th

39
Iranian
Mathematical
Olympiad
 39th Iranian Mathematical Olympiad
Selected Problems with Solutions

This booklet is prepared by Ali Ahmadvand, Sina Azizeddin,

Amirmohammad Bandari Masooleh, Alireza Dadgarnia, Mohammad
Davodabadi Farahani, Amirmohammad Derakhshandeh, Pooria
Mahmoodkhan Shirazi, Ali Partofard, Pooria Rahmani, Hesam Rajabzadeh,
Navid Safaei, Mohammadamin Sharifi and Mehran Talaei.
With special thanks to Mahdi Etesamifard.
Copyright ©Young Scholars Club 2021-2022. All rights reserved.
Ministry of Education, Islamic Republic of Iran.

1
 Iranian Team Members at
at the 63rd IMO (Oslo - Norway)

Top row from left to right:

• Sina Azizeddin
• Seyed Mobin Razavi
• Amirmohammad Bandari Masooleh

Bottom row from left to right:

• Pooria Mahmoodkhan Shirazi
• Mehran Talaei Khajehroshanaei
• Danial Parnian

2
Contents

Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

Problems
Second Round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Third Round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Team Selection Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Solutions
Second Round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Third Round . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Team Selection Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3
Preface

The 39th Iranian National Mathematical Olympiad consisted of four rounds.

All the rounds of this edition of national Olympias was affected by the covid
19 pandemic. The First Round was held on February 2021 nationwide. The
exam consisted of 16 multiple-choice and short answer questions and a time
of 2 hours. In total, more than 11000 students participated in the exam and
more than 700 of them were admitted for participation in the next round.
The Second Round was held on two days in May 2021. In each day, partic-
ipants were given 3 problems and 4.5 hours to solve them. After this round,
the top 80 students were selected to participate in the Third Round.
The selection process of the Third Round consisted of four separate exams.
At the end of this round, the top 14 students were awarded a gold medal and
majority of the others were awarded by silver and bronze medals. The fol-
lowing list represents the names of the gold medalists:

1. Ali Ahmadvand
2. Abolfazl Daniali
3. Mohammad Davodabadi Farahani
4. Arshia Eslami
5. Ehsan Heidari
6. Amirmohammad Karami Abolverdi
7. Pooria Mahmoodkhan Shirazi
8. Matin Mohammadi
9. Mohammad Mohammadian Bisheh
10. Danial Parnian
11. Pooria Rahmani
12. Seyed Mobin Razavi
13. Mehran Talaei Khajehroshanaei
14. Pooria Zareei

The Team Selection Test was held on 4 days, having the same structure
as the International Mathematical Olympiad (IMO). In the end, the top 6
participants were selected to become members of the Iranian Team at the
63rd IMO.

4
 In this booklet, we present the 6 problems of the Second Round, 16 prob-
lems of the Third Round, and 12 problems of the Team Selection Tests,
together with their solutions.
It’s a pleasure for the authors to offer their grateful appreciation to all the
people who have contributed to the conduction of the 39th Iranian Mathemat-
ical Olympiad, including the National Committee of Mathematics Olympiad,
problem proposers, problem selection groups, exam preparation groups, coor-
dinators, editors, instructors and all those who have shared their knowledge
and effort to increase the Mathematics enthusiasm in our country, and as-
sisted in various ways to the conduction of this scientific event.

The 39th Iranian Mathematical Olympiad Problem Selection Committee:

Mohammadreza Badri, Amirhossein Bagheri, Alireza Dadgarnia, Alireza
Danaei, Mahdi Etesamifard (The supervisor of team selection test), Seyed
Hesam Firoozi, Faraz Ghahremani, Alireza Haghi, Seyedreza Hosseini
Dolatabadi, Parsa Hosseini, Seyed Amir Parsa Hosseini Nayyeri, Erfan
Moeini, Ali Partofard, Ali Mirzaei, Amirabbas Mohammadi, Mohammad
Moshtaghifar, Alireza Rezaeimoghaddam, Morteza Saghafian,
Mohammadamin Sharifi, Matin Yadollahi, Elahe Zahiri, Arian Zamani.

5
Problems
Second Round

1. (Amin Bahjati) Let A, B be two points on a plane and M be the midpoint

of AB. We firstly choose a point P on the segment AB, other than A, B, M .
At step i we choose a red point Pi then choose one ofA and B, call it Xi , and
reflect Pi with respect to Xi to get Qi , then color the midpoint of Qi Xi red.
Is it possible that after a few steps we color the M by red?
(→ p.17)
2. (Mohammad Shahverdi) We call a natural number n nice if it has no zero
digit in its decimal representation and it has a digit b in this representation
that the resulting number would be a divisor of n, after its removal . Prove
that there are only finitely many nice numbers.
(→ p.17)
3. (Mohammad Javad Shabani) Let ABCD be a quadrilateral and w is a
circle inscribed in it. The circle ω is tangent to BC and AD at E and F ,
respectively and DE meets ω at X, for the second time. If AB, CD are
tangent to the circumcircle of DXF , prove that AF XC is cyclic.
(→ p.17)
4. (Mohammad Shahverdi) There are n point on the perimeter of a circle w.
The interior of circle ω contains all the n points and its diameter is less than
w. Prove that there is a diagonal of w such that its endpoints are not on of
the n points and all the n points are lying on the one side of this diagonal.
(→ p.18)
5. (Erfan Salavati) Given1400 real numbers, prove that among them there
are at least three numbers x, y, z such that
 
 (x − y)(y − z)(z − x)  9
 1 + x4 + y 4 + z 4  < 1000 .
 

(→ p.19)

7
6. (Morteza Saghafian) Is it possible to write 1400 natural numbers(not
necessarily distinct) around a circle such that 2021 used at least once and
each number be the sum of greatest common divisor of the two previous
numbers and the greatest common divisor of the two next numbers? For
example, if a, b, c, d, e are five consecutive numbers on the circle then c =
gcd(a, b) + gcd(d, e).
(→ p.19)

8
Third Round

Algebra
1. (Mohammad Sharifi Kiasari) Let a, b, c, d be positive real numbers such
that a + b + c + d = 4. Prove that
ab bc cd da
+ + + ≤ 4.
a2 − 34 a + 4
3 b2 − 43 b + 4
3 c2 − 43 c + 4
3 d2 − 43 d + 4
3

(→ p.21)
2. (Navid Safaei) Let a, b, c, d be four non-zero complex numbers such that

2|a − b| ≤ |b|, 2|b − c| ≤ |c|, 2|c − d| ≤ |d|, 2|d − a| ≤ |a|.

Prove that  
b
 + c + d + a > 7.

a b c d 2
(→ p.22)
3. (Mojtaba Zare and Ali Mirzayee) Given a polynomial P (x) with non-
negative real coefficients and a function f : R+ → R+ such that for all
positive real numbers x, y :

f (x + P (x)f (y)) = (y + 1)f (x).

i. Prove that deg P (x) ≤ 1.

ii. Find all function f : R+ → R+ and all non-constant polynomials P (x)
with non-negative real coefficients satisfying above equation.

(→ p.23)

9
Combinatorics
1. Let S be an infinite set of positive integers, such that there exist four
pairwise distinct a, b, c, d ∈ S with gcd(a, b) ̸= gcd(c, d). Prove that there exist
three pairwise distinct x, y, z ∈ S such that gcd(x, y) = gcd(y, z) ̸= gcd(z, x).
(→ p.27)
2. (Morteza Saghafian) Is it possible to write a permutation of positive in-
tegers on the cells of an infinite table (infinite from all sides) such that the
sequence of numbers in each column from bottom to top and also in each row
from left to right be increasing?
(→ p.28)
3. Let n ≥ 3 be an integer. An integer m ≥ n + 1 is called n-colorful if,
given infinitely many marbles in each of n colors C1 , C2 , . . . , Cn , it is possible
to place m of them around a circle such that for each i = 1, 2, . . . , n, in any
group of n + 1 consecutive marbles there is at least one marble of color Ci .
a. Prove that there are only finitely many positive integers which are not
n-colorful.

b. Find the largest number which is not n-colorful.

(→ p.29)

Geometry
1. (Ali Zamani) Let ABC be an acute=angled triangle and D be the foot of
altitude from A. Let K, L be the touching points of tangent lines from D to
the circles with diagonals AB and AC, respectively. Point S is given on the
plane such that

∠ABC + ∠ABS = ∠ACB + ∠ACS = 180◦ .

Prove that A, K, L and S lie on a circle.

(→ p.29)
2. (Alireza Dadgarnia) Let ABC be an acute triangle and M be the midpoint
of AB. Let K be a point such that KB intersects the line AC, ∠KM C = 90◦
and ∠KAC = 180◦ − ∠ABC. The tangent line to circumcircle of triangle
ABC at A intersects line CK at E. Prove that the reflection of line BC with
respect to CM passes through the midpoint of M E.
(→ p.30)

10
3. (Amir Mahdi Mohseni) Given triangle ABC, variable points X and Y are
chosen on segments AB and AC, respectively. Let Z be a point on the line
BC such that ZX = ZY . The circumcircle of XY Z intersects the line BC at
T , for the second time. Point P is chosen on line XY such that ∠P T Z = 90◦ .
Let Q be a point on the same side of line XY as A, satisfying ∠QXY = ∠ACP
and ∠QY X = ∠ABP . Prove that the circumcircle of triangle QXY passes
through a fixed point (as X and Y vary).
(→ p.31)

Number Theory
1. (Morteza Saghafian) For each positive integer n, denote by f (n) the total
number of positive integers less than n such that neither are coprime to n nor
are a divisor of n. For example f (12) = 3 because among positive integers less
than 12 there are only three numbers, i.e., 8, 9, 10 with the aforementioned
property.
Prove that for each positive integer k the equation f (n) = k has only finitely
many solutions.
(→ p.32)
2. (Ilia Mahrooghi) Find all functions f : N → N such that for all positive
integers a, b we have

f a (b) + f b (a)  2(f (ab) + b2 − 1) .



By f a (x) we mean the a−times fold composition of f.

