IMO Winter Camp 2010 Warm-Up Problems 1

IMO Winter Camp 2010 Warm Up Problems

Algebra

A1 Let x, y, z ∈ R be non-negative real numbers such that 0 ≤ x, y, z ≤ 1. Find the maximum possible value
of
x + y + z − xy − yz − zx.
Determine all triples (x, y, z) for which this maximum is attained.
A2 For any positive integer k, prove that
2k p
X 1
2k 2 + k < k 2 + j < 2k 2 + k + .
j=1
2

A3 Let a, b, c be positive real numbers such that a2 + b2 + c2 = 3. Prove that

1 1 1 3
+ + ≥ .
1 + ab 1 + bc 1 + ca 2

A4 P (x) is a polynomial of odd degree with real coefficients. Show that the equation P (P (x)) = 0 has at
least as many distinct real roots, as the equation P (x) = 0.
A5 Find all functions f : R 7→ R such that ∀x, y, z ∈ R, we have that if x3 + f (y) · x + f (z) = 0, then
f (x)3 + y · f (x) + z = 0.

Number Theory

N1 Let n be a positive integer. The sum of the positive divisors of n is s. Prove that the sum of the reciprocal
of these divisors is ns .

N2 A rational number x is written on a blackboard. In each step, you erase x and replace it with either x + 1
or − x1 . (If x = 0, you must choose x + 1). Prove that for any rational number p, if p currently appears on
the blackboard, then you can make 0 appear after a finite number of steps.
N3 Find all triples of positive integers (a, b, c) such that a2 + 2b+1 = 3c .

N4 Find all positive integers that can be written in the form

m 2 + n2 + 1
,
mn
for all positive integers m, n.

N5 Find all prime numbers p such that the following statement is true: there are exactly p ordered pairs of
integers (x, y) such that 0 ≤ x, y < p and y 2 ≡ x3 − x mod p.

1
IMO Winter Camp 2010 Warm-Up Problems 2

Geometry

G1 A circle C and a point P are given on the same plane. Given any point Q on the circumference of C, let
M be the midpoint of P Q. Find the locus of point M , i.e. find all possible locations for point M .

G2 Let A be a point outside of a circle C. Two lines pass through A, one intersecting C at B, C, with B closer
to A than C, and the other intersecting C at D, E, with D closer to A than E. The line passing through
D parallel to AC intersects C a second time at F and the line AF intersects C a second time at G. Let
M = EG ∩ AC. Prove that
1 1 1
= + .
|AM | |AB| |AC|

G3 Let ABC be a triangle and D be a point on side BC. The internal angle bisector of ∠ADB and that of
∠ACB intersect at P . The internal angle bisector of ∠ADC and that of ∠ABC intersect at Q. Let M
be the midpoint of P Q. Prove that |M A| < |M D|.
G4 A convex quadrilateral ABCD has |AD| = |CD| and ∠DAB = ∠ABC < 90◦ . The line through D and
the midpoint of BC intersects line AB at point E. Prove that ∠BEC = ∠DAC.
G5 Let ABC be a triangle with |AB| > |AC|. Let its incircle touch side BC at E. Let AE intersect this
incircle again at D. Let F be the second point on AE such that |CE| = |CF |. Let CF intersect BD at
G. Prove that |CF | = |F G|.

Combinatorics

C1 Find the number of subsets of {1, 2, · · · , 10} that contain its own size. For example, the set (1, 3, 6) has 3
elements and contains 3.
C2 A sequence of non-negative integers is defined by G0 = 0, G1 = 0 and Gn = Gn−1 + Gn−2 + 1 for every
n ≥ 2. Prove that for every positive integer m, there exists a positive integer a such that Ga , Ga+1 are
both divisible by m.
C3 A set S of ≥ 3 points in a plane has the property that no three points are collinear, and if A, B, C are
three distinct points in S, then the circumcentre of ∆ABC is also in S. Prove that S is infinite.
C4 Let n, k be positive even integers. A survey was done on n people where on each of k days, each person
was asked whether he/she was happy on that day and answered either ”yes” or ”no”. It turned out that
on any two distinct days, exactly half of the people gave different answers on the two days. Prove that
there were at most n − nk people who answered ”yes” the same number of times he/she answered ”no”
over the k days.
C5 There are n ≥ 5 people in a room, where each pair is classified as friends or strangers. No three people are
mutually friends. There also exist an odd number of people P1 , · · · , Pm such that Pi is friends with Pi+1
for all i ∈ {1, · · · , m}, where the indices are taken modulo m. Prove that there exists one person who is
friends with at most 2n/5 people.

