after she turns at a red point.
The given lines cut the plane into regions. Further we need the following property of the good set of arrows
(this property directly follows from the definition): the boundary of any bounded region is a directed cycle
of arrows; the boundary of any unbounded region is a directed chain of arrows.
We construct a good set of arrows by induction on n with trivial base n = 1. Now erase one of n given lines
and assume we have a good set of arrows for remaining n − 1 lines. Now restore the n-th line `, assume that `
is horizontal. Denote by A1 , . . . , An−1 all new red points on ` from the left to the right. Each of Ai belongs
to some old segment mi of the line `i . Let us call Ai ascending if the arrow on mi goes up, and descending
if the arrow on mi goes down. Consider the region containing the segment Ai Ai+1 . By the property, Ai and
Ai+1 can not be both ascending or both descending. Thus we can choose arrows on all pieces of ` so that
each arrow goes from a descending to an ascending vertex.
2
Each of points Ai cuts mi into two new pieces; the direction of new pieces supposed to be the same as on
mi . Now simultaneously change the direction of arrows on all pieces below the line `. It is easy to see that
A1 , . . . , An−1 become saddles, while the other red points remain saddles. This completes the induction step.
3
Problem 4
Let n ≥ 1 be an integer and let t1 < t2 < . . . < tn be positive integers. In a
group of tn + 1 people, some games of chess are played. Two people can play
each other at most once. Prove that it is possible for the following conditions
to hold at the same time:
i) The number of games played by each person is one of t1 , t2 , . . . , tn ,
ii) For every i with 1 ≤ i ≤ n, there is someone who has played exactly ti
games of chess.
Comment
In graph theory terms the problem is to prove that for any finite nonempty set
T of positive integers there exists a graph of size max T + 1 such that the degree
set of the graph is equal to T .
Among graph theory specialists a generalization of this problem is known
[1]. Nevertheless, the problem still suited the contest.
2. Solution
Let T = {t1 , . . . , tn }. The proof proceeds by induction on n = |T |. If n = 1
and T = {t}, we choose a group of t + 1 people such that everyone plays with
1
everyone else. If n = 2 and T = {t1 , t2 } with t1 < t2 , divide the t2 + 1 people
into groups A resp. B of size t1 resp. t2 − t1 + 1 such that everyone from
group A played with everyone else whereas people from group B only played
with the people from group A. Then the people from group A resp. B played
with exactly t2 resp. t1 other people. In the inductive step, suppose that T has
n > 2 elements t1 < . . . < tn . Consider the set
3. Solution
The proof proceeds by induction on |tn |. If tn = 1 we have n = 1 and we
can consider two persons that play against each other. Then every player has
played 1 game and the conditions of the problem are satisfied. If tn > 1 we
distinguish the two cases t1 > 1 and t1 = 1. If t1 > 1 there exists, by the
induction hypothesis, a group A of size tn that satisfies the conditions of the
problem for t01 = t1 − 1, . . . , t0n = tn − 1. Now add a new person to A and let
him/her play against everyone from A. The new group will be of size tn + 1
and there exists a person which has played t games if and only if there exists a
person that has played t − 1 games within A, i.e. if and only if t ∈ {t1 , . . . , tn }.
Hence the conditions of the problem are satisfied.
If t1 = 1 there exists, by the induction hypothesis, a group B of size tn−1
that satisfies the conditions of the problem for t01 = t2 − 1, . . . , t0n−2 = tn−1 − 1.
Now add a new person P and let him/her play with everyone from group B and
a group C of size tn − tn−1 > 0 and let them play with P . The new group will
be of size tn−1 + 1 + (tn − tn−1 ) + 1 = tn + 1. Since person P has played against
everyone he will have played tn games. The people in C will have played 1 = t1
games. There exists a person in B that has played t games if and only if there
exist a person in B that has played t − 1 games within B, i.e. if and only if
t ∈ {t2 , . . . , tn−1 }. Hence the conditions of the problem are satisfied.
4. Solution
We generalize the construction for T = {1, ..., n}
Construction
Take sets of people A1 , ..., An . Let all people of Ai play chess with all people in
Aj with j ≥ n − i + 1
2
Now
P the number of games
P played by anyone in Ai is
( j≥n−i+1 |Aj |) or ( j≥n−i+1 |Aj |) − 1 if i ≥ n − i + 1.
Now if we start with one person in each Ai and two people in Ad n2 e . The number
of played games for anyone in Ai is equal to i. In particular this is a construction
for T = {1, .., n}
Now to get to numbers of general sets T of size n we can change the sizes of Ai
but keep the construction.
Variant 1
Variant 2
3
References
[1] Timothy A. Sipka. The orders of graphs with prescribed degree sets. Journal
of Graph Theory, 4(3):301–307, 1980.
[2] Amitabha Tripathi and Sujith Vijay. A short proof of a theorem on degree
sets of graphs. Discrete Appl. Math., 155(5):670–671, 2007.
4
Problem 5
Let n ≥ 2 be an integer. An n-tuple (a1 , a2 , . . . , an ) of positive integers is expensive if there exists a positive
integer k such that
(a1 + a2 )(a2 + a3 ) · · · · · (an−1 + an )(an + a1 ) = 22k−1 .
a) Find all positive integers n ≥ 2 for which there exists an expensive n-tuple.
b) Prove that for every positive integer m there exists an integer n ≥ 2 such that m belongs to an expensive
n-tuple.
There are exactly n factors in the product on the left hand side.
Solution 1
a) Notice that for odd integers n > 2, the tuple (1, 1, . . . , 1) is expensive. We will prove that there are no
expensive n-tuples for even n.
Lemma 0.1. If an expensive n-tuple exists for some n ≥ 4, then also an expensive n − 2-tuple.
Proof. In what follows all indices are considered modulo n. Let (a1 , a2 , . . . , an ) be an expensive n-tuple and
at the largest element of the tuple. We have the inequalities
Since both at−1 + at and at + at+1 are powers of 2 (they are divisors of a power of 2), we deduce from (1)
and (2)
at−1 + at = at + at+1 = 2r
for some positive integer r, and in particular at−1 = at+1 .
Consider now the n − 2-tuple (b1 , . . . , bn−2 ) obtained by removing at and at+1 from (a1 , a2 , . . . , an ). By what
we just said we have
n−2 Qn
i=1 (ai + ai+1 )
Y
(bi + bi+1 ) = = 22(k−r)−1 ,
i=1
(at−1 + at )(at + at+1 )
From the lemma we now conclude that if there exists an expensive n-tuple for some even n, then also an
expensive 2-tuple i.e.
(a1 + a2 )2 = 22k−1
for some positive integers a1 , a2 , which is impossible since the right hand side is not a square.
b) We prove this by induction. In a) we saw that 1 belongs to an expensive n-tuple. Assume now that all
odd positive integers less that 2k belong to an expensive n-tuple, for some k ≥ 1. Hence for any odd r < 2k
there is an integer n and an expensive n-tuple (a1 , . . . , r, . . . , an ). We notice that then also (a1 , . . . , r, 2k+1 −
r, r, . . . , an ) is expensive. Since 2k+1 − r can take all odd values between 2k and 2k+1 the induction step is
complete.
1
Solution 2
a) For odd n the tuple (1, 1, . . . , 1) is a solution.
Q
Now consider n even. Since the product (ai + ai+1 ) is a power of two, every factor needs to be a power
of two. We are going Qto prove that for all tuples (a1 , . . . , an ) such that ai + ai+1 is always a power of two
, it is the casePthat (ai + ai+1 ) is equal to an even power of two. We are going to prove this with strong
induction on ai . When all ai are equal to one this is certainly the case. Since ai + ai+1 > 1 it is even
and we conclude that the ai are either all odd or all even. In the case they are all even, thenP considerPthe
tuple (b1 , . . . , bn ) withQbi = ai /2. This tuple clearly satisfies the hypothesis as well and we have b i < ai .
Furthermore we have (ai + ai+1 ) = 2n (bi + bi+1 ) and since n is even we are done in this case.
Q
Now all ai are odd. Suppose none of are one, then consider the tuple (b1 , . . . , bn ) with bi = (ai + (−1)i )/2.
Since all ai are odd andP strictlyPlarger than
Q one, the bi are positive integers and satisfy bi +bi+1 = (ai +ai+1 )/2,
bi < ai and (ai + ai+1 ) = 2n (bi + bi+1 ) we are done in this case again. Now
Q
a power of two. Since
there is at least one ai being one. We may assume i = 1, because the condition is cyclic. Moreover we may
also assume that a2 > 1 since not all of the ai are equal to one. Let now k be the smallest index larger than
one such that ak is equal to one. We are not excluding the case k = n + 1, yet. Now for i = 1, . . . , k − 1
we have ai + ai+1 > 2 and thus divisible by four. By induction it easily follows that ai ≡ (−1)i+1 mod (4)
for i = 1, . . . , k − 1. In particular, since ak = 1 we find that k is odd and at least three. Now consider the
tuple (b1 , . . . , bn ) with bi = (ai − (−1)i )/2 for i = 1, . . . , k and bi = ai otherwise. This P is again
P a tuple that
satisfies the hypothesis, since b1 = a1 = 1 = b k = a k . Moreover b2 < a 2 and thus b i < ai . Finally we
have (ai + ai+1 ) = 2k−1 (bi + bi+1 ) and since k is odd we conclude the proof.
Q Q
b) We use some of the ideas from a). Consider the operators T± (n) = 2n ± 1. We claim that for every
odd integer m there is an integer r and signs i ∈ {+, −} for i = 1, . . . r such that Tr ◦ · · · ◦ T1 (1) = m.
This is certainly true for m = 1 and for m > 1 we find that m = T− ((m + 1)/2) if m ≡ 1 mod (4) and
m = T+ ((m − 1)/2) if m ≡ 3 mod (4). Note that both (m + 1)/2 and (m − 1)/2 are odd integers in their
respective cases and (m − 1)/2 ≤ (m + 1)/2 < m for m > 1. Therefore iterating the procedure will eventually
terminate in one.