(→ p.33)

3. (Mohammadamin Sharifi and Navid Safaei) Let x1 be a fixed positive

integer. Prove that there is no positive integer m > 2500 such that the
S(n)
sequence xn+1 = xn + 1 is eventually periodic modulo m. Where S(n) is
sum of digits of n in base 10.
(We say a sequence eventfully periodic mod m if there are positive integers
N, T such that xn ≡ xn+T (mod m) for all n ≥ N ).
(→ p.34)

Final Exam
1. (Morteza Saghafian) Is it possible to assign numbers 1, 2, . . . , 8 to the ver-
tices of a cube in a way that the assigned number of each vertex divides the
sum of numbers assigned to its neighbours? (Note that every number should
be used once).
(→ p.36)

11
2. (Mahdi Etesamifard) Given an acute-angled triangle ABC with an altitude
AD and orthocenter H. Let E be the reflection of H with respect to A. A
point X lies on the circumcircle of the triangle BDE such that DX ∥ AC,
and similarly a point Y lies on the circumcircle of the triangle CDE such
that DY ∥ AB. Prove that the circumcircles of triangles AXY and ABC are
tangent to each other.
(→ p.37)
3. (Matin Yousefi) Find all functions f : Q[X] → R such that for all polyno-
mials P (x)andQ(x) in Q[X];

i. f (P (Q(x))) = f (Q(P (x)));

ii. f (P (x)Q(x)) = f (P (x)) + f (Q(x)), where P (x)Q(x) ̸= 0.

(→ p.38)

4. (Sam Khanaki) Arash and Babak play the following game on a 1400×1401
table. Starting with Arash, He in his turns colors k number of L-shaped
trominos on the table (rotation and reflection are allowed). Also, Babak in
his turns colors one 2 × 2 square from the table. Every cell of the table can
be colored at most once and a player who can not color the cells in his turn
will lose. Find all values of k for which Arash has a winning strategy.
(→ p.38)

12
Team Selection Test

1. (Morteza Saghafian) Morteza has 100 sets. At each step Mahdi can choose
two distinct sets from them and Morteza tell him the intersection and union
of those two sets. Find the least possible number of steps that Mahdi can
determine all of 100 sets.
(→ p. 40)
2. (Mohammadamin Sharifi) For a positive integer n, let τ (n) and σ(n) be
the number of positive divisors of n and the sum of positive divisors of n,
respectively. Let a and b be positive integers such that σ (an ) divides σ (bn )
for all n ∈ N. Prove that each prime factor of τ (a) divides τ (b).
(→ p. 40)
3. (Mahdi Etesamifard and Alireza Dadgarnia) Incircle ω of triangle ABC
is tangent to sides CB and CA at D and E, respectively. Point X is the
reflection of D with respect to B. Suppose that the line DE is tangent to the
A-excircle at Z. Let the circumcircle of triangle XZE intersects ω at K, for
the second time. Prove that the intersection of BK and AZ lies on ω.
(→ p. 41)
4. (Amir Parsa Hosseini Nayeri) Cyclic quadrilateral ABCD with circumcen-
ter O is given. Point P is the intersection of diagonals AC and BD. Let M
and N be midpoints of the sides AD and BC, respectively. Suppose that ω1 ,
ω2 and ω3 are circumcircles of triangles ADP , BCP and OM N , respectively.
Let E and F be intersection points of ω1 and ω3 , which is not on the arc
AP D of ω1 , and the intersection point of ω2 and ω3 , which is not on the arc
BP C of ω2 , respectively. Prove that OE = OF .
(→ p. 42)
5. (Navid Safaei) Find all real numbers C such that every sequence an of
integers which is bounded from below and for all n ≥ 2 satisfies:

0 ≤ an−1 + Can + an+1 < 1.

it is periodic. (→ p. 43)

13
6. (Shayan Gholami) Let m, n and a1 , a2 . . . , am be arbitrary positive inte-
gers. Ali and Mohammad play the following game; At each step, Ali chooses
b1 , b2 . . . , bm ∈ N, then Mohammad chooses a positive integer s and obtains
a new sequence {ci = ai + bi+s }m i=1 , where

bm+1 = b1 , bm+2 = b2 , . . . , bm+s = bs .

Through a finite number of steps, Ali intends to make all the numbers divisible
by n. Find all positive integers m and n such that Ali,independent of the
initial values a1 , a2 . . . , am , has a winning strategy.
(→ p. 45)
7. (Mohammad Ahmadi) Suppose that n is a positive integer. Consider a
regular 2n-gon such that one of its largest diagonals is parallel to the x-axis.
Find the smallest integer d such that there is a polynomial P (x) of degree
d whose graph intersects all sides of the polygon on points other than its
vertices.
(→ p. 46)
8. (Amirmahdi Mohseni) In triangle ABC, with AB < AC, I is the incenter,
E is the intersection of A-excircle, and BC. Point F lies on the external
angle bisector of ∠BAC such that E and F lies on the same side of the line
AI and ∠AIF = ∠AEB. Point Q lies on BC such that ∠AIQ = 90◦ . Circle
ωb is tangent to F Q and AB at B, circle ωc is tangent to F Q and AC at C
and both circles are passing through the inside of triangle ABC. If M is the
midpoint of the arc BC, which does not contain A, prove that M lies on the
radical axis of ωb and ωc .
(→ p. 47)
9. (Morteza Saghafian) Consider n ≥ 6 points x1 , x2 , . . . , xn on the plane such
that no three of them are colinear. We call a graph with vertices x1 , x2 , . . . , xn
a "road network " if it is connected, each of its edges is a line segment, and
no two edges intersect at a point other than its vertices. Prove that there are
three road networks G1 , G2 , G3 such that Gi and Gj don’t have a common
edge for 1 ≤ i, j ≤ 3.
(→ p. 48)
10. (Seyed Reza Hosseini Dolatabadi) We call an infinite set S ⊆ N good if
ac − bc
for all pairwise distinct integers a, b, c ∈ S, all positive divisors of are
a−b
in S. For all positive integers n > 1, prove that there exists a good set S such
that n ̸∈ S.
(→ p. 49)

14
11. (Seyed Mohammad Seyedjavadi and Alireza Tavakoli) Consider a table
with n rows and 2n columns. We put some blocks in some of the cells. After
putting blocks on the table we put a robot on a cell and it starts moving in
one of the following four directions: right, left, down, or up. It can change the
direction only when it reaches a block or a border. Find the smallest value
of m such that we can put m blocks on the table and choose a starting point
for the robot so it can visit all of the unblocked cells. (the robot cannot enter
the blocked cells.)
(→ p. 49)

12. (Matin Yousefi) Suppose that A is the set of all closed intervals [a, b] ⊂ R.
Find all functions f : R → A such that
i x ∈ f (y) if and only if y ∈ f (x).

ii. |x − y| > 2 if and only if f (x) ∩ f (y) = ∅.

iii. For all real numbers 0 ≤ r ≤ 1; f (r) = [r2 − 1, r2 + 1].
(→ p. 50)

15
Solutions
Second Round

1. We can assume that AB is the real line,A = 0, B = 2. At each step we

choose a red point x and we color on of the −x
2 or 2 . consider the converse:
3−x

we choose a red point x and color the −2x or 6 − 2x red.

If we can color M red at some point, it would be possible to start from M
and use inverse steps to reach P , so the problem is equivalent to this: is it
possible to start at 1 and do the converse steps to reach a point 0 < P < 2.
From now on we prove this formulation and call converse steps simply steps.
If we start at 1 after one step we are either at −2 or 4. We claim that
if you start at some point x ∈ (−∞, −2] ∪ [4, ∞) you will remain in this
sets. To see this note that if x ≤ −2, −2x ≥ 4, 6 − 2x ≥ 10, and if x > 4,
−2x ≤ −8, 6 − 2x ≤ −2. So we can not reach from 1 to some 0 < x < 2. ■

2. We prove that a nice number has at most 20 digits which proves that
there are finitely many nice numbers. Assume that n = abc where a, c can
have more than one digit and b is a digit such that ac = nk for some natural
number k. If c has t digits, then n = ac (mod 10t−1 ) therefore we have kn = n
(mod 10t−1 ). Because there is no zero digit in n, we know that gcd(n, 2) = 1
or gcd(n, 5) = 1. If gcd(n, 2) = 1 then we have k = 1 (mod 2t−1 ), on the other
hand n < 100ac so k < 7 which implies that t < 9. Similarly if gcd(n, 2) = 5
we have t < 4.
Now assume that a has more than 10 digits, we have ac | ac0,  ac | abc
therefore ac | abc − ac0. But the previous paragraph implies that abc − ac0
has at most 10 digits which implies that abc − ac0 | ac. So we should have
n = ac0 which is in contradiction with the fact that n is an interesting number.
■

3. Let Ω be the circumcircle of DXF and l be the perpendicular bisector of

F X. Note that lines AB, CD are external common tangents to Ω, w, therefore
they are reflection of each other with respect to l. Suppose that AB is tangent
to Ω at Y , and Z is the reflection of A with respect to l. The line XY is
the reflection of F D with respect to l hence F D passes through A, and XY
passes through Z.

17
 It is clear that AF XZ is an isosceles trapezoid therefore AF XZ is cyclic
and it is suffices to prove that the circumcircle of F XZ passes through C.
>
Note that ∠BEF = ∠F XE = F XD 2 = ∠F DC therefore F ECD is cyclic and
we have

∠F CZ = ∠F CD = ∠F ED = ∠F EX = ∠F XY = 180 − ∠F XZ.

Which implies that F XZC is cyclic.

B

A ω

Y F

l E

X
D
Z

4. Let O be the center of w and O′ be the center of Ω. Suppose that the

perpendicular from O to OO′ intersects w at A and B, we prove that all the
n points are on the side of AB that contains O′ . Assume P and O′ are on
different side of AB, it is clear that ∠P OO′ ≥ 90◦ , therefore P O′ > P O.
This implies that P can’t be one of the n points because otherwise P O is the
diameter of w and P O′ is less than the diameter of Ω which contradicts to
the fact that the diameter of w is larger then the diameter of Ω . ■

18
5. Putting C = 9
1000 , n = 1400. Assume to the contrary that for all z ≥ y ≥ x
we have
(z − y)(y − x)(z − x) ≥ C(1 + x4 + y 4 + z 4 ).
Notice that 4(y − x)(z − y) ≤ (y − x + z − y)2 = (z − x)2 . Whence,

(z −x)3 ≥ 4(z −y)(y −x)(z −x) ≥ 4C(1+x4 +y 4 +z 4 ) ≥ 4C(1+x4 +z 4 ) ≥ 4C.