2
IMO Winter Camp 2010 Warm-Up Problems 3

IMO Winter Camp 2010 Warmup Solutions

Algebra

A1 Let x, y, z ∈ R be non-negative real numbers such that 0 ≤ x, y, z ≤ 1. Find the maximum possible value
of
x + y + z − xy − yz − zx.
Determine all triples (x, y, z) for which this maximum is attained.

Solution: The maximum possible value is 1 and is attained when (x, y, z) = (1, t, 0) for any 0 ≤ t ≤ 1, or
any of its permutation solutions.

Note that
x + y + z − xy − yz − zx = −(1 − x)(1 − y)(1 − z) + 1 − xyz ≤ 1.
Equality holds when at least one of x, y, z is equal to 1 and at least one of x, y, z is equal to 0. Clearly,
this is possible. Therefore, the maximum possible value of x + y + z − xy − yz − zx is 1 and is attained
when (x, y, z) = (1, t, 0) for any 0 ≤ t ≤ 1, and any of its permutation solutions. 

Source: Diamonds by Tran Phuong

Comments: When the expressions x + y + z, xy + yz + zx, xyz appear, you should always consider poly-
nomials of degree three and/or terms of the form (c − x)(c − y)(c − z) for some constant c. This may even
help you factor terms in other similar problems.

A2 For any positive integer k, prove that

2k p
X 1
2k 2 + k < k 2 + j < 2k 2 + k + .
j=1
2

Solutions: By subtracting 2k 2 from all three terms, it suffices to show that

2k p
X 1
k< ( k 2 + j − k) < k + .
j=1
2

Note that
p j
k2 + j − k = p ,
k2 + j + k
and
j j j
<p < .
2k + 1 k2 + j + k 2k
Therefore,
2k 2k 2k
X j X j X j
< p < .
j=1
2k + 1 j=1 k 2 + j + k j=1
2k

3
IMO Winter Camp 2010 Warm-Up Problems 4

Hence,
2k
2k(2k + 1) X j 2k(2k + 1)
< p < .
2(2k + 1) j=1
2
k +j+k 2(2k)
j
p
Since √ = k 2 + j − k, we conclude that
k2 +j+k

2k p
X  1
k< k2 + j − k < k + .
j=1
2

This solves the problem. 

Source: Mongolian Mathematical Olympiad 2009

Comments: The appearance of the square root suggests that p you rationalize the pnumerator of some
you to consider the term k 2 + j − k, since k < k 2 + j < k + 1 for
expression. It should be natural for p
j j
1 ≤ j ≤ 2k. Bounding the term j/( k 2 + j + k) between 2k+1 and 2k should also be natural since you
j
are looking for rational lower and upper bounds of √ 2 . 
k +j+k

A3 Let a, b, c be positive real numbers such that a2 + b2 + c2 = 3. Prove that

1 1 1 3
+ + ≥ .
1 + ab 1 + bc 1 + ca 2
a2 +b2 1 1
Solution 1: Note that ab ≤ 2 . Therefore, 1+ab ≥ 2 2 for all a, b ≥ 0. Therefore, by AM-HM
1+ a +b
2
inequality, we have
X 1 X 1 9 9 3
≥ a2 +b2 ≥ P a2 +b2 = 2 2 2
= ,
cyc
1 + ab cyc 1 + 2 (1 + 2 ) 3+a +b +c 2
cyc

as desired. 

Solution 2: By adding the terms on the left-hand side and cross multiplying, the inequality becomes
equivalent to
3 + (ab + bc + ca) ≥ abc(a + b + c) + 3a2 b2 c2 .
By homogenizing (i.e. using the condition a2 + b2 + c2 = 3 to make every term have the same degree), we
have
(a2 + b2 + c2 )3 (a2 + b2 + c2 )2 (ab + bc + ca) (a2 + b2 + c2 )abc(a + b + c)
3· + ≥ + 3a2 b2 c2 .
27 9 3
By clearing denominators, this inequality becomes equivalent to
(a2 + b2 + c2 )3 + (a2 + b2 + c2 )2 (ab + bc + ca) ≥ 3(a2 + b2 + c2 )abc(a + b + c) + 27a2 b2 c2 .
For any non-negative real numbers x, y, z, let
X
[x, y, z] = ax by cz .
sym