For the construction it is most convenient to set n = 2l + 1 and label the tuple (a−l , a−l+1 , . . . , al ). For
m = 1 we have the expensive tuple (1, 1, 1). For m > 1 we will define operators T± on expensive tuples
with the condition a−l = al = 1 that give rise to a new expensive tuple (b−l0 , . . . , bl0 ) with b−l0 = bl0 = 1
and b0 = T± a0 . It is then clear that Tr ◦ · · · ◦ T1 ((1, 1, 1)) is an expensive tuple containing m. We define
T± as follows: set l0 = l + 1 and b−l0 = bl0 = 1 and bi = T±(−1)i (ai ) for i = −l, . . . , l. Here we identify +
Q +1 and − with2l −1.
with Q We are left to prove that the new tupleQ is indeed expensive. If ±(−1)l = −1, then
l 2l+2
Q
(bi + bi+1 ) = 4 · 2 (ai + ai+1 ), and if ±(−1) = +1, then (bi + bi+1 ) = 4 · 2 (ai + ai+1 ). In both
cases we end up with an expensive tuple again.
2
Problem 6
Let ABC be an acute-angled triangle in which no two sides have the same length. The
reflections of the centroid G and the circumcentre O of ABC in its sides BC, CA, AB are
denoted by G1 , G2 , G3 , and O1 , O2 , O3 , respectively. Show that the circumcircles of the tri-
angles G1 G2 C, G1 G3 B, G2 G3 A, O1 O2 C, O1 O3 B, O2 O3 A and ABC have a common point.
The centroid of a triangle is the intersection point of the three medians. A median is a
line connecting a vertex of the triangle to the midpoint of the opposite side.
X3
e2
e3 O3
A
G3
G2
H
X
O2
O
B X2
O1
C
G1
X1
e1
1
Now consider for example the circumcircles of O1 O2 C and G1 G2 C. By construction, it
is clear that ∠O2 CO1 = ∠G2 CG1 = 2∠ACB. Let G1 O1 and G2 O2 meet at X, and let e
meet e1 , e2 at E1 , E2 , respectively, as shown in the diagram below. Chasing angles,
E2
e2
e
A
G2
O2
X
G
O
B
E1
C
O1
G1
e1
2
e
X10 = X20
H
B
X2
X1
It thus remains to show that e1 , e2 , e3 are concurrent, and intersect on the circumcircle
of ABC. Let e meet the circumcircles of the triangles BCH, ACH, ABH at X1 , X2 , X3 ,
respectively. It is well known that the reflections of H in the sides of ABC lie on the
circumcircle of ABC. For this reason, the circumcircles of BCH, ACH, ABH have the
same radius as the circumcircle of ABC, and hence the reflections X10 , X20 , X30 of X1 , X2 , X3
in the sides [BC], [CA], [AB] lie on the circumcircle of ABC. By definition, X10 , X20 , X30
lie on e1 , e2 , e3 , respectively. It thus remains to show that they coincide.
To show that, for example, X10 = X20 , it will be sufficient to show that ∠X2 AC = ∠X10 AC,
since we have already shown that X10 and X20 lie on the circumcircle of ABC. But, chasing
angles in the diagram above,
where we have used the fact that BHCX1 and AHX2 C are cyclic by construction, and
the fact that ∠BHC = 180◦ − ∠BAC. This shows that X10 = X20 . Similarly, X10 = X30 ,
which completes the proof.
Remark. The statement of the problem remains true if O and G are replaced with two points
that are aligned with the orthocenter H of the triangle, and indeed, the proof above did not
require any property of G and O other than the fact that they lie on a line through H, the Euler
line.
3
Solution 2
The proof consists of two parts. First, we show that if P is any point inside the triangle
ABC and P1 , P2 , P3 are its reflections in the sides BC, CA, AB, then the circumcircles
of the triangles P1 P2 C, P1 P3 B, P2 P3 A intersect in a point TP on the circumcircle of the
triangle ABC. In the second part, we show that TG coincides with TO .
Now let P be any point inside the triangle ABC and let P1 , P2 , P3 be the reflections in
the sides as above. Let TP be the second intersection of the circumcircles of the triangles
P1 P2 C and ABC. We want to show that TP lies on the circumcircles of the triangles
P1 P3 B and P2 P3 A.
4
then T = TG = TO and we are done.
Let H be the orthocenter of the triangle ABC and let H2 be the reflection in the side
AC. It is known that H2 lies on the circumcircle of the triangle ABC. G, O, H lie on the
Euler line, so G2 , O2 , H2 are collinear as well. Let T be the second intersection of G2 H2
and the circumcircle of the triangle ABC. We can now complete the proof by seeing that
5
Solution 3 (complex numbers)
For every point P , let p denote the corresponding complex number. Set O to be the
origin, so o = 0, and without loss of generality we can assume that a, b and c lie on the
a+b+c
unit circle. Then the centroid can be expressed as g = .
3
The segments oo1 and bc have a common midpoint, so o1 + o = b + c, and then o1 = b + c.
Similarly o2 = a + c and o3 = a + b. In order to compute g1 , define y to be the projection
of g onto bc. Since b and c are on the unit circle, it is well known that y can be expressed
as
1
y = (b + c + g − bcḡ).
2
1 1 1
By using ā = , b̄ = and c̄ = (points on the unit circle), we obtain
a b c
ab + bc + ca
g1 = b + c − .
3a
ab + bc + ca ab + bc + ca
Similarly, we get g2 = a + c − and g3 = a + b − .
3b 3c
1) Proof that circumcircles of triangles abc, o1 o2 c, o1 o3 b and o2 o3 a have common point.
Let x be the point of intersection of circumcircles of triangles o1 o2 c and abc (x 6= c). We
x−c x − o2
know that x, o1 , o2 and c are concyclic if and only if : is real number,
o1 − c o1 − o2
which is equivalent to
x − c o1 − o2 o1 − c x − o2
· = · . (1)
x̄ − c̄ o¯1 − o¯2 o¯1 − c̄ x̄ − o¯2
x−c o1 − o2 b−a
Since x and c are on the unit circle = −xc. Also, = = −ab, and
x̄ − c̄ o¯1 − o¯2 b̄ − ā
o1 − c b 1
= = b2 . Since x̄ = , from (1) and previous relations, we have:
o¯1 − c̄ b̄ x
ab + bc + ca
x= .
a+b+c
This formula is symmetric, so we conclude that x also belongs to circumcircles of o1 o3 b
and o2 o3 a.
2) Proof that x belongs to circumcircles of g1 g2 c, g1 g3 b and g2 g3 a.
Because of symmetry, it is enough to prove that x belongs to circumcircle of g1 g2 c, i.e. to
prove the following:
x − c g1 − g2 g1 − c x − g2
· = · . (2)
x̄ − c̄ g¯1 − g¯2 g¯1 − c̄ x̄ − g¯2
Easy computations give that
a b
2ab − bc − ac 2− c
− c
g1 − g2 = (b − a) , g¯1 − g¯2 = (b̄ − ā) ,
3ab 3
and then
g1 − g2 c(bc + ac − 2ab)
= .
g¯1 − g¯2 2c − a − b
6
On the other hand we have
2ab − bc − ac 2c − a − b
g1 − c = , g¯1 − c̄ = .
3a 3bc
This implies
g1 − c 2ab − bc − ac bc
= · .
g¯1 − c̄ 2c − a − b a
Then (2) is equivalent to
The last equality can easily be verified, which implies that x belongs to circumcircle of
triangle g1 g2 c. This concludes our proof.
7
Solution 4 (rotation)
The first part of the first solution and the second part of the second solution can also be
done by the following rotation argument: The rotation through 2∠BAC about A takes
O3 to O2 , G3 to G2 and H3 to H2 (again, H2 , H3 are the reflections of the orthocenter
H in the sides CA, AB). Let X be the intersection of the Euler line reflections e2 (going
through O2 , G2 , H2 ) and e3 (going through O3 , G3 , H3 ). We now use the well-known fact
that if a rotation about a point A takes a line l and a point P on l to the line l0 and the
point P 0 , then the quadrilateral AP P 0 X is cyclic, where X is the intersection of l and l0 .
For this reason, AO3 O2 X, AG3 G2 X and AH3 H2 X are cyclic quadrilaterals. Since H2 and
H3 lie on the circumcircle of the triangle ABC, the circumcircles of the triangles AH3 H2
and ABC are the same, hence X lies on the circumcircle of ABC.
This proves the first part of the first solution as well as the second part of the second
solution.
8
Problem 1 Let ABC be a triangle with CA = CB and ∠ACB = 120◦, and let M be the
midpoint of AB. Let P be a variable point on the circumcircle of ABC, and let Q be the point
on the segment CP such that QP = 2QC. It is given that the line through P and perpendicular
to AB intersects the line MQ at a unique point N.
Prove that there exists a fixed circle such that N lies on this circle for all possible positions
of P .
(Velina Ivanova, Bulgaria)
Solution Let O be the circumcenter of ABC. From the assumption that ∠ACB = 120◦ it
follows that M is the midpoint of CO.
Let ω denote the circle with center in C and radius CO. This circle in the image of the
circumcircle of ABC through the translation that sends O to C. We claim that N lies on ω.
N b
C C
b b
P b Q
A b b b
B A b b b
B
b
M M
b
Q
N
b b
O O
Let us consider the triangles QNP and QMC. The angles in Q are equal. Since NP is parallel
to MC (both lines are perpendicular to AB), it turns out that ∠QNP = ∠QMC, and hence the
two triangles are similar. Since QP = 2QC, it follows that
NP = 2MC = CO,
Comment The possible positions of N are all the points of ω with the exception of the two
points lying on the line CO. Indeed, P does not lie on the line CO because otherwise the point
N is not well-defined, and therefore also N does not lie on the same line.