√
It follows that z − x ≥ 3 4C. On the other hand, we can also deduce that

(z − x)3
≥ 4C.
x4 + z 4
Since 2(x4 +z 4 ) ≥ (x2 +z 2 )2 and 2(x2 +z 2 ) ≥ (x−z)2 . Therefore, 8(x4 +z 4 ) ≥
(x − z)4 . We would obtain

8 8(z − x)3 (z − x)3

= 4
≥ 4 ≥ 4C.
z−x (x − z) x + z4

That is,
√
3 2
4C ≤ z − x < .
C
Finally,
√ assume that x1 ≤ · · · ≤ xn are given. We would then have xk+2 −xk >
3
4C. Hence,
n−1 √
 
2 3
> xn − x1 ≥ 4C.
C 2
Hence, √
3
2
C< 3 .
n−1 2
2

Yielding, C < 87
10000 . A contradiction. Thus, we are done. ■

6. We shall show that this is impossible. Assume that there are such 2021
numbers. First of all, note that of we divide all the numbers by k then the
new numbers satisfy the second condition and one of them is a divisor of 2021,
so we can assume that the greatest common divisor of the numbers is 1.
Lemma. the gcd of each three consecutive number on the circle is 1.
Proof. Assume that a, b, c, d, e are five consecutive numbers on the circle, if
a, b, c has a common divisor d, then the equality c = gcd(a, b) + gcd(d, e)
implies that d, e are also divisible by d.Similarly all the numbers should be
divisible by d which is in contradiction with our assumption. This completes
our proof.

19
 Assume that m is the maximum number on the circle, and x, y, m, z, t
be five consecutive numbers. We know that m > 4 because at least one of
the numbers should be a divisor of 2021 and 2021 is not divisible by 2, 3, 4.
We have m = gcd(x, y) + gcd(z, t) so at least one of the gcd(x, y), gcd(z, t)
should be at least m 2 . Without the loss of generality we can assume that
gcd(x, y) ≥ m 2 .
If x ̸= y then max(x, y) ≥ m, yielding max(x, y) = m. If y = m then by
the assumption x > gcd(m, y) = m we shall arrive at a contradiction hence we
have x = m, y = m 2 . This implies that gcd(z, t) = 2 therefore z = 2 , y = m
m m

which contradicts to the fact that y > gcd(z, m).

So we have x = y ≥ m 2 , let w be the number before x. From the lemma
we know that gcd(w, x) = 1 and we have

x = gcd(w, x) + gcd(m, z) = 1 + (m, z) =⇒ x − 1 | m.

If m is odd this contradicts to the fact that x − 1 ≥ m

2 − 1, if m is even we
have x = m2 + 1 and gcd(m, z) = 2 . But we have
m

m m m  m
m = gcd(x, y) + gcd(z, t) = + 1 + ( , t) =⇒ −1  .

2 2 2 2
Which contradicts to the fact that m
2 > 2. ■

20
Third Round

Algebra
1. First Solution Note that a2 − 43 a+ 34 = a2 +1− 43 a+ 13 ≥ 2a− 34 a+ 13 = 2a+1
3 .
Hence,  
ab ab 3ab 1 3b
≤ = = 3b − .
a2 − 43 a + 43 2a+1
3
2a + 1 2 2a + 1
That is,
 
X ab 1X 3b 3 3X b
≤ 3b − = (a + b + c + d) −
cyc
a − 43 a +
2 4
3
2 cyc 2a + 1 2 2 cyc 2a + 1
3X b
=6− .
2 cyc 2a + 1

Now, it suffices to prove

X b 4
≥ .
cyc
2a + 1 3

Notice that
X b X b2 (a + b + c + d)2 8
= ≥ = .
cyc
2a + 1 cyc
2ab + b 2(a + c)(b + d) + 4 2 + (a + c)(b + d)

We need to prove (a + c)(b + d) ≤ 4. Indeed,

1
(a + c)(b + d) ≤ (a + b + c + d)2 = 4.
4
The equality case occurs whenever a = b = c = d = 1. ■

Second Solution. We can prove that a2 − 4aa+ 4 ≤ a+2 3 by using the equation
3 3
of the tangent line to graph of f (a) = a2 − 4 a+ 4 at a = 1 or prove it by
a
3 3

21
algebraic calculations leading to (a − 1)2 (3a + 8) ≥ 0. Then, the left side is
less than or equal to
X a + 2 1 8 4 8
b = (a + c)(b + d) + ≤ + = 4.
cyc
3 3 3 3 3

■
Comment. The original statement of the problem was:
Let λ ∈ [1, 2] and a, b, c, d be positive real numbers such that a + b + c + d = 4.
Prove that
ab bc cd da
+ 2 + 2 + 2 ≤ 4.
a2 − λa + λ b − λb + λ c − λc + λ d − λd + λ
Indeed, since the left side is a convex function with respect to λ, its maximum
achieved at the endpoints. So, it suffices to prove the statement for λ =
1, 2. Both cases are so easy. That was the chief reason why we changed the
statement to a particular case, i.e., λ = 34 .

2. We can rewrite the assumptions of our problem as follows

a  1  b 
 1 c  1  d 
 1
 − 1 ≤ ,  − 1 ≤ ,  − 1 ≤ ,  − 1 ≤ .
     
b 2 c 2 d 2 a 2

Putting ( ab , cb , dc , ad ) = (x, y, z, t) such that |x − 1| ≤ 21 , |y − 1| ≤ 12 , |z − 1| ≤

2 , |t − 1| ≤ 2 we are going to prove that
1 1

 
1
 + 1 + 1 + 1 > 7.

x y z t 2

by letting x = u + iv, for some real numbers u, v such that u2 + v 2 ≤ 2u − 34 .

It follows that u ≥ 38 . Since

1 u − iv
= 2 ,
x u + v2
We would obtain
  X  !
1
 + 1 + 1 + 1 = 
  u − iv  X u − iv
≥ |ℜ |
x y z t  u2 + v 2  cyc
u2 + v 2
 
X u  X u
=  2 2
≥ .
u +v 
cyc
u + v2
2

22
 Since u ≥ 1 3
we would obtain


2 u2 + v 2 + 4

3
X u 1 X u2 + v 2 + 4 3X 1 3X 1
≥ =2+ =2+ .
cyc
u2 +v 2 2 cyc u2 + v 2 2
8 cyc u + v 2 8 cyc |x|2

Notice that xyzt = 1 thus, according to AM-GM inequality,

s
X 1 1
2 ≥4 2 2 = 4.
4
2 2
cyc |x| |x| |y| |z| |t|

Hence,
X u 3 7
≥2+4· = .
cyc
u2 + v 2 8 2

Notice that this inequality has no equality case. ■

Comment 1. Since u2 +v 2 ≤ 2u− 34 = 4 .
8u−3
Hence, according to AM-GM;
P 4
cyc (8u − 3) Y Y
≥ (8u − 3) ≥ 44 (u2 + v 2 ) = 44 .
4 cyc cyc

It follows that X 7
u≥ .
cyc
2

Thus, |x + y + z + t| ≥ u ≥ 72 .
P
cyc

Comment 2. In the original proposal it was asked to prove

√
   
a b c d   b c d a 
max  + + +  ,  + + +  > 2 3.

b c d a a b c d

3. First Solution. We shall provide the solution for each part, separately;
Part i. Since f (x) is injective and limx→+∞ f (x) = +∞ set f (y) − 1 instead
of y, it follows that

f (x + P (x)f (f (y) − 1)) = f (x)f (y).

if you change the place of x, y by injectivity it follows that

x + P (x)f (f (y) − 1) = y + P (y)f (f (x) − 1).

Set y = a, b = f (a) > 1 we have

x + P (x)f (b − 1) = a + P (a)f (f (x) − 1).

23
 Therefore
x − a + P (x)f (b − 1) x − a + AP (x)
f (f (x) − 1) = = .
P (a) B
That is
   
y − a + AP (y) x − a + AP (x)
x + P (x) = y + P (y) .
B B
Hence, Bx+(y −a)P (x) = By +(x−a)P (y). That is, setting y = c ̸= a,

Bx + (c − a)P (x) = Bc + (x − a)P (c).

Thus,
(P (c) − B)x + Bc − aP (c)
P (x) = .
c−a
Hence, P (x) = rx + s or P (x) = C.
Part ii. If P (x) = rx + s, then f (f (y) − 1) = ux + v. Then

f (x + P (x)f (f (y) − 1)) = f (N xy + M x + T y + k) = f (x)f (y).

Therefore, since the right side is symmetric with respect to x, y; then

M = T;
√  √  
f M x − r1 M y − r1 + k ′ = f (x)f (y),

Set f (x + k ′ ) = h(x); Then,

√  √ 
h M x − r1 M y − r1 = h(x − k ′ )h(y − k ′ ).
√ √
By setting M x − r1 = Z, M y − r1 = Y. and plugging x = Z+r
√ 1,
M
y=
Y√+r1
M
.

  
Z + r1 Y + r1
√
h(ZY ) = h h √ .
M M
in the equation then for Y = 1, we have:
 
Z + r1
h(Z) = h √ D,
M
Hence,
h(Z)h(Y )
h(ZY ) = .
D
Now we set
h(Z)
g(Z) = .
D

24
 Hence,
g(Z)g(Y ) = g(Y Z).
So, we have a multiplicative function which is bounded from below, thus
g(x) = xk for some real number k. We can easily find that g(x) = x.
The rest is clear.

■
Comment 1. If P (x) = C then f (f (x) − 1) = ux + v and hence,

f (x + Cy + D) = f (x)f (y).

Then, symmetry implies that C = 1,and

f (x + y + D) = f (x)f (y),

is another shifted Cauchy’s equation.

Second solution. We shall again provide the solution for each part, sepa-
rately;
Part i. It is clear that the function is injective. By putting x+P (x)f (z) instead
of x in the original equation for some x, z > 0 we obtain

f (x + P (x)f (z) + P (x + P (x)f (z))) = (y + 1)f (x + P (x)f (z))

= (y + 1)(z + 1)f (x).