4
IMO Winter Camp 2010 Warm-Up Problems 5

We recall Muirhead’s Majorization inequality, which states that if x, y, z, u, v, w are non-negative real
numbers such that x ≥ y ≥ z, u ≥ v ≥ w, x + y + z = u + v + w, x ≥ u, x + y ≥ u + v, then
[x, y, z] ≥ [u, v, w]. [x, y, z] is said to majorize [u, v, w]. Note that
1
(a2 + b2 + c2 )3 = [6, 0, 0] + 3[4, 2, 0] + [2, 2, 2],
2
1
(a2 + b2 + c2 )2 (ab + bc + ca) = [5, 1, 0] + [4, 1, 1] + [3, 3, 0] + 2[3, 2, 1],
2
3
3(a2 + b2 + c2 )abc(a + b + c) = [4, 1, 1] + 3[3, 2, 1],
2
9
27a2 b2 c2 = [2, 2, 2].
2
Hence, after cancellation of terms, the previous inequality becomes equivalent to
1 7
[6, 0, 0] + 3[4, 2, 0] + [5, 1, 0] + [3, 3, 0] ≥ [4, 1, 1] + [3, 2, 1] + [2, 2, 2].
2 2
By Muirhead’s inequality, [5, 1, 0] ≥ [4, 1, 1], [3, 3, 0] ≥ [3, 2, 1], 21 [6, 0, 0] + 3[4, 2, 0] ≥ 21 [2, 2, 2] + 3[2, 2, 2] =
7
2 [2, 2, 2]. Hence, the inequality holds. 

Source: Belarus Mathematical Olympiad 1999

Comments: You should always be able to solve any three-variable symmetric inequality, that can be
homogenized, has equality case a = b = c and/or a = b, c = 0 and its permutation solutions, and whose
variables are non-negative real numbers, using Muirhead’s and Schur’s inequality. Muirhead’s inequality
states that [x, y, z] ≥ [u, v, w] whenever [x, y, z] majorizes [u, v, w]. An equivalent version of Schur’s in-
equality is [x + 2, 0, 0] + [x, 2, 0] ≥ 2[x + 1, 1, 1]. For proofs of these inequalities, see Chapter 3.2 of [2].

You may be thinking that solving problems in this manner is not elegant and can be considered ugly.
But the truth of the matter is that a lot of mathematics is initially done by getting your hands dirty and
working with cumbersome expressions. Muirhead’s and Schur’s inequality is in the arsenal of every top
math olympian in the world and should always be used as a last resort to solve inequalities of this type.
To test your ability to use these inequalities, do try the following problem.

Exercise: Let a, b, c be positive real numbers such that a + b + c = 3. Prove that

1 1 1
2
+ 2 + 2 ≥ a2 + b2 + c2 .
a b c

A4 P (x) is a polynomial of odd degree with real coefficients. Show that the equation P (P (x)) = 0 has at
least as many distinct real roots, as the equation P (x) = 0.

Solution: Let x1 , · · · , xn be all of the distinct roots of the equation P (x) = 0. We want to show that
P (P (x)) = 0 has at least n distinct real roots.

For each i = 1, · · · , n, consider the equation P (x) = xi . It has at least one real root ai since P is a
polynomial of odd degree. Now, for i, j ∈ {1, 2, · · · , n}, i 6= j, if ai = aj , then P (ai ) = P (aj ). Therefore,

5
IMO Winter Camp 2010 Warm-Up Problems 6

xi = xj , which is impossible since x1 , · · · , xn are pairwise distinct. Therefore, the ai ’s are also pairwise
distinct. But each ai is a solution to P (P (x)) = 0, since P (P (ai )) = P (xi ) = 0. Hence the equation
P (P (x)) = 0 has at least n distinct real roots a1 , a2 , ..., an . The result follows. 

Source: Russian Mathematical Olympiad 2000

Comments: When polynomials in question have certain properties such as having odd degree, a question
you should ask yourself is what are the differences between polynomials of odd degree and those of even
degree? The answer will often give you clues as to how to approach a problem. If you have not used the
fact that the polynomial has odd degree, you may not be approaching the problem in a correct way.

A5 Find all functions f : R 7→ R such that ∀x, y, z ∈ R, we have if x3 + f (y) · x + f (z) = 0, then
f (x)3 + y · f (x) + z = 0.

Solution: We will call x3 + f (y) · x + f (z) = 0 relation (A) and f (x)3 + y · f (x) + z = 0 relation (B).