Conversely, let N be any point on ω and not lying on the line CO. Let P be the corresponding
point on the circumcircle of ABC, namely such that NP is parallel and equal to CO. Let Q be
1
the intersection of CP and NM. As before, the triangles QNP and QMC are similar, and now
from the relation NP = 2MC we deduce that QP = 2QC. This proves that N can be obtained
from P through the construction described in the statement of the problem.
Computational solution There are many computation approaches to this problem. For ex-
ample, we can set Cartesian coordinates so that
√ ! √ !
3 1 3 1 1
A= − , , B= , , C = (0, 1) , M = 0, .
2 2 2 2 2
Setting P = (a, b), we obtain that Q = (a/3, (2 + b)/3). The equation of the line through P
and perpendicular to AB is x = a. The equation of the line MQ (if a 6= 0) is
1 x 1
y− = +b .
2 a 2
This shows that the map P → N in the translation by the vector (0, 1). This result is
independent of the position of P (provided that a 6= 0, because otherwise N is not well-defined).
When P lies on the circumcircle of ABC, with the exception of the two points with a = 0,
then necessarily N lies on the translated circle (which is the circle with center in C and radius 1).
2
Problem 2 Consider the set
1
A = 1 + : k = 1, 2, 3, . . . .
k
(a) Prove that every integer x ≥ 2 can be written as the product of one or more elements of A,
which are not necessarily different.
(b) For every integer x ≥ 2, let f (x) denote the minimum integer such that x can be written as
the product of f (x) elements of A, which are not necessarily different.
Prove that there exist infinitely many pairs (x, y) of integers with x ≥ 2, y ≥ 2, and
which is enough to establish part (a). We now consider part (b). Notice that for any positive
integer k we have
f (2k + 1) ≤ k + 1,
because 2k + 1 = 1 + 21k · 2k is a representation of 2k + 1 as a product of k + 1 elements of A.
We claim that all the pairs (x, y) of the form
24k+2 + 1
x = 5, y=
5
satisfy the required inequality. Notice that y is an integer for any positive value of k, because
24k+2 + 1 ≡ 16k · 4 + 1 ≡ 5 ≡ 0 (mod 5). Furthermore, f (xy) = f (24k+2 + 1) ≤ 4k + 3 (and
f (x) = f (22 + 1) ≤ 3) by the above. We now need some lower bounds on the values of f . Notice
that no element of A exceeds 2, and therefore the product of at most k elements of A does not
exceed 2k : it follows that
f (n) ≥ ⌈log2 (n)⌉, (Q2.1)
and in particular that
f (5) = f (22 + 1) ≥ ⌈log2 (5)⌉ = 3.
We have thus proven f (x) = f (5) = 3. We want to show f (xy) < f (x) + f (y), and since we
know f (xy) ≤ 4k + 3 and f (x) = 3 we are reduced to showing f (y) > 4k. Since y > 24k−1 , from
(Q2.1) we already know that f (y) ≥ 4k, and hence we just need to exclude that f (y) = 4k. Let
us assume that we can represent y in the form a1 · . . . · a4k with every ai in A. At least one of the
ai is not 2 (otherwise the product would be a power of 2, while y is odd), and hence it is less than
or equal to 3/2. It follows that
3 24k−2 24k+2
a1 · . . . · a4k ≤ 24k−1 · = 15 · < < y,
2 5 5
3
which contradicts the fact that a1 · . . . · a4k is a representation of y.
Note. Using a similar approach one can also prove that all pairs of the form
22k+1 + 1 210k+5 + 1
3, and 11,
3 11
Second solution As in the previous solution we obtain the lower bound (Q2.1).
Now we claim that all the pairs of the form
x = 2k + 1, y = 4k − 2k + 1
satisfy the required inequality when k is large enough. To begin with, it is easy to see that
2k + 1 23k + 1
2k + 1 = · |2 · .{z
. . · 2} and 23k + 1 = . . · 2},
· 2| · .{z
2k 23k
k terms 3k terms
which shows that f (2k + 1) ≤ k + 1 and f (23k + 1) ≤ 3k + 1. On the other hand, from (Q2.1) we
deduce that the previous inequalities are actually equalities, and therefore
Therefore, it remain to show that f (y) > 2k. Since y > 22k−1 (for k ≥ 1), from (Q2.1) we
already know that f (y) ≥ 2k, and hence we just need to exclude that f (y) = 2k. Let us assume
that we can represent y in the form a1 · . . . · a2k . At least one of the factors is not 2, and hence it
is less than or equal to 3/2. Thus when k is large enough it follows that
3 3
a1 · . . . · a2k ≤ 22k−1 · = · 22k < 22k − 2k < y,
2 4
which contradicts the fact that a1 · . . . · a2k is a representation of y.
Third solution Let’s start by showing that (x, y) = (7, 7) satisfies f (xy) < f (x) + f (y). We
have f (7) ≥ 4 since 7 cannot be written as the product of 3 or fewer elements of A: indeed 23 > 7,
and any other product of at most three elements of A does not exceed 22 · 23 = 6 < 7. On the
other hand, f (49) ≤ 7 since 49 = 2 · 2 · 2 · 2 · 2 · 32 · 48
49
.
Suppose by contradiction that there exist only finitely many pairs (x, y) that satisfy f (xy) <
f (x) + f (y). This implies that there exists M large enough so that whenever a > M or b > M
holds we have f (ab) = f (a)+f (b) (indeed, it is clear that the reverse inequality f (ab) ≤ f (a)+f (b)
is always satisfied).
Now take any pair (x, y) that satisfies f (xy) < f (x) + f (y) and let n > M be any integer. We
obtain
f (n) + f (xy) = f (nxy) = f (nx) + f (y) = f (n) + f (x) + f (y),
which contradicts f (xy) < f (x) + f (y).
4
Problem 3 The n contestants of an EGMO are named C1 , . . . , Cn . After the competition they
queue in front of the restaurant according to the following rules.
• The Jury chooses the initial order of the contestants in the queue.
– If contestant Ci has at least i other contestants in front of her, she pays one euro to
the Jury and moves forward in the queue by exactly i positions.
– If contestant Ci has fewer than i other contestants in front of her, the restaurant opens
and the process ends.
(a) Prove that the process cannot continue indefinitely, regardless of the Jury’s choices.
(b) Determine for every n the maximum number of euros that the Jury can collect by cunningly
choosing the initial order and the sequence of moves.
(Hungary)
n+1 n+1 n 1
n 1 n−1 2
n−1 2 .. ..
M moves
n
−−− −−−→
n moves
−−−−→ . M moves
n
−−− −−−→ .
.. ..
. . 2 n−1
2 n−1 1 n
1 n n+1 n+1
Let us show now that at most 2n − n − 1 moves are possible. To this end, let us identify a line
of contestants with a permutation σ of {1, . . . , n}. To each permutation we associate the set of
reverse pairs
R(σ) := {(i, j) : 1 ≤ i < j ≤ n and σ(i) > σ(j)},
and the nonnegative integer X
W (σ) := 2i ,
(i,j)∈R(σ)
which we call the total weight of the permutation. We claim that the total weight decreases after
any move of the contestants. Indeed, let us assume that Ci moves forward in the queue, let σ
be the permutation before the move, and let σ ′ denote the permutation after the move. Since Ci
jumps over exactly i contestants, necessarily she jumps over at least one contestant Cj with index
5
j > i. This means that the pair (i, j) is reverse with respect to σ but not with respect to σ ′ , and
this yields a reduction of 2i in the total weight. On the other hand, the move by Ci can create
new reverse pairs of the form (k, i) with k < i, but their total contribution is at most
20 + 21 + . . . + 2i−1 = 2i − 1.
In conclusion, when passing from σ to σ ′ , at least one term 2i disappears from the computation
of the total weight, and the sum of all the new terms that might have been created is at most
2i − 1. This shows that W (σ ′ ) ≤ W (σ) − 1.
We conclude by observing that the maximum possible value of W (σ) is realized when all pairs
are reverse, in which case
Xn
W (σ) = (i − 1)2i = 2n − n − 1.
i=1
This proves that the number of moves is less than or equal to 2n − n − 1, and in particular it
is finite.
Alternative solution As in the previous solution, the fundamental observation is again that,
when a contestant Ci moves forward, necessarily she has to jump over at least one contestant Cj
with j > i.
Let us show now that the process ends after a finite number of moves. Let us assume that
this is not the case. Then at least one contestant moves infinitely many times. Let i0 be the
largest index such that Ci0 moves infinitely many times. Then necessarily Ci0 jumps infinitely
many times over some fixed Cj0 with j0 > i0 . On the other hand, we know that Cj0 makes only
a finite number of moves, and therefore she can precede Ci0 in the line only a finite number of
times, which is absurd.
In order to estimate from above the maximal number of moves, we show that the contestant
Ci can make at most 2n−i − 1 moves. Indeed, let us argue by “backward extended induction”. To
begin with, we observe that the estimate is trivially true for Cn because she has no legal move.
Let us assume now that the estimate has been proved for Ci , Ci+1 , . . . , Cn , and let us prove
it for Ci−1 . When Ci−1 moves, at least one contestant Cj with j > i − 1 must precede her in the
line. The initial configuration can provide at most n − i contestants with larger index in front
of Ci−1 , which means at most n − i moves for Ci−1 . All other moves are possible only if some
contestant in the range Ci , Ci+1 , . . . , Cn jumps over Ci−1 during her moves. As a consequence,
the total number of moves of Ci−1 is at most
n
X
n−i+ (2n−k − 1) = 2n−i+1 − 1.
k=i
Comment In every move of the example, the moving contestant jumps over exactly one con-
testant with larger index (and as a consequence over all contestants with smaller index).
6
Problem 4 A domino is a 1 × 2 or 2 × 1 tile.
Let n ≥ 3 be an integer. Dominoes are placed on an n × n board in such a way that each
domino covers exactly two cells of the board, and dominoes do not overlap.
The value of a row or column is the number of dominoes that cover at least one cell of this
row or column. The configuration is called balanced if there exists some k ≥ 1 such that each row
and each column has a value of k.