Interchange y, z and using the injectivity we have

P (x)(f (y) − f (z)) = f (y)P (x + p(x)f (z)) − f (z)P (x + p(x)f (y)).

now if y, z are numbers such that y ̸= z and letting f (y) = A, f (z) = B

it follows that A ̸= B and A, B > 0. Yielding

(A − B)P (x) = BP (x + AP (x)) − AP (x + BP (x)).

Assume to the contrary that d = deg P > 1. Let ad be the leading coeffi-

2
cient of P (x), the coefficient of xd in the right is ad+1
d AB Ad−1 − B d−1 .
2
Since A ̸= ±B, we find that the coefficient of xd in the right side is not
zero. That is, the right side is of degree d2 However, since d > 1 the
left side is of degree d < d2 . A contradiction. Thus, d ≤ 1.
Part ii. Let P (x) = ax + b for some a > 0, b ≥ 0. Then, the original equation
can be rewritten as

f (x + (ax + b)f (y)) = (y + 1)f (x).

25
We want to prove that b = 0. Assume that b > 0. Plugging bf (x) instead
of x yielding

f (abf (x)f (y) + b(f (x) + f (y))) = (y + 1)f (bx).

Interchanging x, y to obtain fx+1

(bx)
= c, for some constant c > 0. Plugging
bf (x) instead of x in the preceding equation yields

f (bc(x + 1)) = c(1 + bf (x)).

Now we shall prove following lemma.

Lemma. There is a positive real t such that f (t) = c.

Proof. Assume to the contrary, notice that if z ∈ ℑ f then z(y + 1) ∈

ℑ f. Hence, the image of function includes [z, +∞). That is, all u > z are
in the image of the function. According to our assumption, all numbers
d ≤ c don’t appear in the image of f. On the other hand f (x) > c and
for all d > c there is a positive real x such that d = c(x + 1) hence,
f (bf (x)) = d, therefore d is in the image of f. Moreover, bf (x) > bc
implies that for all d > c there is a real number z > bc such that
f (z) = d. Since f (bc) > c taking d = f (bc) to obtain that there is
a positive real number z > bc such that f (z) = f = f (bc), This is
impossible since the function is injective. Thus our assumption was
false and there is a positive real number t such that f (t) = c. This
completes our proof.

Now, multiplying both sides of original equation by b then taking f

from both sides and using the above equations yielding

f (bf (x + (ax + b)f (y))) = f (bf (x)(y + 1)) = c(x + (ax + b)f (y) + 1).

Hence,
f (b(y + 1)f (x)) = c(x + (ax + b)f (y) + 1).
Put x = t to obtain f (bc(y + 1)) = c(t + (at + b)f (y) + 1) = c(1 +
bf (y)). This equation is not true. Hence, b = 0. Rewrite the original
equation in the form f (x(1 + af (y))) = (1 + y)f (x). Define ff (x)
(1) = g(x),
if we rewrite the preceding equation in terms of g(x) we would obtain
g(x(1 + rg(y))) = (1 + y)g(x), for some positive real number r, i.e.,
r = af (1). This equation is quite similar to the aforementioned equation
while g(1) = 1. If x = 1 we have g(1 + rg(y)) = 1 + y this implies
that g(x) is surjective on big enough numbers, hence g(x(1 + rg(y))) =
g(x)g(1 + rg(y)). Since 1 + rg(y) covers all real numbers after some

26
 certain point, we can find that g(xz) = g(x)g(z) for all z > M. Now,
we can extend it. Let z < M choose y such that yz, y > M then

g(xz)g(y) = g(xyz) = g(x)g(yz) = g(x)g(y)g(z).

Hence, g(yz) = g(y)g(z). The function g is multiplicative. Moreover,

g(1 + rg(y)) = 1 + y implies that it is bounded from below in some
interval and tends to infinity as y tends to infinity. Hence, g(x) = xs
for some real number s, it follows from aforementioned equation that
c = 1. Hence, g(x) = x. The rest is clear.
■
Comment 2. There is also an alternative way for finishing this problem.
Indeed, g(1 + r) = 2 plugging  r to obtain g(1 + 2r) = 2 + r now
 y = 1+
2
plugging y = 1 + 2r to get g (1 + r) = (g(1 + r))2 = 4 = 2 + 2r. Hence,
r = 1. That is,
g(1 + g(y)) = 1 + y.
Plugging 1 + g(y) instead of y gives g(2 + y) = 2 + g(y). Hence,

g(x(y + 2)) = g(x)(2 + g(y)) = 2g(x) + g(x)g(y) = 2g(x) + g(xy).

Putting y = z
x for some z > 0 yielding

g(z + 2x) = 2g(x) + g(z)

Setting 2z instead of z and interchanging z, x implying that g(2x)−2g(x) = C.

For some constant C. Therefore,

g(2x + 2z) = 2g(x + z) + C = 2g(x) + g(2z) = 2g(x) + 2g(z) + C.

Therefore,
g(z + z) = g(z) + g(x).
Thus, g(x) is both additive and multiplicative. Hence, g(x) = x.

Combinatorics
1. Note that for any natural number x ∈ S we have gcd(a, x) ≤ a. Therefore,
according to the pigeonhole principle there is an infinite subset S ′ ⊂ S − {a}
such that for x, y ∈ S ′ , gcd(a, x) = gcd(a, y). If there are x, y ∈ S ′ such that
gcd(x, y) ̸= gcd(a, x), then x, a, y satisfy the desired condition. So for every
x, y ∈ S ′ , we have gcd(x, a) = gcd(y, a) = gcd(x, y). Fix the sum s ∈ S ′
and replace S ′ with the set consisting of all the elements x ∈ S such that

27
(s, a) = (x, a). We claim that S − (S ′ ∪ {a}) intersects with {b, c, d} and so is
non-empty. Besides, if b, c, d ∈ S ′ then we have

gcd(a, b) = gcd(a, c) = gcd(a, d) = gcd(c, d),

and this contradicts gcd(a, b) ̸= gcd(c, d). Fix s′ ∈ S − (S ′ ∪ {a}), clearly

(s′ , a) ̸= (s, a). Similar to the previous part, there is an infinite subset S ′′ ⊆ S ′
such that for every x, y ∈ S ′′ , gcd(x, s′ ) = gcd(y, s′ ). Again if there are
x, y ∈ S ′′ such that gcd(x, y) ̸= gcd(s′ , x), then x, y, s′ satisfy the desired
condition. So for every x, y ∈ S ′′ , we have

gcd(x, s′ ) = gcd(y, s′ ) = gcd(x, y) = gcd(y, a) = gcd(x, a).

From this and since gcd(s′ , a) ̸= gcd(s, a) we conclude that triple (a, s′ , s)
satisfies the desired condition. ■

Comment. This problem was IMO 2021 Shortlist-C1.

2. The answer is yes. First, we choose a cell as origin and fill it with zero, then
we fill the table in a way that the numbers in the cells of (2k + 1) × (2k + 1)
table centered at the origin be a permutations of numbers

(−2k 2 − 2k, −2k 2 − 2k + 1, . . . , 0, . . . , 2k 2 + 2k − 1, 2k 2 + 2k),

We proceed the filling of the table inductively. Suppose that in step k ≥ 0,

the (2k + 1) × (2k + 1) table centered at the origin is filled with the before-
mentioned numbers such that the sequence of numbers in each row and in each
column in increasing. Putting m = 2k 2 +2k. Now, consider (2k +3)×(2k +3)
table centered at the origin. Fill this table as follows.

−(m + 4k + 4) −(m + 4k + 3) ··· −(m + 2k + 3) m+1

−(m + 2k + 2) m+2
..
−(m + 2k + 1) (2k + 1) × (2k + 1) table from the k-th step .
..
. m + 2k + 1
−(m + 2) m + 2k + 2
−(m + 1) m + 2k + 3 ··· m + 4k + 3 m + 4k + 4

It can be easily verified that the columns and rows of this table form
increasing sequences. So it is clear that with this algorithm the table will be
filled with a permutation of integers and has the desired properties. ■

28
3. A block of n marble with colors 1, 2, 3, . . . , n is called a nice block. We
prove that every m ≥ n(n − 1) is colorful. Assume that m = nk + r where
0 ≤ r ≤ n−1 and n−1 ≤ k. Put k nice blocks around the circle, now consider
r consecutive nice blocks and put a marble between any two adjacent blocks
among these r blocks, inclusively (note that k ≥ n − 1 ≥ r, so r blocks are
distinct). It is easy to see that these marbles satisfy the desired condition.
Now we want to prove that n(n − 1) − 1 is not n-colorful. Assume the
contrary. Then by the pigeonhole principle, we at least have n − 2 marbles of
the same color, and by our assumption between each two of these marbles we
have at most n marbles, hence we have at most (n − 2)(n + 1) marbles. But
n(n − 1) − 1 = n(n − 2) + n − 1 > n(n − 2) + n − 2 = (n + 1)(n − 2),
which is a contradiction. ■
Comment. This problem was IMO 2021 Shortlist-C2.

Geometry
1. Suppose that L and K lie on the circumcircle of ABD and ADC, respec-
tively.

L
M
K

B D C

29
Note that ∡LAB = 90◦ − ∡ABL = 90◦ − ∡ADL = 90◦ − ∡ACB. Similarly
we have ∡CAK = 90◦ − ∡CBA. Summing these two implies that ∡LAK =
2∡BAC. Denote by T the intersection of BL, CK, then

∡T BC = 180◦ − ∡CBA − ∡ABL = 180◦ − ∡CBA − ∡ACB = ∡BAC, ,

and similarly ∡T CB = ∡BAC. Therefore ABC is an isosceles triangle and

∡CT B = 180◦ − 2∡BAC. Similarly, one can show that ∡CSB = 180◦ −
2∡BAC and so quadrilateral BCT S is cyclic. Properties of S imply that
A is the S-excenter of the triangle BCS and so SA is the angle bisector of
∠BSC. Let M be the midpoint of the arc BC (the one that does not contain
S) in the circumecircle of BCS. Clearly M lies on SA.
On the other hand, since triangle BT C is isosceles, M T is a diagonal of
the circumcircle of BCS. Thus, we can conclude that

∡AST = ∡M ST = 90◦ = ∡AKT = ∡ALT.