We first prove that f is surjective. Let m be an arbitrary real number. We will show that f (x) = m
for some x ∈ R. Since (A) is a cubic polynomial in x, given any fixed y, z ∈ R, there exists x ∈ R such
that (x, y, z) satisfies (A). We choose y = 1, z = −m3 − m. We can choose x ∈ R such that (x, y, z)
satisfies (A). Therefore, (x, y, z) satisfies (B), i.e. f (x)3 + f (x) − (m3 + m) = 0. This implies that
(f (x) − m)(f (x)2 + m · f (x) + m2 + 1) = 0. The latter factor, as a quadratic equation in f (x), has negative
discriminant, hence cannot be zero. Therefore, f (x) = m. This proves that f is surjective.

We now show that f (x) = 0 if and only if x = 0. Since f is surjective, f (r) = 0 for some r ∈ R.
Note that (x, y, z) = (0, a, r) satisfies (A) for any choice of a ∈ R. Hence, (0, a, r) satisfies (B), i.e.
f (0)3 + af (0) + r = 0 holds for all a ∈ R. Therefore, af (0) = r − f (0)3 for all a ∈ R. The right hand side
is a constant. Therefore, f (0) = 0. This implies r = 0. Hence, f (x) = 0 if and only if x = 0.

We now show that f is injective. Assume f (a) = f (b) for some a, b ∈ R. Take any z 6= 0. Then f (z) 6= 0.
There exists x for which x3 + f (a)x + f (z) = 0. Note that x 6= 0, since f (z) 6= 0. Then (x, a, z) and
(x, b, z) satisfy (A), since f (a) = f (b) and x3 + f (a)x + f (z) = 0. Therefore, (x, a, z), (x, b, z) satisfy (B),
i.e. x3 + af (x) + z = x3 + bf (x) + z = 0, hence (a − b)f (x) = 0. But x 6= 0, implying f (x) 6= 0. Hence
a = b. Therefore, f is injective.

We now show that (A) and (B) are equivalent, i.e. (x, y, z) satisfies (A) if and only if (x, y, z) satisfies (B).
From the condition given in the problem, (x, y, z) satisfies (A) implies (x, y, z) satisfies (B). To prove the
converse, let x, y, z ∈ R be a triple such that (x, y, z) satisfies (B), i.e. we have have f (x)3 +y ·f (x)+z = 0.
Let z 0 ∈ R such that (x, y, z 0 ) satisfies (A). This is possible since f is surjective. Therefore, (x, y, z 0 ) sat-
isfies (B), i.e. z 0 = −f (x)3 − yf (x). But this term is also equal to z. Therefore, z 0 = z, implying (x, y, z)
satisfies (A). Therefore, (A) and (B) are equivalent.

Take (x, y, z) such that x = 1, f (y) = −1, z = 0. Choosing such a y is possible since f is surjective. Then
(x, y, z) satisfies (A). Hence, f (1)3 + yf (1) = 0. Therefore, y = −f (1)2 , which implies f (−f (1)2 ) = −1,
by the definition of y. Take (x, y, z) such that x = 1, y = 0, f (z) = −1. Then (x, y, z) satisfies (A). Hence,
f (1)3 + z = 0, implying z = −(f (1))3 . Therefore, f (−f (1)3 ) = −1. But f is injective and f (z) = 1.

6
IMO Winter Camp 2010 Warm-Up Problems 7

Therefore, −f (1)2 = −f (1)3 , or equivalently, f (1)2 (f (1) − 1) = 0. Since f is injective, f (1) 6= f (0) = 0.
Therefore, f (1) = 1.

Take x = 1, z = −y − 1. Since f (1) = 1, (x, y, z) satisfies (B). Since (A) and (B) are equivalent, (x, y, z)
satisfies (A), i.e.
f (−y − 1) = −f (y) − 1, ∀y ∈ R. (1)

Substitute y = 0 into (1) yields

f (−1) = −1. (2)

Take x = −1, z = y + 1. Since f (−1) = −1, (x, y, z) satisfies (B), which implies (x, y, z) satisfies (A), i.e.
f (y + 1) = f (y) + 1, ∀y ∈ R. Inductively, we can show that

f (y + k) = f (y) + k, ∀y ∈ R, k ∈ Z. (3)

Because f (0) = 0, from (3), we have that

f (x) = x, ∀x ∈ Z. (4)

From (1) and (3), we get that f (−y − 1) = −f (y) − 1 = −(f (y + 1) − 1) − 1 = −f (y + 1). Therefore,

f (y) = f (−y), ∀y ∈ R. (5)