Prove that a balanced configuration exists for every n ≥ 3, and find the minimum number of
dominoes needed in such a configuration.
(Merlijn Staps, The Netherlands)
2nk = 3D.
Any n ≥ 8 can be written in the form 4A + r where A is a positive integer and r ∈ {4, 5, 6, 7}.
Therefore, we can obtain a balanced configuration with n ≥ 8 and k = 3 by using one block with
size r × r, and A blocks with size 4 × 4 along the principal diagonal of the board. In particular,
this construction covers all the cases where n is not a multiple of 3.
7
Problem 5 Let Γ be the circumcircle of triangle ABC. A circle Ω is tangent to the line segment
AB and is tangent to Γ at a point lying on the same side of the line AB as C. The angle bisector
of ∠BCA intersects Ω at two different points P and Q.
Prove that ∠ABP = ∠QBC.
(Dominika Regiec, Poland)
Solution 1 Let M be the midpoint of the arc AB that does not contain C, let V be the
intersection of Ω and Γ, and let U be the intersection of Ω and AB.
C
b
Ω
b
V
P b
U
b b b
A B
8
Solution 2 The second solution is in fact a different proof of the first part of Solution 1.
Let us consider the inversion with respect to circle with center M and radius MA = MB.
This inversion switches AB and Γ, and fixes the line passing through M, U, V . As a consequence,
it keeps Ω fixed, and therefore it switches P and Q. This is because they are the intersections
between the fixed line MC and Ω, and the only fixed point on the segment MC is its intersection
with the inversion circle (thus P and Q are switched). This implies that MP · MQ = MB 2 .
Solution 3 This solution is instead a different proof of the second step of Solution 1.
Let I and J be the incenter and the C-excenter of △ABC respectively. It is well-known that
MA = MI = MJ, therefore the relation MP · MQ = MA2 implies that (P, Q, I, J) = −1.
Now observe that ∠IBJ = 90◦ , thus BI is the angle bisector of ∠P BQ as it is well-known
from the theory of harmonic pencils, and this leads easily to the conclusion.
Solution 4 Let D denote the intersection of AB and CM. Let us consider an inversion with
respect to B, and let us use primes to denote corresponding points in the transformed diagram,
with the gentlemen agreement that B ′ = B.
′
Q′ C b
b
ω′
′
Ω
P′ A′
D′
b b b
B′
M′
9
Problem 6
(a) Prove that for every real number t such that 0 < t < 21 there exists a positive integer n
with the following property: for every set S of n positive integers there exist two different
elements x and y of S, and a non-negative integer m (i.e. m ≥ 0), such that
|x − my| ≤ ty.
1
(b) Determine whether for every real number t such that 0 < t < 2
there exists an infinite set
S of positive integers such that
|x − my| > ty
for every pair of different elements x and y of S and every positive integer m (i.e. m > 0).
Solution
• If si+1 > (1 + t)si for every i ∈ {1, . . . , n − 1}, then by induction we obtain that
sn > (1 + t)n−1 s1 .
10
Part (b) (Explicit formula) We claim that an infinite set with the required property exists.
To this end, we rewrite the required condition in the form
x
− m > t.
y
This is equivalent to saying that the distance between the ratio x/y and the set of positive
integers is greater than t.
Now we construct an increasing sequence sn of odd coprime positive integers satisfying
1 1
− >t ∀n ≥ 1, (Q6.2)
2 2sn
and such that for every j > i it turns out that
si 1 sj 1
< and t< < , (Q6.3)
sj 2 si 2
where {α} denotes the fractional part of α. This is enough to show that the set S := {sn : n ≥ 1}
has the required property.
To this end, we consider the sequence defined recursively by
(s1 · . . . · sn )2 + 1
sn+1 = ,
2
with s1 large enough. An easy induction shows that this is an increasing sequence of odd positive
integers. For every i ∈ {1, . . . , n} it turns out that
si 2 2 1
≤ ≤ <
sn+1 si s1 2
because s1 is large enough, which proves the first relation in (Q6.3). Moreover, it turns out that
sn+1 (s1 · . . . · sn )2 1
= + .
si 2si 2si
The first term is a positive integer plus 1/2, from which it follows that the distance of sn+1 /si
from the positive integers is greater than or equal to
1 1 1 1
− ≥ − ,
2 2si 2 2s1
which is greater than t if s1 is large enough. This proves the second relation in (Q6.3).
Part (b) (Arithmetic approach) We produce an increasing sequence sn of odd and coprime
positive integers that satisfies (Q6.3) every j > i. As in the previous solution, this is enough to
conclude.
We argue by induction. To begin with, we choose s1 to be any odd integer satisfying the
inequality in (Q6.2). Let us assume now that s1 , . . . , sn have already been chosen, and let us
choose sn+1 in such a way that
si − 1
sn+1 ≡ (mod si ) ∀i ∈ {1, . . . , n}.
2
11
We can solve this system because the previously chosen integers are odd and coprime. More-
over, any solution of this system is coprime with s1 , . . . , sn . Indeed, for every 1 ≤ i ≤ n it turns
out that
si − 1
sn+1 = + k i si
2
for some positive integer ki . Therefore, any prime p that divides both sn+1 and si divides also
(2ki + 1)si − 2sn+1 = 1, which is absurd. Finally, we observe that we can assume that sn+1 is odd
and large enough. In this way we can guarantee that
si 1
< ∀i ∈ {1, . . . , n},
sn+1 2
which is the first requirement in (Q6.3), and
1 1 sn+1 1
ki + t < ki + − = < ki + ∀i ∈ {1, . . . , n},
2 2si si 2
which implies the second requirement in (Q6.3).
Part (b) (Algebraic approach) Again we produce an increasing sequence sn of positive integers
that satisfies (Q6.3) every j > i.
To this end, for every positive integer x, we define its security region
[
S(x) := (n + t)x, (n + 21 )x .
n≥1
The security region S(x) is a periodic countable union of intervals of length ( 21 − t)x, whose
left-hand or right-hand endpoints form an arithmetic sequence. It has the property that
ny o 1
t< < ∀y ∈ S(x).
x 2
Now we prove by induction that we can choose a sequence sn of positive integers satisfying
(Q6.3) and in addition the fact that every interval of the security region S(sn ) contains at least
one interval of S(sn−1).
To begin with, we choose s1 large enough so that the length of the intervals of S(s1 ) is larger
than 1. This guarantees that any interval of S(s1 ) contains at least a positive integer. Now let
us choose a positive integer s2 ∈ S(s1 ) that is large enough. This guarantees that s1 /s2 is small
enough, that the fractional part of s2 /s1 is in (t, 1/2), and that every interval of the security region
S(s2 ) contains at least one interval of S(s1 ), and hence at least one positive integer.
Let us now assume that s1 , . . . , sn have been already chosen with the required properties. We
know that every interval of S(sn ) contains at least one interval of S(sn−1 ), which in turn contains
an interval in S(sn−2 ), and so on up to S(s1 ). As a consequence, we can choose a large enough
positive integer sn+1 that lies in S(sk ) for every k ∈ {1, . . . , n}. Since sn+1 is large enough, we are
sure that
sk
<t ∀k ∈ {1, . . . , n}.
sn+1
Moreover, we are sure also that all the intervals of S(sn+1 ) are large enough, and therefore
they contain at least one interval of S(sn ), which in turn contains at least one interval of S(sn−1 ),
and so on. Finally, the condition
sn−1 1
t< <
sn 2
is guaranteed by the fact that sn+1 was chosen in an interval that is contained in S(sk ) for every
k ∈ {1, . . . , n}. This completes the induction.
12
Day 1. Solutions
Problem 1 (Netherlands). Find all triples (a, b, c) of real numbers such that ab + bc +
ca = 1 and
a2 b + c = b2 c + a = c2 a + b.
c2 + a2 + ca = 1,
ab + bc + ca = 1.
Adding first three equations and subtracting the fourth yields 2(a2 + b2 + c2 ) = 2 =
2(ab + bc + ca). Consequently, (a − b)2 + (b − c)2 + (c − a)2 = 0. Now we can easily
conclude that a = b = c = ± √13 .
Solution by Achilleas Sinefakopoulos, Greece. We have
c(1 − b2 ) = a(1 − ab) = a(bc + ca) = c(ab + a2 ),
and so
c(a2 + ab + b2 − 1) = 0.
Similarly, we have
b(a2 + ac + c2 − 1) = 0 and a(b2 + bc + c2 − 1) = 0.
1
If c = 0, then we get ab = 1 and a2 b = a = b, which give us a = b = 1, or a = b = −1.
Similarly, if a = 0, then b = c = 1, or b = c = −1, while if b = 0, then a = c = 1, or
a = c = −1.
So assume that abc 6= 0. Then
a2 + ab + b2 = b2 + bc + c2 = c2 + ca + a2 = 1.
a2 + b2 + c2 = 1 = ab + bc + ca.
Hence
a2 b + c = b2 c + a = c2 a + b. (1)
We have
a2 b + c = b2 c + a ⇐⇒ a2 b − a = b 2 c − c
⇐⇒ a(ab − 1) = c(b2 − 1)
⇐⇒ a(−bc − ac) = c(b2 − 1)
⇐⇒ −ac(a + b) = c(b2 − 1) (2)
First, consider the case where one of a, b, c is equal to 0. Without loss of generality, assume
that a = 0. Then bc = 1 and b = c from (1), and so b2 = 1 giving us b = 1 or −1. Hence
b = c = 1 or b = c = −1.
Therefore, (a, b, c) equals one of the triples (0, 1, 1), (0, −1, −1), as well as their rearrange-
ments (1, 0, 1) and (−1, 0, −1) when b = 0, or (1, 1, 0) and (−1, −1, 0) when c = 0.
Now consider the case where a 6= 0, b 6= 0 and c 6= 0. Then (2) gives us
−a(a + b) = b2 − 1 ⇐⇒ −a2 − ab = b2 − 1 ⇐⇒ a2 + ab + b2 − 1 = 0.