This implies AKLS is cyclic as desired. ■

2. Note that KA is parallel to the tangent line from C to the circumcircle of

ABC. Therefore, if we denote by O the circumcenter of ABC, then CO⊥KA.
So, denoting by F the intersection point of KA and CO, we have ∠KF C =
90◦ . On the other hand, ∠KM C = 90◦ , therefore KCF M is cyclic. We then
have ∠AF O = ∠AM O = 90◦ , hence the quadrilateral OF AM is also cyclic.
Thus,

∡M CE = ∡M CK = ∡M F K = ∡M F A = ∡M OA = ∡BCA.

On the other hand, 180 − ∡BCA = ∡BAE = ∡M AE. These facts together
imply that AECM is cyclic.
Note that ∡CEM = ∡CAM = ∡CAB, hence △BCA ∼ △M CE. Now
denote the midpoint of M E by D. As M is the midpoint of BA, similarity of
triangles BCA and M CE implies that that ∡BCM = ∡M CD. Analogously,
D lies on the reflection of line BC with respect to CM , as desired.

30
 K

A
E

O
F
B C

3. Let E, F be the intersections of the circumcircle of ABC with the lines

BP, CP , respectively. If S is the intersection of EY, F X, then by Pascal ’s
theorem it lies on the circumcircle of ABC. It is also easy to see that XY QS
is cyclic.
Let F ′ be the intersection of CX with the circumcircle of ABC and G
be the intersection of XY, BC. Note that T Z is the external angle bisector
of ∠XT Y and P T is perpendicular to T Z, so P T is the angle bisector of
∠XT Z. Thus (GP, XY ) = −1 and BY, CX, AP are concurrent. Let R be the
intersection of these lines, and D be the intersection of AP, BC. By looking
through point C, we have (DP, RA) = (GP, XY ) = −1. By projecting X to
the circumcircle of ABC and also projecting this circle onto AP through C
we have
(BC, SA) = (AF ′ , F B) = (AR, P D) = −1.

31
This shows that S is the intersection of the circumcircle of ABC with sym-
median and the proof is complete as the circumcircle of triangle QXY passes
through this fixed point S.

F
E
F′

P Q

Y
R

T
G
B Z D C

Number Theory
1. First Solution. We first state a well-known bound for the left side.
Lemma. For each composite n we have
√
n − φ(n) ≥ n.
φ(n) Qt pi −1
Proof. Let n = pα
1 . . . pt such that p1 < · · · < pt . Then,
1 αt
n = i=1 p1 ≤
p1 −1
p1 . Hence,
n √
n − φ(n) ≥ ≥ n.
p1
This completes our proof.

32
 On the other hand, none of integers from pn1 to n are not a divisor of n.
Hence,
    
n n n φ(n)
f (n) ≥ n − − φ(n) − φ ≥n− − φ(n) −
p1 p1 p1 p1
  √
1 n
≥ 1− (n − φ(n)) ≥ .
p1 2
Hence, √
n
k = f (n) ≥ .
2
It follows that n ≤ 4k 2 . ■

Second Solution. It is easy to verify that

f (n) = n − φ(n) − d(n) + 1

. It is also clear than n is not a prime number. Rewrite The equation f (n) = k
as
n − d(n) = φ(n) + k + 1.
The left side is equal to number of integers between 2 and n that doesn’t divide
n. The right sides is equal to k + 1 plus total number of integers between 2
and n such that are coprime to n. Whence, there are k + 1 numbers like a
such that a doesn’t divide n and gcd(a, n) > 1. Let p be a prime divisor of
n. If n = pm then all numbers of the form a = pr, gcd(r, m) = 1 such that
2 ≤ r ≤ m have the desired property. Hence, we at least have φ(m) − 1
choices for a. Thus,  
n
φ(m) = φ ≤ k + 2.
p
Since p was arbitrary, it follows that all primes dividing n as well as their
exponents must be bounded. Hence, we at most have finitely many choices
for n. ■

2. Let b = 1 we have f a (1) + f (a) 2f (a). Hence, 2f (a) = C(f a (1) + +f (a)),
for some positive integer C. Thus, C = 1 and f a (1) = f (a). Whence, f a (b) =
f a−1 (f (b)) = f a−1 (f b (1)) = f a+b−1 (1) = f (a + b − 1). By the same argument
f b (a) = f (a + b − 1). It then follows that f (a + b − 1) divides f (ab) + b2 − 1.
Interchanging a, b to obtain f (a + b − 1) | b2 − a2 . Hence, f (2a) divides 2a + 1.
If the function is injective, it follows that

f 2 (n) = f (f (n)) = f (n + 1).

Hence f (n) = n + 1. This is indeed a solution. If the function is not injective

then f (r) = f (s) for some r > s. Thus, f r (1) = f s (1). Hence, for each

33
positive integer m; f (m + r) = f m (f r (1)) = f m (f s (1)) = f (m + s). Hence,
the function is periodic with a period of r − s. This implies that the function
is indeed bounded. That is, for all n we have f (n) < M. Choose a prime
p > M it follows that f (p − 1) divides p and f (p − 1) < M < p. Yielding
f (p − 1) = 1. Whence, for all large enough p; f (p − 1) = f p−1 (1) = 1.
Let d be the smallest positive integer such that f d (1) = 1 by using the
divison algorithm, we can easily prove that d divides p − 1 . This means that
for all large enough p we have p ≡ 1 (mod d). This implies that d ∈ {1, 2}.
2k
Thus f (f (1)) = 1. Hence, f (2k) = f (1) = 1 for all positive integers k. On
2k+1
the other hand, f (2k + 1) = f (1) = f (1) for all non-negative integers k.
Finally putting a = b = 2 to obtain f (1) divides 4. Hence, f (2k+1) ∈ {1, 2, 4}.
Whence, we have four functions satisfying the statement of the problem. ■

3. We shall prove a more general statement. That is, we prove that xn is

a-periodic modulo m, for each m. In doing so, we need the following lemmas.
Lemma 1. Let m ≥ 3 be an integer which is not a power of 2. Let x1 be
a fixed positive integer and yn be a sequence of positive integers such that
xn+1 = xynn + 1, n = 1, . . . If the sequence xn is eventually periodic mod
m then there are positive integers b, d, T where 2 ≤ d ≤ m − 1 such that
yb+kT , k = 0, 1, . . . is periodic mod d.

Proof. Since m is not a power of two it would have an odd prime divisor,
say p. Since the sequence xn (mod m) is eventually periodic it follows that
xn (mod p) is also eventually periodic. Hence, for all large n, xn+T − xn is
divisible by p. Fix b, large enough such that a ≡ xb (mod p) and a ̸= 0, 1.
Since p ≥ 3, such a b exists because each 0 (mod p) followed by 1 which is
followed by 2. Clearly xb+kT ≡ a (mod p) for each a ≥ 0.
Notice that p divides ayb+kT − ayb . Hence, |yb+kT − yb | is divisible by the
order of a mod p, namely d. That is, yb+kT (mod d) is periodic. As desired.
Now, we shall prove a nice fact about the periodicity of S(b + kT ), k =
0, . . .
Lemma 2. Let b, T, d be positive integers, if the sequence S(b + kT ), k =
0, . . . is constant mod d then d divides T and d ∈ {1, 3, 9}.
Proof. Letting k = 10r s for some large enough r it follows that S(b+10r sT ) =
S(b) + S(sT ) ≡ S(b) (mod d). Whence, S(sT ) ≡ 0 (mod d). We can also
assume gcd(10, T ) = 1 indeed, if T = 2α 5β T1 , gcd(T1 , 10) = 1 then, by a
suitable choice of s we can make power of 10.
Then, there are s0 , s1 such that

T s0 ≡ 1 (mod 100)

34
and
T s1 ≡ 9 (mod 10r+1 ).
Further, by a suitable choice of r we can ensure that T s0 < 10r . It follows
that
S((s0 + s1 )T ) = S(s0 T ) + S(s1 T ) − 9.
Since d divides S(s0 T ), S(s1 T ), S((s0 + s1 )T ) it follows that d divides 9. That
is, S((s0 + s1 )T ). And since S(sT ) ≡ sT (mod 9) it follows that sT ≡ 0
(mod d), Hence, d divides T. This completes our proof.

Back to our problem. We shall firstly prove that m has no odd prime
divisor. Assume the sequence is periodic with the minimal period of T. Since
there are infinitely many i such that S(i) is divisible by p − 1 if p divides xi
then xi+2 ≡ 2 (mod p). If p doesn’t divide xi then xi+1 ≡ 2 (mod p). Hence,
2 is among the residues of xn modulo p. Take a = 2 in the first lemma. It
follows that the sequence S(kT + i) is constant modulo the order of 2 modulo
p, namely d. Hence, d divides 9. That is, p divides 29 −1 = 7×73. If p = 7 then
for i ≡ 0 (mod 3) we have xi+1 ≡ 2 (mod 7) and xi+2 ≡ 3 (mod 7) since
the order of 3 mod 7 is 6 we reached to a contradiction. The same holds for the
case i ≡ 1 (mod 3). Finally, for i ≡ 2 (mod 3) we have xi+1 ≡ 5 (mod 7)
and since the order of 5 modulo 7 is 6 we again reached to a contradiction.
If p = 73 the numbers with orders that dividing 9 are

1, 2, 4, 8, 16, 32, 64, 55, 37,

surprisingly powers of 2. If i is divisible by 9 then xi+1 ≡ 2 (mod 73) and

xi+2 ≡ 3 (mod 73) while the order of 3 modulo 73 doesn’t divide 9. If i is
not divisible by 9 then xi+1 is not congruent to a power of 2 mod 73. Hence,
its order modulo 73 doesn’t divide 9.
Finally, for powers of 2 we prove that the only solutions are m = 1, 2, 4.
Let m = 2c we claim that the period is of the form 1, 2, 1, 2, . . . Indeed if xi
is odd then since the order of xi modulo 2c is a power of two and for sake of
periodicity it must divide 9 we would obtain that xi ≡ 1 (mod 2c ). Hence,
xi+1 ≡ 2 (mod 2c ) and by the same argument xi+2k ≡ 2S(i−1+2k) + 1 ≡ 1
(mod 2c ) yielding S(i − 1 + 2k) ≥ c for all k. Thus, c ≤ min S(i − 1 + 2k) = 2.
Whence, m = 1, 2, 4. ■
Comment. Here are the original proposal. The PSC changed it to a more
challenging problem. Let x1 be an integer, find all positive integers m ≥ 3
such that the sequence √
xn+1 = x⌊n
n
2⌋
+ 1,
is eventually periodic mod m.