2
Using a2 b + c = c2 a + b we obtain b(a2 − 1) = c(ac − 1) yielding a2 + ac + c2 − 1 = 0 in
a similar way. The quadratic Q(x) = x2 + cx + c2 − 1 has x = a as a root. Let x2 be its
second root (which could be equal to a in the case where the discriminant is 0). From
Vieta’s formulas we get
(
x2 + a = −c ⇐⇒ x2 = −c − a, and
2
x2 a = c2 − 1 ⇐⇒ x2 = c a−1 .
Then (
x1 + x2 = −b − a − c − a, and
2 2
x1 + x2 = b a−1 + c a−1 ,
which give us
b2 − 1 c 2 − 1
−(2a + b + c) = + ⇐⇒ −2a2 − ba − ca = b2 + c2 − 2
a a
⇐⇒ bc − 1 − 2a2 = b2 + c2 − 2
⇐⇒ 2a2 + b2 + c2 = 1 + bc. (3)
By symmetry, we get
2b2 + a2 + c2 = 1 + ac, and (4)
2c2 + a2 + b2 = 1 + bc (5)
Adding equations (3), (4), and (5), we get
4(a2 + b2 + c2 ) = 3 + ab + bc + ca ⇐⇒ 4(a2 + b2 + c2 ) = 4 ⇐⇒ a2 + b2 + c2 = 1.
3
Thus, after substracting the cyclicly symmetric part abc + cyc a2 b we find the condition
P
is eqivalent to
D := bc(c − b) = ca(a − c) = ab(b − a).
Ending 1. It is easy to see that if e.g. a = 0 then b = c = ±1, and if e.g. a = b then either
a = b = c = ± √13 or a = b = ±1, c = 0, and these are indeed solutions. So, to show that
these are all solutions (up to symmetries), we may assume by contradiction that a, b, c
are pairwise different and non-zero. All conditions are preserved under cyclic shifts and
under simultaenously switching signs on all a, b, c, and by applying these operations as
necessary we may assume a < b < c. It follows that D3 = a2 b2 c2 (c − b)(a − c)(b − a) must
be negative (the only negative term is a − c, hence D is negative, i.e. bc, ab < 0 < ac. But
this means that a, c have the same sign and b has a different one, which clearly contradicts
a < b < c! So, such configurations are impossible.
P 2
c b − b2 c = (c − b)(c − a)(b − a) and D3 = a2 b2 c2 (c −
P
Ending 2. Note that 3D =
b)(a − c)(b − a) = −3a2 b2 c2 D. Since 3D and D3 must have the same sign, and −3a2 b2 c2
is non-positive, necessarily D = 0. Thus (up to cyclic permutation) a = b and from there
we immediately find either a = b = ±1, c = 0 or a = b = c = ± √13 .
4
Problem 2 (Luxembourg). Let n be a positive integer. Dominoes are placed on a
2n × 2n board in such a way that every cell of the board is adjacent to exactly one cell
covered by a domino. For each n, determine the largest number of dominoes that can be
placed in this way.
(A domino is a tile of size 2 × 1 or 1 × 2. Dominoes are placed on the board in such a
way that each domino covers exactly two cells of the board, and dominoes do not overlap.
Two cells are said to be adjacent if they are different and share a common side.)
Solution 1. Let M denote the maximal number of dominoes that can be placed on the
chessboard. We claim that M = n(n + 1)/2. The proof naturally splits into two parts:
we first prove that n(n + 1)/2 dominoes can be placed on the board, and then show that
M ≤ n(n + 1)/2 to complete the proof.
We construct placings of the dominoes by induction. The base cases n = 1 and n = 2
correspond to the placings
and
or
depending on whether n is odd or even. In these constructions, the interior square is filled
with the placing for the 2n × 2n board. This construction adds 2n + 3 dominoes, and
therefore places, in total,
n(n + 1) (n + 2)(n + 3)
+ (2n + 3) =
2 2
dominoes on the board. Noticing that the contour and the interior mesh together appro-
priately, this proves, by induction, that n(n + 1)/2 dominoes can be placed on the 2nn
board.
To prove that M ≤ n(n+1)/2, we border the 2n×2n square board up to a (2n+2)×(2n+2)
square board; this adds 8n + 4 cells to the 4n2 cells that we have started with. Calling
a cell covered if it belongs to a domino or is adjacent to a domino, each domino on the
2n × 2n board is seen to cover exactly 8 cells of the (2n + 2) × (2n + 2) board (some of
which may belong to the border). By construction, each of the 4n2 cells of the 2n × 2n
board is covered by precisely one domino.
5
If two adjacent cells on the border, away from a corner, are covered, then there will be at
least two uncovered cells on both sides of them; if one covered cell lies between uncovered
cells, then again, on both sides of it there will be at least two uncovered cells; three or
more adjacent cells cannot be all covered. The following diagrams, in which the borders
are shaded, × marks an uncovered cell on the border, + marks a covered cell not belonging
to a domino, and − marks a cell which cannot belong to a domino, summarize the two
possible situations,
Close to a corner of the board, either the corner belongs to some domino,
× + + × × ···
+ + −
× + + −
× − −
..
.
or one of the following situations, in which the corner cell of the original board is not
covered by a domino, may occur:
× × + + × × ··· × × × × + + ···
× + + − × + + +
× + + + − + + + +
+ + − or × + +
+ + − × − −
.. ..
. .
It is thus seen that at most half of the cells on the border, i.e. 4n+2 cells, may be covered,
and hence 2
4n + (4n + 2) n(n + 1) 1 n(n + 1)
M≤ = + = ,
8 2 2 2
which completes the proof of our claim.
Solution 2. We use the same example as in Solution 1. Let M denote the maximum
number of dominoes which satisfy the condition of the problem. To prove that M ≤
n(n + 1)/2, we again border the 2n × 2n square board up to a (2n + 2) × (2n + 2) square
board. In fact, we shall ignore the corner border cells as they cannot be covered anyway
and consider only the 2n border cells along each side. We prove that out of each four
border cells next to each other at most two can be covered. Suppose three out of four
cells A, B, C, D are covered. Then there are two possibilities below:
+ + × + + + + ×
or
+ +
6
The first option is that A, B and D are covered (marked with + in top row). Then the
cells inside the starting square next to A, B and D are covered by the dominoes, but
the cell in between them has now two adjacent cells with dominoes, contradiction. The
second option is that A, B and C are covered. Then the cells inside the given square next
to A, B and C are covered by the dominoes. But then the cell next to B has two adjacent
cells with dominoes, contradiction.
Now we can split the border cells along one side in groups of 4 (leaving one group of 2
if n is odd). So when n is even, at most n of the 2n border cells along one side can be
covered, and when n is odd, at most n + 1 out of the 2n border cells can be covered. For
all four borders together, this gives a contribution of 4n when n is even and 4n + 4 when
n is odd. Adding 4n2 and dividing by 8 we get the desired result.
Solution (upper bound) by ISR5. Consider the number of pairs of adjacent cells,
such that one of them is covered by a domino. Since each cell is adjacent to one covered
cell, the number of such pairs is exactly 4n2 . On the other hand, let n2 be the number
of covered corner cells, n3 the number of covered edge cells (cells with 3 neighbours), and
n4 be the number of covered interior cells (cells with 4 neighbours). Thus the number of
pairs is 2n2 + 3n3 + 4n4 = 4n2 , whereas the number of dominoes is m = n2 +n23 +n4 .
Considering only the outer frame (of corner and edge cells), observe that every covered
cell dominates two others, so at most half of the cells are ccovered. The frame has a total
of 4(2n − 1) cells, i.e. n2 + n3 ≤ 4n − 2. Additionally n2 ≤ 4 since there are only 4 corners,
thus
8m = 4n2 +4n3 +4n4 = (2n2 +3n3 +4n4 )+(n2 +n3 )+n2 ≤ 4n2 +(4n−2)+4 = 4n(n+1)+2
n(n+1) n(n+1)
Thus m ≤ 2
+ 14 , so in fact m ≤ 2
.
Solution (upper and lower bound) by ISR5. We prove that this is the upper bound
(and also the lower bound!) by proving that any two configurations, say A and B, must
contain exactly the same number of dominoes.
Colour the board in a black and white checkboard colouring. Let W be the set of white
cells covered by dominoes of tiling A. For each cell w ∈ W let Nw be the set of its adjacent
(necessarily black) cells. Since each black cell has exactly one neighbour (necessarily
white) covered by a domino of tiling A, it follows that each black cell is contained in
exactly one Nw , i.e. the Nw form a partition of the black cells. Since each white cell has
exactly one (necessarily black) neighbour covered by a tile of B, each Bw contains exactly
one black tile covered by a domino of B. But, since each domino covers exactly one white
and one black cell, we have
as claimed.
7
Problem 3 (Poland). Let ABC be a triangle such that ∠CAB > ∠ABC, and let I
be its incentre. Let D be the point on segment BC such that ∠CAD = ∠ABC. Let ω
be the circle tangent to AC at A and passing through I. Let X be the second point of
intersection of ω and the circumcircle of ABC. Prove that the angle bisectors of ∠DAB
and ∠CXB intersect at a point on line BC.
Solution 1. Let S be the intersection point of BC and the angle bisector of ∠BAD, and
let T be the intersection point of BC and the angle bisector of ∠BXC. We will prove
that both quadruples A, I, B, S and A, I, B, T are concyclic, which yields S = T .
Firstly denote by M the middle of arc AB of the circumcenter of ABC which does not
contain C. Consider the circle centered at M passing through A, I and B (it is well-known
that M A = M I = M B); let it intersect BC at B and S 0 . Since ∠BAC > ∠CBA it is
easy to check that S 0 lies on side BC. Denoting the angles in ABC by α, β, γ we get
Secondly let N be the middle of arc BC of the circumcenter of ABC which does not
contain A. From ∠BAC > ∠CBA we conclude that X lies on the arc AB of circumcircle
of ABC not containing C. Obviously both AI and XT are passing through N . Since
∠N BT = α2 = ∠BXN we obtain 4N BT ∼ 4N XB, therefore
N T · N X = N B2 = N I 2.
8
Since ∠DAC = ∠CBA, triangles ADC and BAC are similar and therefore
DC AC
= .
AC BC
By the Law of sines
BW2 BA BC sin α
= = = .
W2 D AD AC sin β
Consequently
BD W2 D sin β
= +1= + 1,
BW2 BW2 sin α
BC BC BD 1 BD 1 BD
= · = · = 2 2
· =
BW2 BD BW2 1 − DC/BC BW2 1 − AC /BC BW2
sin2 α sin β + sin α sin α
2 2 · = .
sin α − sin β sin α sin α − sin β
Note that AXBC is cyclic and so ∠BXC = ∠BAC = α. Hence, ∠XBC = 180◦ −
∠BXC − ∠BCX = 180◦ − α − φ. By the Law of sines for the triangle BXC, we have
BC W1 C CX sin ∠CBX
= +1= +1= +1=
W1 B W1 B BX sin φ
sin (α + φ)
+ 1 = sin α cot φ + cos α + 1.
sin φ
Since AC is tangent to the circle AIX, we have ∠AXI = ∠IAC = α/2. Moreover
∠XAI = ∠XAB +∠BAI = φ+α/2 and ∠XIA = 180◦ −∠XAI −∠AXI = 180◦ −α −φ.
Applying the Law of sines again XAC, XAI, IAC we obtain
AX AI
= ,
sin (α + φ) sin α/2
AX AC AC
= = ,
sin (γ − φ) sin ∠AXC sin β
AI AC
= .
sin γ/2 sin (α/2 + γ/2)
9
sin α sin γ cot φ − sin α cos γ 2 sin β/2 cos α/2
=
sin γ sin α cot φ + sin γ cos α cos γ/2
2 sin β/2 cos α/2 − sin (α/2 + β/2) sin β/2 cos α/2 − sin α/2 cos β/2
= ,
cos γ/2 cos γ/2
− sin β
= 2 sin γ/2 sin (β/2 − α/2) =
sin α cot φ + cos α
and the result follows. We are left to note that none of the denominators can vanish.
Solution by Achilleas Sinefakopoulos, Greece. We first note that
Let CX and AD meet at K. Then ∠CXA = ∠ABC = ∠KAC. Also, we have ∠IXA =
∠A/2, since ω is tangent to AC at A. Therefore,
(the absolute value depends on whether ∠B ≥ ∠A/2 or not) which means that XKIA is
cyclic, i.e. K lies also on ω.
Let IK meet BC at E. (If ∠B = ∠A/2, then IK degenerates to the tangent line to ω at
I.) Note that BEIA is cyclic, because
We have ∠EKA = 180◦ − ∠AXI = 180◦ − ∠A/2 and ∠AEI = ∠ABI = ∠B/2. Hence
This means that AE is the angle bisector of ∠BAD. Next, let M be the point of inter-
section of AE and BI. Then
10
and so, its supplement is
∠AM I = ∠A/2 = ∠AXI,
so X, M, K, I, A all lie on ω. Next, we have
∠XM A = ∠XKA
= 180◦ − ∠ADC − ∠XCB
= 180◦ − ∠A − ∠XCB
= ∠B + ∠XCA
= ∠B + ∠XBA
= ∠XBE,
∠EXC = ∠EXM + ∠M XC
= ∠M BE + ∠M AK
= ∠B/2 + ∠BAD/2
= ∠A/2
= ∠BXC/2.
This means that XE is the angle bisector of ∠BXC and so we are done!
11
Solution based on that by Eirini Miliori (HEL2), edited by A. Sinefakopoulos,
Greece. It is ∠ABD = ∠DAC, and so AC is tangent to the circumcircle of 4BAD at
A. Hence CA2 = CD · CB.
12
Triangle 4ABC is similar to triangle 4CAD, because ∠C is a common angle and
∠CAD = ∠ABC, and so ∠ADC = ∠BAC = 2ϕ.
Let Q be the point of intersection of AD and CX. Since ∠BXC = ∠BAC = 2ϕ, it
follows that BDQX is cyclic.Therefore, CD · CB = CQ · CX = CA2 which implies that
Q lies on ω.
Next let P be the point of intersection of AD with the circumcircle of triangle 4ABC.
Then ∠P BC = ∠P AC = ∠ABC = ∠AP C yielding CA = CP. So, let T be on the side
BC such that CT = CA = CP . Then
∠C ∠A − ∠B ∠BAD
∠T AD = ∠T AC − ∠DAC = 90 − ◦
− ∠B = = ,
2 2 2
that is, line AT is the angle bisector of ∠BAD. We want to show that XT is the angle
bisector of ∠BXC. To this end, it suffices to show that ∠T XC = ϕ.
It is CT 2 = CA2 = CQ · CX, and so CT is tangent to the circumcircle of 4XT Q at
T. Since ∠T XQ = ∠QT C and ∠QDC = 2ϕ, it suffices to show that ∠T QD = ϕ, or, in
other words, that I, Q, and T are collinear.
Let T 0 is the point of intersection of IQ and BC. Then 4AIC is congruent to 4T 0 IC,
since they share CI as a common side, ∠ACI = ∠T 0 CI, and
13
Let the line AD meet ω at K. Then it is ∠AXK = ∠CAD = ∠ABC, because the angle
that is formed by a chord and a tangent to the circle at an endpoint of the chord equals
the inscribed angle to that chord. Therefore, ∠AXK = ∠AXC = ∠AXG. This means
that the point G coincides with the point K and so G belongs to the circle ω.
Let E be the point of intersection of the angle bisector of ∠DAB with BC. It suffices to
show that
CE XC
= .
BE XB
Let F be the second point of intersection of ω with AB. Then we have ∠IAF = ∠CAB 2
=
∠IXF, where I is the incenter of 4ABC, because ∠IAF and ∠IXF are inscribed in the
same arc of ω. Thus 4AIF is isosceles with AI = IF. Since I is the incenter of 4ABC,
we have AF = 2(s − a), where s = (a + b + c)/2 is the semiperimeter of 4ABC. Also, it
is CE = AC = b because in triangle 4ACE, we have
BF = BA − AF = c − 2(s − a) = a − b = CB − CE = BE.
Therefore
CE AC XC
= = ,
BE BF XB
as desired. The proof is complete.
Solution by IRL1 and IRL 5. Let ω denote the circle through A and I tangent to
AC. Let Y be the second point of intersection of the circle ω with the line AD. Let L
14
be the intersection of BC with the angle bisector of ∠BAD. We will prove ∠LXC =
1/2∠BAC = 1/2∠BXC.
We will refer to the angles of 4ABC as ∠A, ∠B, ∠C. Thus ∠BAD = ∠A − ∠B.
On the circumcircle of 4ABC, we have ∠AXC = ∠ABC = ∠CAD, and since AC is
tangent to ω, we have ∠CAD = ∠CAY = ∠AXY . Hence C, X, Y are collinear.
Also note that 4CAL is isosceles with ∠CAL = ∠CLA = 21 (∠BAD) + ∠ABC = 12 (∠A +
∠B) hence AC = CL. Moreover, CI is angle bisector to ∠ACL so it’s the symmetry axis
for the triangle, hence ∠ILC = ∠IAC = 1/2∠A and ∠ALI = ∠LIA = 1/2∠B. Since
AC is tangent to ω, we have ∠AY I = ∠IAC = 1/2∠A = ∠LAY + ∠ALI. Hence L, Y, I
are collinear.
Since AC is tangent to ω, we have 4CAY ∼ 4CXA hence CA2 = CX · CY . However
we proved CA = CL hence CL2 = CX · CY . Hence 4CLY ∼ 4CXL and hence
∠CXL = ∠CLY = ∠CAI = 1/2∠A.
X I
B L D C
Solution by IRL 5. Let M be the midpoint of the arc BC. Let ω denote the circle
through A and I tangent to AC. Let N be the second point of intersection of ω with AB
and L the intersection of BC with the angle bisector of ∠BAD. We know DL LB
= AD
AB
and
want to prove XB
XC
= LB
LC
.
First note that 4CAL is isosceles with ∠CAL = ∠CLA = 21 (∠BAD) + ∠ABC hence
AC = CL and LB LC
LB
= AC .
XB
Now we calculate XC
:
Comparing angles on the circles ω and the circumcircle of 4ABC we get 4XIN ∼
4XM B and hence also 4XIM ∼ 4XN B (having equal angles at X and proportional
XB
adjoint sides). Hence XM =N
IM
B
.
Also comparing angles on the circles ω and the circumcircle of 4ABC and using the
XC
tangent AC we get 4XAI ∼ 4XCM and hence also 4XAC ∼ 4XIM. Hence XM =
AC
IM
.
XB NB LB LB
Comparing the last two equations we get XC
= AC
. Comparing with LC
= AC
, it remains
to prove N B = LB.
15
A
N I
B D
L C
16
Let B 0 , D0 be the intersections of AB, AD with the circle AXI, respectively. We wish to
show that this D0 is the concurrency point defined above, i.e. that CD0 X and ID0 E are
collinear. Additionally, we will show that I is the circumcenter of AB 0 E.
Consider the inversion with center C and radius CA. The circles AXI and ABD are
tangent to CA at A (the former by definition, the latter since ∠CAD = ∠ABC), so
they are preserved under the inversion. In particular, the inversion transposes D and B
and preserves A, so sends the circle CAB to the line AD. Thus X, which is the second
intersection of circles ABC and AXI, is sent by the inversion to the second intersection
of AD and circle AXI, which is D0 . In particular CD0 X are collinear.