35
Final Exam
1. The answer is no. First for every 1 ≤ i ≤ 8 we define Ni to be the set of
numbers assigned to the vertices adjacent to the vertex with number i. By
the problem’s assumption the sum of elements of Ni is divisible by i. Since
the sum of every three numbers less than 8 is at most 7 + 6 + 5 = 18, the sum
of elements of N8 is either 8 or 16. By checking different cases it is easy to
see N8 equals to either {3, 6, 7}, {4, 5, 7}, {5, 2, 1} or {4, 3, 1}. We will check
these different cases.
• If N8 = {3, 6, 7}, let N6 = {x, y, 8}. Evidently N8 ∩ N6 = ∅. Now, by
assumption, 6 | x + y + 8, and 11 = 1 + 2 + 8 ≤ x + y + 8 ≤ 17 = 4 + 5 + 8.
These imply x + y = 4 and so {x, y} = {1, 3}. This contradicts to
N8 ∩ N6 = ∅.
• The case N8 = {4, 5, 7} is similar to the previous case. We only need to
consider N7 instead of N6 .
• If N8 = {1, 3, 4}, denote by x, y, z, t the numbers of other vertices as
is the figure below. We have {x, y, z, t} = {2, 5, 6, 7}. It follows from
t | x + y + z that t is a divisor of t + x + y + z = 20 so t = 5. On the other
hand, y | 1 + 3 + t. Therefore y must be either 1, 3 which is impossible.

• If N8 = {5, 2, 1}, denote the numbers of other vertices by x, y, z, t as

is the figure below. In this case {x, y, z, t} = {3, 4, 6, 7}. By a similar
argument to the previous case, we have t = 4. On the other hand,
x | 1 + 5 + t, so x is a divisor of 10, again impossible.

36
Therefore in each case the numbers can not be assigned to the vertices of the
cube and the solution is complete. ■

2. Denote by Γ the circumcircle of the triangle ABC. Letting F and G be the

reflections of H with respect to AC and AB, respectively. It is well-known
that F and G lie on Γ. Also EF is parallel to AC since A is the midpoint
of the segment EH. Therefore ∠EF B = 90◦ and the pentagon EF DBX is
cyclic. Now notice that

∠XF B = ∠XDB = ∠ACB = ∠AF B,

Y
E

A F

X
G

B C
D

Yielding that the line F X passes through A. Then, since AF = AH = AE

we should have AD = AX. Similarly one can show that G, A, and Y are
collinear and AD = AY . So AX = AY and AF = AG. This yields that
XY ∥ F G and

∠XAG = ∠AY X + ∠AXY = ∠AY X + ∠AF G.

Hence the result follows. ■

37
3. Let P (x) is a non-zero polynomial and d = deg P ,We shall prove that
f (P (x)) = Cd, for some constant C, furthermore f (0) = 0. It is easy to check
that such a function properly works.
Letting (P (x), Q(x)) = (P (x), 1), (b, P (x)) in the second equation as well
as (P (x), Q(x)) = (a, 1) in the first equation for some non-zero constants a, b
we conclude that f (a) = 0 and f (bP (x)) = f (P (x)). Letting (P (x), Q(x)) =
(bx, xb + 1) to obtain f (x + b) = f (x + 1) for all b ̸= 0. Let c, d be two arbitrary
real numbers such that cd ̸= 0 it follows that
    
d d
f x+ =f c x+ = f (cx + d) = f (x + 1).
c c

Finally, letting (P (x), Q(x)) = cx, x − 1c for some c ̸= 0, 1 in the first

equation yields
 
1
f (x) = f (cx) = f cx + 1 − = f (x + 1).
c

Hence, if P (x) is a linear polynomial then f (P (x)) = f (x + 1), i.e., a constant

function, say C. We then finish our proof based on the induction on the degree
of polynomial. The base is true for d = 0, 1. Assume that the statement holds
true for all polynomials of degree less than d. Choose a non-zero rational
number r such that P (r) ̸= 0. Letting R(x) = P (r) r
P (x + r) it follows that
R(0) = r. Hence, R(x) = r + xS(x), for some polynomial S(x) of degree d − 1.
Putting (P (x), Q(x)) = (R(x), x − r) in the first equation yielding

f (R(x) − r) = f (xS(x)) = f (x) + f (S(x)) = f (R(x − r))

 
r
=f P (x) = f (P (x)).
P (r)

According to the induction hypothesis f (S(x)) = C(d − 1) and f (x) = C.

Hence,
f (P (x)) = Cd.
As desired. ■

4. We claim that Arash has a winning strategy if and only if k ≤ 1400×1401

3 =
1400 × 467.
First of all, it is obvious if k > 1400 × 467, then Arash can not color k
trominos in his first turn and he will lose. Indeed, there are not k disjoint
trominos in the table.
Next, we shall prove that for every k ≤ 1400 × 467, Arash has a winning
strategy. We can partition the table into 700 × 467 number of 2 × 3 rectangles
(intersection of two consecutive columns and three consecutive rows). We
count the rectangles in such a way that the rectangles filling the left columns

38
have numbers 1, 2, . . . , 467 (from top to bottom), then the rectangles filling
the third and forth column have numbers 468, 469, . . . , 934, etc. We consider
two different trominos in each 2 × 3 table, as in the figure below.

Now if k ≥ 700 × 467, Arash will color the trominos of type 1 in all the
rectangles and then k ′ = k − 700 × 467 trominos of type 2 in rectangles of
numbers 1, 2, . . . , k ′ . It is easy to find that after Arash’s turn, every 2 × 2
square has a colored cell. So, Babak can not color any 2 × 2 square in his
turn and Arash will win.
If k ≤ 700 × 467, Arash in his first turn colors the trominos of type 1 in
rectangles 1, 2, . . . , k. For the next turns, suppose that Babak has filled some
square and now this is Arash’s turn. If he can find k disjoint trominos do
not intersecting rectangles 1, 2, . . . , k, he will color them. In the case that
there is not k disjoint trominos, let k ′ be the maximal number such trominos
(k ′ < k). Now, he will color k ′ trominos without intersection with rectangles
1, 2, . . . , k, and also color k − k ′ trominos in rectangles 1, 2, . . . , k − k ′ . After
this step, there would be no uncolored 2 × 2 square left in the table and so
Babak will lose the game. ■

39
Team Selection Test

1. Suppose that the sets are A1 , A2 , . . . , A100 . We claim that Mahdi needs
100 steps. First of all, notice that 100 would be enough; by calling (1, 2),
(1, 3) and (2, 3) he can understand A1 , A2 and A3 . Because we have

A1 = ((A1 ∪ A2 ) − (A2 ∪ A3 )) ∪ (A1 ∩ A2 ) ∪ (A1 ∩ A3 ) .

Then by calling (i, 3) for 4 ≤ i ≤ 100 he can determine all sets.

To prove that 100 is necessary, consider a graph with 100 vertices, and
at each step connect vertices i and j if Mahdi calls i and j. After 99 steps
the graph has 99 edges so one of the connected components should be a tree.
Hang the tree on a vertex and consider at each step Ai ∪Aj = {1}, Ai ∩Aj = ∅
then Mahdi can’t distinguish between the case that all the sets at odd levels
of the tree are empty and the sets on even levels are {1} and the case where
all the sets at odd levels are {1} and the sets at even levels are empty. ■

2. Let
a = pα1 α2 α3 αn
1 p2 p3 . . . pn and b = q1β1 q2β2 q3β3 . . . qm
βm

We have τ (a) = (α1 + 1) . . . (αn + 1) and τ (b) = (β1 + 1) . . . (βm + 1) so

it suffices to prove that for each 1 ≤ i ≤ n there is an integer j such that
αi = βj .
Assume the contrary, suppose that there is some integer i such that αi ̸∈
{β1 , β2 , . . . , βm }. It is well known that the sequence (ak − 1)n∈N has infinitely
many prime divisors. Consider a large prime s, such that s > max(αi ), s >
max(βj ) and
pkαi +1 − 1
s| i .
pi − 1
We can also assume that gcd(k, s) = 1 by taking k = ordpi (s). Now by
Chinese remainder theorem for each t, there is an integer z such that z ≡ 1
(mod s − 1), zkαi ≡ −1 (mod sM ) for some positive integer M .
By the lifting the exponent lemma, sM +1 | pzkα
i
i +1
− 1. But because
αi ̸= βj we have zkβj ̸≡ −1 (mod s ). Hence if s | q kβi +1 − 1 we have
M

40
    
kβ +1 zkβ +1
νs qj j − 1 = νs qj j − 1 and if s ∤ q kβi +1 − 1 then by Fermat’s
little theorem s ∤ q zkβi +1 − 1 and we again have
   
zkβ +1 zkβ +1
νs qj j − 1 = νs qj j − 1 .