In the circle AIB 0 , AI is the angle bisector of B 0 A and the tangent at A, so I is the
midpoint of the arc AB 0 , and in particular AI = IB 0 . By angle chasing, we find that
ACE is an isosceles triangle:
thus the angle bisector CI is the perpendicular bisector of AE and AI = IE. Thus I is
the circumcenter of AB 0 E.
We can now show that ID0 E are collinear by angle chasing:
Solution inspired by ISR2. Let W be the midpoint of arc BC, let D0 be the second
intersection point of AD and the circle ABC. Let P be the intersection of the angle
bisector XW of ∠CXB with BC; we wish to prove that AP is the angle bisector of
DAB. Denote α = ∠CAB 2
, β = ∠ABC.
Let M be the intersection of AD and XC. Angle chasing finds:
and ∠P XB = α = ∠CAI = ∠AXI, and it follows that 4XIA ∼ 4XBP . Let S be the
second intersection point of the cirumcircles of XIA and XBP . Then by the spiral map
lemma (or by the equivalent angle chasing) it follows that ISB and ASP are collinear.
17
Let L be the second intersection of ω and AB. We want to prove that ASP is the angle
bisector of ∠DAB = ∠M AL, i.e. that S is the midpoint of the arc M L of ω. And this
follows easily from chasing angular arc lengths in ω:
˜ = ∠CAI = α
AI
ˆ = ∠IAL = α
IL
M
¯ I = ∠M XI = α − β
˜ = ∠ABI =
˜ − SL
AI β
2
And thus
M
¯ L=M
¯ I + IL ˜ − β ) = 2SL.
ˆ = 2α − β = 2(AI ˜
2
A, X ∗ , E ∗ , N ∗ are concyclic ⇐⇒ Y X ∗ · Y N ∗ = Y A · Y E ∗
⇐⇒ Y X ∗ · Y N ∗ = Y B ∗ · Y C ∗ .
(A, B ∗ , C ∗ , E ∗ are concyclic)
18
Define Q, R as described in the figure, and we get by simple angle chasing
19
Day 2. Solutions
Problem 4 (Poland). Let ABC be a triangle with incentre I. The circle through B
tangent to AI at I meets side AB again at P . The circle through C tangent to AI at I
meets side AC again at Q. Prove that P Q is tangent to the incircle of ABC.
Solution 1. Let QX, P Y be tangent to the incircle of ABC, where X, Y lie on the
incircle and do not lie on AC, AB. Denote ∠BAC = α, ∠CBA = β, ∠ACB = γ.
Since AI is tangent to the circumcircle of CQI we get ∠QIA = ∠QCI = γ2 . Thus
α γ
∠IQC = ∠IAQ + ∠QIA = + .
2 2
Similarly one can prove that ∠AP Y = γ. This means that Q, P, X, Y are collinear which
leads us to the conclusion that X = Y and QP is tangent to the incircle at X.
thus J, P , I, Q are concyclic. Let S be the intersection of AI and BC. It follows that
This means that IQ is the angle bisector of ∠CQP , so QP is indeed tangent to the incircle
of ABC.
Comment. The final angle chasing from the Solution 2 may simply be replaced by the
observation that since J, P , I, Q are concyclic, then I is the A-excenter of triangle AP Q.
Solution 3. Like before, notice that AQ · AC = AP · AB = AI 2 . Consider the positive
inversion Ψ with center A and power AI 2 . This maps P to B (and vice-versa), Q to C
1
(and vice-versa), and keeps the incenter I fixed. The problem statement will follow from
the fact that the image of the incircle of triangle ABC under Ψ is the so-called mixtilinear
incircle of ABC, which is defined to be the circle tangent to the lines AB, AC, and the
circumcircle of ABC. Indeed, since the image of the line QP is the circumcircle of ABC,
and inversion preserves tangencies, this implies that QP is tangent to the incircle of ABC.
We justify the claim as follows: let γ be the incircle of ABC and let ΓA be the A-mixtilinear
incircle of ABC. Let K and L be the tangency points of γ with the sides AB and AC,
and let U and V be the tangency points of ΓA with the sides AB and AC, respectively.
It is well-known that the incenter I is the midpoint of segment U V . In particular, since
AI A
also AI ⊥ U V , this implies that AU = AV = cos A . Note that AK = AL = AI · cos 2 .
2
Therefore, AU · AK = AV · AL = AI 2 , which means that U and V are the images of K
and L under Ψ. Since ΓA is the unique circle simultaneously tangent to AB at U and to
AC at V , it follows that the image of γ under Ψ must be precisely ΓA , as claimed.
Solution by Achilleas Sinefakopoulos, Greece. From the power of a point theorem,
we have
AP · AB = AI 2 = AQ · AC.
Hence P BCQ is cyclic, and so, ∠AP Q = ∠BCA. Let K be the circumcenter of 4BIP
and let L be the circumcenter of 4QIC. Then KL is perpendicular to AI at I.
Let N be the point of intersection of line KL with AB.Then in the right triangle 4N IA,
we have ∠AN I = 90◦ − ∠BAC 2
and from the external angle theorem for triangle 4BN I,
we have ∠AN I = ∠ABC2
+ ∠N IB. Hence
∠BCA ∠BCA
∠QP I = 180 − ∠AP Q − ∠BP I = 180 − ∠BCA − 90 −
◦ ◦ ◦
= 90◦ − ,
2 2
as well. Hence I lies on the angle bisector of ∠BP Q, and so it is equidistant from its
sides P Q and P B. Therefore, the distance of I from P Q equals the inradius of 4ABC,
as desired.
2
Solution by Eirini Miliori (HEL2). Let D be the point of intersection of AI and BC
and let R be the point of intersection of AI and P Q. We have ∠RIP = ∠P BI = ∠B 2
,
∠C
∠RIQ = ∠ICQ = 2 , ∠IQC = ∠DIC = x and ∠BP I = ∠BID = ϕ, since AI is
tangent to both circles.
RQ sin ∠B
2 sin ω
· = . (2)
RP sin ∠C
2
sin y
DC ID
Similarly, from the sine law in triangles 4IDC and 4IDB, it is = and
sin x sin ∠C
2
BD ID
= , and so
sin ϕ sin ∠B
2
DC sin ϕ sin ∠B
2
· = . (3)
BD sin x sin ∠C
2
RQ DC sin ϕ sin ω
By multiplying equations (2) with (3), we obtain · · = , which combined
RP BD sin x sin y
with (1) and cross-multiplying yields
or
1 1
(cos(ϕ − θ + ω) − cos(ϕ + θ − ω)) = (cos(ω − θ + ϕ) − cos(ω + θ − ϕ)) ,
2 2
which is equivalent to
cos(ϕ + θ − ω) = cos(ω + θ − ϕ).
So
ϕ + θ − ω = 2k · 180◦ ± (ω + θ − ϕ), (k ∈ Z.)
If ϕ + θ − ω = 2k · 180◦ + (ω + θ − ϕ), then 2(ϕ − ω) = 2k · 180◦ , with |ϕ − ω| < 180◦
forcing k = 0 and ϕ = ω. If ϕ + θ − ω = 2k · 180◦ − (ω + θ − ϕ), then 2θ = 2k · 180◦ , which
contradicts the fact that 0◦ < θ < 180◦ . Hence ϕ = ω, and so P I is the angle bisector of
∠QP B.
Therefore the distance of I from P Q is the same with the distance of I from AB, which
is equal to the inradius of 4ABC. Consequently, P Q is tangent to the incircle of 4ABC.
4
Problem 5 (Netherlands).
Let n ≥ 2 be an integer, and let a1 , a2 , . . . , an be positive integers. Show that there exist
positive integers b1 , b2 , . . . , bn satisfying the following three conditions:
1. ai ≤ bi for i = 1, 2, . . . , n;
2. the remainders of b1 , b2 , . . . , bn on division by n are pairwise different; and
n−1 a1 + · · · + an
3. b1 + · · · + bn ≤ n + .
2 n
(Here, bxc denotes the integer part of real number x, that is, the largest integer that does
not exceed x.)
Solution 1. We define the bi recursively by letting bi be the smallest integer such
that bi ≥ ai and such that bi is not congruent to any of b1 , . . . , bi−1 modulo n. Then
bi − ai ≤ i − 1, since of the i consecutive integers ai , ai + 1, . . . , ai + i − 1, at most i − 1
are
Pn congruentPn to one of b11, . . . , bi−1 modulo n. Since all bi are
Pn distinct 1
modulo n, we have
i=1 bi ≡
P i=1 (i −P1) = 2 n(nP − 1) modulo n, so n divides P i=1 bi − 2 n(n − 1). Moreover,
we have ni=1 bi − ni=1 ai ≤ ni=1 (i − 1) = 21 n(n − 1), hence ni=1 bi − 21 n(n − 1) ≤ ni=1 .
P
As the left hand side is divisible by n, we have
n
! " n #
1 X 1 1X
bi − n (n − 1) ≤ ai
n i=1 2 n i=1
as required.
Solution 2. Note that the problem is invariant under each of the following operations:
h(t) ≥ 1,
h(t) + h(t + 1) ≥ 2,
h(t) + h(t + 1) + h(t + 2) ≥ 3,
..
.
5
multiples of n and reordering our elements, we may assume a1 = t, a2 ∈ {t, t + 1},
a3 ∈ {t, t + 1, t + 2}, et cetera. Finally subtracting t from the complete sequence, we may
assume a1 = 0, a2 ∈ {0, 1}, a3 ∈ {0, 1, 2} et cetera. Now simply setting bi = i − 1 for all i
suffices, since ai ≤ bi for all i, the bi are all different modulo n, and
n Pn
X n(n − 1) n(n − 1) i=1 ai
bi = ≤ +n .
i=1
2 2 n
Put xi = h(i) − 1 for all i = 0, . . . , n − 1. Note that xi ≥ −1, because h(i) ≥ 0. If we have
xi ≥ 0 for all i = 0, . . . , n − 1, then taking t = 0 completes the proof. Otherwise, we can
pick some index j such that xj = −1. Let yi = xi where i = 0, . . . , j − 1, j + 1, . . . , n − 1
and yj = 0. For sequence {yi } we have
n−1
X n−1
X n−1
X
yi = xi + 1 = h(i) − n + 1 = 1,
i=0 i=0 i=0
di mod n, when i ∈ I,
bi = n + 1 + (di mod n), when i ∈ {1, . . . , n} \ I,
n, for i = n + 1.
6
Now, ai ≤ di mod n ≤ bi for i ∈ I, while for i ∈ / I we have ai ≤ n ≤ bi . Thus the sequence
n+1
(bi )i=1 satisfies the first condition from the problem statement.
By the induction hypothesis, the numbers di mod n are distinct for i ∈ {1, . . . , n}, so the
values bi mod (n + 1) are distinct elements of {0, . . . , n − 1} for i ∈ {1, . . . , n}. Since
bn+1 = n, the second condition is also satisfied.
Denote k = |I|. We have
n+1
X n
X n
X
bi = bi + n = di mod n + (n − k)(n + 1) + n =
i=1 i=1 i=1
n(n + 1)
+ (n − k)(n + 1),
2
hence we need to show that
"P #
n+1
n(n + 1) n(n + 1) ai
i=1
+ (n − k)(n + 1) ≤ + (n + 1) ;
2 2 n+1
X X n(n − 1)
di mod n + (n + di mod n) = + (n − k)n
i∈I
2
i∈I
/
or Pn
i=1 ai
n−k ≤ .
n
7
We start at step 1 by choosing f (1) to be the maximum i in {1, . . . , n} such that ai ≤ n−1,
that is f (1) = n. We set bf (1) = n − 1.
Having performed steps 1 through j, at step j + 1 we set f (j + 1) to be the maximum i in
{1, . . . , n} \ {f (1), . . . , f (j)} such that ai ≤ n − j − 1, if such an index exists. If it does,
we set bf (j+1) = n − j − 1. If there is no such index, then we define T = j and assign to
the terms bi , where i ∈ / f ({1, . . . , j}), the values n, n + 1 . . . , 2n − j − 1, in any order, thus
concluding the run of our algorithm.
Notice that the sequence (bi )ni=1 satisfies the first and second required conditions by con-
struction. We wish to show that it also satisfies the third.
Notice that, since the values chosen for the bi ’s are those from n − T to 2n − T − 1, we
have n
X n(n − 1)
bi = + (n − T )n.
i=1
2
8
with its residue modulo n). Then in fact in fact in whichever order we would flatten the
two stacks, the total number of forward steps would be the same, and the total sum of
the corresponding bt (such that at mod n ∈ {i, j}) would be the same. Moreover, we can
merge the stacks to a single stack of k + l numbers at residue i, by replacing each at ≡ j
(mod n) by a0t = at − (j − i), and this stack would be flattened forward into the same
positions as the separate stacks would have
P been, so applying our algorithm to theP new
stacks will yieldjthe same total sum of bi – but the ai s are strictly decreased, so ai
P k
ai
is decreased, so n is not increased – so by merging the stacks, we can only make the
inequality we wish to prove tighter.
Thus, as long as there is some stack that when flattened will overtake another stack, we
may merge stacks and only make the inequality tighter. Since the amount of numbers
equals the amount of places, the merging process terminates with stacks of sizes k1 , . . . , km ,
such that the stack j, when flattened, will exactly cover the interval
Pkj to the next stack.
−1 k (k −1)
Clearly the numbers in each such stack were advanced by a total of t=1 = j 2j , thus
P P P k (k −1) P
bi = ai + j j 2j . Writing ai = n · r + s with 0 ≤ s < n, we must therefore
show
X kj (kj − 1) n(n − 1)
s+ ≤ .
j
2 2
Ending 1. Observing that both sides of the last inequality are congruent modulo n (both
are congruent to the sum of all different residues), and that 0 ≤ s < n, the inequality is
P kj (kj −1)
eqivalent to the simpler
P j 2
≤ n(n−1)
2
. Since x(x − 1) is convex, and kj are non-
negative integers with j kj = n, the left hand side is maximal when kj 0 = n and the rest
are 0, and then eqaulity is achieved. (Alternatively it follows easily for any non-negative
reals from AM-GM.)
Ending 2. If m = 1 (and k1 = n), then all numbers are in a single stack and have the
P k (k −1)
same residue, so s = 0 and equality is attained. If m ≥ 2, then by convexity j j 2j
is maximal for m = 2 and (k1 , k2 ) = (n − 1, 1), where it equals (n−1)(n−2)
2
. Since we always
have s ≤ n − 1, we find
X kj (kj − 1) (n − 1)(n − 2) n(n − 1)
s+ ≤ (n − 1) + =
j
2 2 2
as required.
Greedy algorithm variant 1’ (ISR). We apply the same algorithm as in the previous
solution. However, this time we note that we may merge stacks not only when they
overlap after flattening, but also when they merely touch front-to-back: That is, we relax
the condition j ∈ (i, i + k) to j ∈ (i, i + k]; the argument for why such merges are allowed
is exactly the same (But note that this is now sharp, as merging non-touching stacks can
cause the sum of bi s to decrease).
We now observe that as long as there at least two stacks left, at least one will spread
to touch (or overtake) the next stack, so we can perform merges until there is only one
stack left. We are left with verifying that the inequality indeed holds for the case of only
one stack which is spread forward, and this is indeed immediate (and in fact equality is
achieved).
9
Greedy algorithm variant 2 (ISR). Let ci = ai mod n. Iteratively define bi = ai + li
greedily, write di = ci + li , and observe that li ≤ n − 1 (since all residues are present in
ai , . . . , ai + n − 1), hence 0 ≤ di ≤ 2n − 2. Let I = {i ∈ I : di ≥ n}, and note that di = bi
mod n if i ∈ / I and di = (bi mod n) + n if i ∈ I. Then we must show
P
X X n(n − 1) ai
(ai + li ) = bi ≤ +n
2 n
P
X X ci
⇐⇒ (ci + li ) ≤ (bi mod n) + n
n
P P P
ci ci ci
⇐⇒ n|I| ≤ n ⇐⇒ |I| ≤ ⇐⇒ |I| ≤
n n n
Let k = |I|, and for each 0 ≤ m < n let Jm = {i : ci ≥ n − m}. We claim that there must
be some m for which |Jm | ≥ m + k (clearly for such m, at least k of the sums dj with
j ∈ Jm must exceed n, i.e. at least k of the elements of Jm must also be in I, so this m is
a “witness” to the fact |I| ≥ k). Once we find such an m, then we clearly have
X
ci ≥ (n − m)|Jm | ≥ (n − m)(k + m) = nk + m(n − (k + m)) ≥ nk = n|I|
10
Problem 6 (United Kingdom).
On a circle, Alina draws 2019 chords, the endpoints of which are all different. A point is
considered marked if it is either
Alina labels each marked point. Of the 4038 points meeting criterion (i), Alina labels
2019 points with a 0 and the other 2019 points with a 1. She labels each point meeting
criterion (ii) with an arbitrary integer (not necessarily positive).
Along each chord, Alina considers the segments connecting two consecutive marked points.
(A chord with k marked points has k − 1 such segments.) She labels each such segment
in yellow with the sum of the labels of its two endpoints and in blue with the absolute
value of their difference.
Alina finds that the N + 1 yellow labels take each value 0, 1, . . . , N exactly once. Show
that at least one blue label is a multiple of 3.
(A chord is a line segment joining two different points on a circle.)
Solution 1. First we prove the following:
Lemma: if we color all of the points white or black, then the number of white-black edges,
which we denote EW B , is equal modulo 2 to the number of white (or black) points on the
circumference, which we denote CW , resp. CB .
Observe that changing the colour of any interior point does not change the parity of EW B ,
as each interior point has even degree, so it suffices to show the statement holds when all
interior points are black. But then EW B = CW so certainly the parities are equal.
Now returning to the original problem, assume that no two adjacent vertex labels differ
by a multiple of three, and three-colour the vertices according to the residue class of the
labels modulo 3. Let E01 denote the number of edges between 0-vertices and 1-vertices,
and C0 denote the number of 0-vertices on the boundary, and so on.
Then, consider the two-coloring obtained by combining the 1-vertices and 2-vertices. By
applying the lemma, we see that E01 + E02 ≡ C0 mod 2.
Using the fact that C0 = C1 = 2019 and C2 = 0, we deduce that either E02 and E12 are
even and E01 is odd; or E02 and E12 are odd and E01 is even.
But if the edge labels are the first N non-negative integers, then E01 = E12 unless N ≡ 0
modulo 3, in which case E01 = E02 . So however Alina chooses the vertex labels, it is not
possible that the multiset of edge labels is {0, . . . , N }.
Hence in fact two vertex labels must differ by a multiple of 3.
Solution 2. As before, colour vertices based on their label modulo 3.
Suppose this gives a valid 3-colouring of the graph with 2019 0s and 2019 1s on the
11
circumference. Identify pairs of 0-labelled vertices and pairs of 1-labelled vertices on the
circumference, with one 0 and one 1 left over. The resulting graph has even degrees except
these two leaves. So the connected component C containing these leaves has an Eulerian
path, and any other component has an Eulerian cycle.
∗ 0
Let E01 denote the number of edges between 0-vertices and 1-vertices in C, and let E01
denote the number of such edges in the other components, and so on. By studying whether
a given vertex has label congruent to 0 modulo 3 or not as we go along the Eulerian path
∗ ∗ ∗ ∗
in C, we find E01 + E02 is odd, and similarly E01 + E12 is odd. Since neither start nor end
∗ ∗
vertex is a 2-vertex, E02 + E12 must be even.
Applying the same argument for the Eulerian cycle in each other component and adding
0 0 0 0 0 0
up, we find that E01 + E02 , E01 + E12 , E02 + E12 are all even. So, again we find E01 + E02 ,
E01 + E12 are odd, and E02 + E12 is even, and we finish as in the original solution.
12