In concise

νs (σ (bz k)) = νs σ bk , νs σ azk > M + 1,

so if we consider M > νs σ bk we would obtain νs σ azk > νs σ bzk ,

a contradiction. ■

3. Let I be the incenter and Ia the A-excenter of triangle ABC and J the
C-excenter of triangle CDE. J lies on ω, since
1
∠EJD = 90◦ − ∠ECD = ∠CDE.
2
We claim that A, J, and Z are collinear. Point T lies on AC such that
IT ∥ DE. To prove our claim, we just need to show that IT IJ
= IIaa C
Z
, since
IT ∥ CIa , IJ ∥ Ia Z and IIa passes through A. Notice that
IE Ia Y
∠IT E = ∠DEC = ∠EDC = ∠Ia CY =⇒ △IET ∼ △Ia Y C =⇒ = ,
IT Ia C

hence IT
IJ
= IIaa C
Z
, since IE = IJ and Ia Y = Ia Z. So the claim is proved. Let
BC touches A-excircle at Y . The extensions of sides of quadrilateral AEDB
are tangent to A-excircle, therefore

AE + ED = AB + BD =⇒ ED = 2BD. (1)

Yielding DX = DE, and it is obvious that DY = DZ so quadrilateral

Y ZXE is an isosceles trapezoid and Y lies on the circumcircle of triangle
XZE. Hence
1
∠EKY = ∠EZY = ∠EDY = ∠EDM,
2
where M is the midpoint of arc ED (not containing J), so KY passes through
M . Let N be the midpoint of segment ED. Notice that △CDI ∼ △CN D,
hence
DN ID (1) BD IJ
= =⇒ = =⇒ JB ∥ ID.
CD IC CD IC
Therefore ∠JBC = 90◦ . From the other hand we have ∠JKY = ∠JKM =
90◦ , it follows that quadrilateral Y KJB is cyclic. Notice that

∠BJD = 90◦ − ∠JDB = ∠M JD = ∠Y KD,

41
therefore ∠Y JD = ∠BKD, since ∠BJY = ∠BKY . According to our as-
sumption, J lies on the angle bisector of ∠Y DZ and Y D = DZ. It yields
that JD is the perpendicular bisector of Y Z and ∠Y JD = ∠ZJD. Hence
∠ZJD = ∠BKD and the result follows.
A

T
E
K
J
I N M

X C
B D Y

Ia

4. Let E ′ be the second intersection of N P and ω1 . Since △BP C ∼ △AP D

we have ∠M P D = ∠N P C = ∠AP E ′ . So AP DE ′ is a harmonic quadrilat-
eral. Then
∠E ′ M A = ∠P M A = ∠P N B =⇒ ∠E ′ M O = ∠E ′ M A + 90◦
= ∠P N B + 90◦ = 180◦ − ∠P N O,

hence E ′ N OM is cyclic and E ′ ≡ E. Similarly, we can show that M , P and

F are collinear. Finally

∠EN O = 90◦ − ∠P N B = 90◦ − ∠P M A = ∠F M O =⇒ OE = OF.

42
 A B
ω3
E′ ≡ E
P
F′ ≡ F
M
ω1 N
O ω2
D

5. The condition of the problem can be rewritten as an+1 + an−1 = −⌊Can ⌋.

If C ≤ −2, setting a1 = 1, a2 = 2 to obtain −⌊Can ⌋ ≥ 2an−1 , and by
induction on n we can show that this sequence is increasing. In particular it
is bounded from below and is a-periodic.
Assume now C > −2. We shall prove a more general statement. That is,
when the statement holds true for all integer n.
Lemma 1. If an integer m appears infinitely often as a member of the se-
quence, then an is periodic.

Proof. Let Jm denotes the infinite set of indices j such that aj = m. Then,
−Cm ≤ aj−1 + aj+1 < −Cm + 1 holds for all j ∈ Jm . Thus, −aj−1 − Cm ≤
aj+1 < −aj−1 − Cm + 1. Assume that the sub-sequence aj−1 is unbounded
from above for j ∈ Jm .
Since an > −K for some constant K and all n, then there is j ∈ Jm such
that aj−1 ≥ K − Cm + 1. Then, aj+1 < −K this is impossible. Therefore,
aj−1 is bounded from above for j ∈ Jm . Thus, there is an integer r and
j1 , j2 ∈ Jm , and j1 ̸= j2 such that aj1 −1 = aj2 −1 = r. Thus, (aj1 −1 , aj1 ) =
(aj2 −1 , aj2 ) = (r, m). Hence the sequence is periodic.
Now, we prove the following lemma.
Lemma 2. If the sequence is unbounded from above then it has infinitely
positive and infinitely many negative terms.

43
Proof. A collection of positive (negative) terms of the sequence is called a
positive (negative) run. We shall prove that a run is always finite. This
implies the statement immediately. If C ≥ 0. then the length of the run is at
most two. If −1 ≤ C < 0, we would obtain

0 ≤ an−2 + (1 + C)an−1 + (2 + C)an+1 + an+2 < 3.

Hence, the length of the run is at most four.

If −2 < C < 1. Consider first a positive run. We prove that it has a local
maximum. Indeed, assume that the run starting with a0 is increasing. Let l
be so large that (l − 2)(2 + C) + 1 + C ≥ 0.
We may assume that a0 ≥ l. If ak = ak+1 for some k and ak+l+1 =
ak+l+2 for some positive integer l then the sequence is periodic with length
2l + 2. That is, we can assume that k = 1. By induction on |n| we can show
that an+2 = a1−n for all n. Thus, if ak+l+1 = ak+l+2 then (a−l , a−l+1 ) =
(a2+l , a3+l ). So, two consecutive terms of the sequence can be identical only
once. Thus, by omitting from consideration at most l terms, we can assume
that a0 ≥ l. Summing up l consecutive inequalities to obtain
l−1
X
l > a0 + (1 + C)a1 + (2 + C) aj + (1 + C)al + al+1
j=2

≥ a0 + (1 + C)a1 + (2 + C)(l − 2)a2 ≥ a0 ≥ l.

Which is a contradiction. Thus an increasing part of run has length at most

2l + 1 and this proves our claim. Now, let a0 be a local maximum of our run.
Only a1 , a−1 but not both can be equal to a0 , thus we assume that a0 > a1 .
Assume that ak > ak+1 for some k ≥ 0. Hence,
ak ak+2 1
0≤ +C + < ,
ak+1 ak+1 ak+1
yielding
ak+2 ak − 1
<− − C ≤ −1 − C < 1,
ak+1 ak+1
i.e., ak+1 > ak+2 . Thus, from the maximum on the run is strictly decreas-
ing. As the definition of an is symmetrical this holds for both directions.
Summing up the length of a positive run is at most 4l + 3. Negative run can
be treated similarly.
Now, if the sequence is bounded then we have nothing to do. Assume that it
is one-side bounded, but unbounded. Then there are infinitely many positive
and infinitely many negative terms in the sequence. Since the sequence is
one-side bounded there is an integer which appears infinitely many often.
Thus, the sequence is bounded and hence periodic.
■

44
6. We claim that the second player has winning strategy for (m, n) = (pα , pβ )
where p is a prime number and α, β ∈ Z≥0 . We call a pair (m, n) a good pair
if you the second player has winning strategy.
Lemma 1. If (m, n) is a good pair and d | m then (d, n) is also a good pair.
Proof. Assume that the games starts with the sequence a1 , a2 , ..., ad . Moham-
mad Considers a second game that starts with the sequence (ã1 , ã2 , ..., ãm )
where ã mi
d
= ai and a˜j = 0, for m
d ∤ j. Now if Ali chooses s in the first game,
Mohammad assumes that Ali chooses m d s in the second game and follows his
strategy and wins the second game. But the first game is just a restriction of
the second game to the components m d i so he can also win the first game.

Lemma 2. (m, nk) is a good pair if and only if (m, n), (m, k) are also good
pairs.
Proof. If (m, nk) is a good pair then it is clear that (m, n), (m, k) are good
pairs. Now if (m, n), (m, k) are good pairs, starting with a sequence a1 , ..., am
then using the afore-mentioned strategy for (m, n), making all the components
divisible by n. Considering the sequence ( ani ), applying the preceding strategy
for (m, k) to ensure that k| ani , by the very end all the terms are divisible by
nk and hence nk is a good pair.
By these lemmas we see that if (m, n) is a good pair and p|m, q|n then
(p, q) is also a good pair. Now by the next lemma we would obtain p = q.
Hence (m, n) should be of the form (pα , pβ ) .
Lemma 3. If (m, n) is a good pair and m, n > 1 then gcd(m, n) > 1.
Proof. Assume that gcd(m, n) = 1. Consider one step before the end of the
game. The sequence (ai ) must be constant; otherwise for each sequence (bi )
there is some j such that ai + bj ̸≡ 0 (mod n) and the first player can choose
j − i. In the second step before the last step, we should have

a0 − am ≡ a1 − a0 = a2 − a1 ≡ · · · ≡ am − am−1 (mod n),

otherwise Ali could choose a shift s such that (ai + bi + s) is not constant. So
we must have m(a1 − a0 ) ≡ 0 (mod n), implying that a1 − a0 ≡ 0 (mod n).
So the sequence is constant. We then use induction to deduce that sequence
should be constant from the starting point. So, for a non-constant sequence
the first player has a winning strategy.

It remains to prove that (pα , pβ ) is a good pair. By the second lemma it

suffices to prove that (pα , p) is a good pair. We claim that if second player
chooses bi = −ai at each step, he shall win. There are two ways to do this:

45
First proof. Let T : (Z/pZ)m → (Z/pZ)m denotes a shift by 1. Then at
each step the sequence changes to (T s − id)(a1 , . . . , apα ). Suffices to prove
that
Ypα
(T sj − id) ≡ 0 (mod p)
j=1
α
but T − id | T sj − id. So it is enough to prove (T − id)p ≡ 0 (mod p) but
α α
(T − id)p ≡ T p − id ≡ 0 (mod p).

Second proof.
Lemma 4. There is an integer-valued polynomial Q(x) such that Q(i) ≡ ai
(mod p) for all 1 ≤ i ≤ pα .
Proof. Consider the polynomial Pi (x) = x−i−1
pα −1 . One can see that Pi (i) ≡ 1

(mod p) and for j ̸≡ i (mod pα ), we have Pi (j) ≡ 0 (mod p). Then we can
choose
pα
X
Q(x) = ai Pi (x).
i=1

Now, in the next step we have ai ≡ Q(i) − Q(i + s) (mod p) but the degree
of Q(x) − Q(x + s) is less than deg(Q) so after pα steps the second player
wins. ■

7. First of all, we show that d should be at least n. The vertices of the

polygon are on n + 1 different vertical lines and between any two such lines
the polynomial should intersect two edges of the polygon, one above and one
below the x-axis. So by the intermediate value theorem the polynomial should
have at least n roots.
Now we want to say that n is sufficient. Choose n + 1 points on the vertical
lines that passing through vertices like below. By the Lagrange Interpolation
formula there is a polynomial of degree at most n that cross these n+1 points.
Again By the intermediate value theorem this polynomial should intersects
all edges. So the polynomoal is of degree n.

46
 4

−2 −1 1 2 3 4 5
−1

−2

8. Let in triangle ABC, points Ia , H, and D be the A-excenter, foot of the

altitude from A, and foot of the angle bisector from A, respectively. Since
(AD, IIa ) = −1 and ∠AHD = 90◦ , we have ∠IHD = ∠DHIa . Further,
AIHQ is cyclic, So ∠QAI = ∠IHD, hence ∠QAI = ∠DHIa and △QAI ∼
△Ia HE. We know △AHE ∼ △F AI, therefore △QIF ∼ △Ia EA and we
have
∠F QI = ∠AIa E = ∠IQB.
Now let ω be the circle centered at M , has the radius M B = M C, and X and
Y be the intersections of line F Q with AB and AC, respectively. Point I lies
on the angle bisector of ∠QY C, since I lies on the angle bisector of ∠Y QB
and ∠Y CQ. Therefore Y , I, and Oc are collinear, where Oc is the center of
ωc . Now we have
1 1
∠IOc C = 90◦ − ∠QY C = (∠Y QC + ∠Y CQ) = ∠IQB + ∠ICB
2 2
1 1 1
= (∠B − ∠C) + ∠C = ∠B
2 2 2
Therefore Oc lies on ω. Similarly we can show that Ob lies on ω, where Ob is
the center of ωb . Notice that I is the A-excenter of triangle AXY so
1
∠Ob IOc = ∠Y IX = 90 − ∠A = ∠BIa C,
2
hence BOb = COc . Finally

PωMb = M Ob2 − Ob B 2 = M Oc2 − Oc C 2 = PωMc ,

we are done.

47
 F

X
I

B D C
Q
H E

OB
M

OC Ia

9. For n = 6 there are 16 combinatorially different point sets (order types).

It is easy to check that all these 16 cases allow for 3 edge-disjoint plane span-
ning trees.

Let S be a set of n points in the plane and let e = (r, b) be an edge spanned
by S having exactly 4 points of S on one side (i.e., on one side of the straight
line ℓe supporting e). Let S ′ be the set of 6 points containing r, b, and the
exactly 4 points on one side of e. By the above claim, S ′ contains 3 edge-
disjoint plane spanning trees. For sake of simplicity, we call them red, blue,
and green. Without loss of generality, assume that e is a part of the red tree.
Note that each point of S ′ is incident to all three trees, and that r and b
are extremal points for S\(S ′ \{r, b}). We construct a red and a blue plane
spanning trees by connecting r and b, respectively, with all points in S\S ′ .

48
Next, we shall construct the third (green) plane spanning tree on S. Note
that the green plane spanning tree on S ′ can be completed by a triangulation
T . Let q be a point of S ′ \{r, b}, such that qrb is a triangle in T . Observe
that any edge incident to q and crossing e does not cross a green edge.
Assume that there exists a point q ′ ∈ (S\S ′ ) such that the edge qq ′ crosses
e. Then we connect q and q ′ with a green edge and complete the green plane
spanning tree by connecting all points in S\(S ′ ∪ {q ′ }) with q ′ . If such a point
q ′ does not exist, then there must be an edge e′ of the convex hull of S, such
that e′ crosses ℓe . Denote by p the endpoint of e′ in S\S ′ . We shall color e
by green and complete the green plane spanning tree by connecting all points
in S\(S ′ ∪ {p}) with p. ■

10. Let L(z) be the smallest prime divisor of z.

Lemma. If a, b, c ∈ N such that a ̸= b we have
 c
a − bc

L ≥ L(abc)
a−b
Proof. Assume the contrary, then there should be an integer p such that
c
−bc
gcd(p, a) = gcd(p, b) = 1 and p | aa−b . Now consider two cases, if a ≡ b
(mod p) then by lifting the exponent lemma νp (ac − bc ) = νp (a − b) which is
a contradiction. If a ̸≡ b (mod p) then ordp (ab−1 ) ̸= 1 (b−1 is multiplicative
inverse of b module p). But we know that ordp (ab−1 ) | c and we have p >
ordp (ab−1 ) > L(c) which is a contradiction.
Consider a set
Ak = {x ∈ N | L(x) ≥ k}
c c
now if a, b, c ∈ Ak and d | aa−b
−b
then by lemma we have L(d) < L(abc) < n
and hence d ∈ Ak . Now it is enough to take k > n. ■

11. We need at least ⌊ n2 ⌋ − 1 blocks. To see this number is sufficient, we need

to provide an example for ⌊ n2 ⌋ − 1 blocks for every n. The idea is to put
blocks on some cells from the first two and the last two columns such that
the number of their rows have specific remainders modulo n. For instance, if
n ≡ 0 (mod 8), we can put blocks in the cells of form (i, j) for

if j = 1


 i ≡ 5 (mod 8),
i ≡ 8 (mod 8), i ≥ 1, if j = 2




 i ≡ 0 (mod 8), if j = 2n − 1
i ≡ 4 (mod 8), if j = 2n.


We need a small modifications in this pattern for other cases modulo 8.

Next, we shall prove that ⌊ n2 ⌋ − 1 blocks are necessary.

49
Lemma 1. Assume that the starting point of the move is not in column j
(1 < j < n), and there is no block in columns j − 1, j, j + 1. Then we do not
ever change direction in column j. (The same is true for rows.)

Proof. Consider the direction firstly changes at cell (i, j) in the column j.
Since there is no block in this column, before reaching this row we have been
moving vertically, so there should be some block in (i, j + 1) or (i, j − 1). A
contradiction.
Lemma 2. If we reach the cell (i, j) then the starting cell or a cell where we
change direction should either be in the row i or in the column j.
Proof. Obvious!
If there are three columns j − 1, j, j + 1 that do not contain the starting
cell or any blocks, then by the first lemma, we cannot change our direction
at any cell in column j. By the second lemma, to reach all the cells in the
column j we need to change direction at all rows except the rows containing
the starting point. Assume that the staring point is (i0 , j0 ). Then by the first
lemma, because we can change direction at rows i ̸= i0 , 1, n there should be
a block in one of the (i − 1)-th or (i + 1)-th rows. Each block could work for
at most two rows so we need at least ⌈ n−3 n
2 ⌉ = [ 2 ] − 1.
If there are not three such columns, then we need at least [2n/3] blocks.
So in each case we need at least [ n2 ] − 1 blocks. ■

12. We shall firstly prove following lemmas;

Lemma 1. For any real number r, the length of the interval f (r) is at most
2.
Proof. If x, y ∈ f (r) we have r ∈ f (x) ∩ f (y) hence we have |x − y| ≤ 2.

Lemma 2. If the interval I has length 2, then there is a real number r such
that f (r) = I.
Proof. If I = [x, x + 2] then by condition 2 there is r ∈ f (x) ∩ f (x + 2) so
we have x, x + 2 ∈ f (r). But the length of f (r) is at most 2 hence we have
f (r) = [x, x + 2], this completes our proof.

Lemma 3. f is an injective function and for any r the length of f (r) is 2.

Proof. Using preceding lemmas it is enough to show that if f (x) ⊂ f (y)
then x = y. Assume the contrary, let I be an interval of length 2 such that
x ∈ I, y ̸∈ I then by lemma 2 there exists r such that I = f (r). Now we have
r ∈ f (x), r ̸∈ f (y) which contradicts f (x) ⊂ f (y).

50
 Define a new function g : R → R such that for all r, g(r) is the midpoint
of f (r). Since we know the length of f (r) we can forget about f and work
with g. We apt to find all bijective functions g such that:
i. |x − g(y)| ≤ 1 ⇐⇒ |y − g(x)| ≤ 1.
ii. |g(x) − g(y)| ≤ 2 ⇐⇒ |x − y| ≤ 2.
iii. g(r) = r2 for all 0 ≤ r ≤ 1.
Lemma 4. g is strictly increasing.
Proof. It suffices to prove if x < y, y − x < 1 then g(x) < g(y). We firstly
consider the case y = 0. For each −2 ≤ x ≤ 0 by condition 2 and g(0) = 0
we have
|g(x) − g(2)| > 2, |g(x)| ≤ 2, |g(2)| ≤ 2
so we can deduce that g(x) and g(2) have different signs, similarly we can
deduce that g(2), g(1) = 1 have the same sign (they are both opposite to the
sign of g(−2)). Hence we have g(2) > 0, g(x) < 0.
We prove by induction on |⌊x⌋| (We only prove the case that 0 ≤ x < y
the other case is similar). For the base if 0 ≤ x < y < 1 then x2 < y 2 . We
know that y − x − 2 > 2 hence by the second condition we have
|g(y) − g(x − 2)| > 2, |g(y) − g(x)| ≤ 2, |g(x) − g(x − 2)| ≤ 2.
We know by induction that g(x) − g(x − 2) > 0 hence we have g(y) > g(x).
Lemma 5. we have g(x + 1) = g −1 (x) + 1.
Proof. We know that g is bijective so it suffices to prove that g(g(x) + 1) =
x + 1. Assume that g(r) = x + 1 then we have r ≤ g(x) + 1. Now if r′ > r we
have g(r′ ) > x + 1 hence by condition 1; r′ > g(x) + 1 so we get r = g(x) + 1.
By the last lemma we obtain that there is only one function that satisfies the
condition of the problem and that function is:
(
(x − ⌊x⌋)2 + ⌊x⌋ ⌊x⌋ ≡ 0 (mod 2);
g(x) = p
x − ⌊x⌋ + ⌊x⌋ ⌊x⌋ ≡ 1 (mod 2).
It is easy to check that this function actually works. ■
Remark. If we replace the condition 3 in the problem with the weaker con-
dition f (0) = [−1, 1], f (1) = [0, 2] we can still finds all the solutions. By the
above proof it is clear that any such f would be of the form [g(r) − 1, g(r) + 1]
where g was constructed by choosing a monotonically increasing yet bijective
function h : [0, 1] → [0, 1] such that:
(
h(x − ⌊x⌋) + ⌊x⌋ ⌊x⌋ ≡ 0 (mod 2);
g(x) = −1
h (x − ⌊x⌋) + ⌊x⌋ ⌊x⌋ ≡ 1 (mod 2).

51