Let x, y be any fixed real numbers and let z = −f (x)3 − yf (x). Then (x, y, z) satisfies (A). Hence, (x, y, z)
satisfies (B), i.e. x3 + f (y)x + f (z) = 0. Therefore, f (z) = −x3 − f (y)x. By the definition of z, we now
have
f (−f (x)3 − yf (x)) = −x3 − f (y)x, ∀x, y ∈ R. (6)

Using (5), we get that

f (f (x)3 + yf (x)) = x3 + f (y)x, ∀x, y ∈ R. (7)

If x is an integer, then by (4), f (x) and x3 are also integers. Therefore, by (3) and (7), we have x3 +xf (y) =
f (f (x)3 + yf (x)) = f (x)3 + f (yf (x)) = x3 + f (yx). (We substitute y ← yf (x) and k ← f (x)3 into (3) for
the second last assertion and the fact that f (x) = x for all x ∈ Z, for the last assertion.) Therefore,

xf (y) = f (yx), ∀x ∈ Z, y ∈ R. (8)

Now, if m, n are integers with n 6= 0, then by (8), we have

 
 m nx + m f (nx + m) nf (x) + m m
f x+ =f = = = f (x) + .
n n n n n
Hence,
f (x + q) = f (x) + q, ∀x ∈ R, q ∈ Q. (9)

Substituting x = 0 into (9) gives us

f (q) = q, ∀q ∈ Q. (10)

Substituting y = −f (x)2 into (7) gives us 0 = f (0) = −x3 − f (−f (x)2 )x. Therefore, f (−f (x)2 ) = −x2 .
By (5), we have
f (f (x)2 ) = x2 , ∀x ∈ R. (11)

7
IMO Winter Camp 2010 Warm-Up Problems 8

Let c > 0 be an arbitrary real number. Then since f is surjective, there exists d ∈ R such that c = f (d)2 .
Since c 6= 0, f (0) = 0 and f is injective, d 6= 0. Therefore, by (11), we have f (c) = f (f (d)2 ) = d2 > 0.
Hence,
c > 0 ⇒ f (c) > 0, ∀c ∈ R (12)

Assume for some x, we have f (x) < x. Then there exists a rational number r for which f (x) < r < x.
Then by (12), we have f (x − r) > 0 and by (9), we have r > f (x) = f (r + (x − r)) = r + f (x − r) > r.
This is a contradiction. So we cannot have f (x) < x. Similarly, we cannot have f (x) > x. So the only
possibility for f is f (x) = x for all x ∈ R. It is easy to check f (x) = x satisfies the conditions of the
problem and so is the only solution.

Source: German Team Selection Test 2009

Comments: This problem is difficult in the sense that the solution requires many steps. However, every
step demonstrates an important technique that you should know to solve functional equations.

The first thing you should almost always do is to find solutions to the functional equation by inspection.
The most common ones are f (x) = c for some constant c, f (x) = x + c for some constant c, quadratic
equations and f (x) = x1d for some integer d.

You next want to establish properties of the functional equation, i.e. what is f (0)? If f (x) = 0, does it
mean x = 0? Is f one-to-one? Is f surjective? (Don’t do the latter two questions if one of the solutions
to the functional equation is, say, f (x) = x2 . You already know that this function is neither one-to-one
or surjective over the reals.)

Another technique is solving the problem over the rationals first, and then use detailed (and careful)
continuity arguments to solve the problem over all of the reals.

8
IMO Winter Camp 2010 Warm-Up Problems 9

Number Theory

N1 Let n be a positive integer. The sum of the positive divisors of n is s. Prove that the sum of the reciprocal
of these divisors is ns .

Solution: Let d1 , · · · , dt be the positive divisors of n with 1 = d1 < d2 < · · · < dt = n. Note that
n = di dt+1−i for each i. Therefore, d1i = dt+1−i
n . Hence,

t t t t
X 1 X dt+1−i 1X 1X s
= = dt+1−i = di = .
d
i=1 i i=1
n n i=1 n i=1 n

Source: Original

Comments: Always remember that the positive divisors√of n come in pairs as shown in this solution.
The exception is when n is a perfect square. In this case, n is paired with itself.

N2 A rational number x is written on a blackboard. In each step, you erase x and replace it with either x + 1
or − x1 . (If x = 0, you must replace x with x + 1). Prove that for any rational number p, if p currently
appears on the blackboard, then you can make 0 appear in a finite number of steps.

Solution: By adding 1 repeatedly to p (which is a legal step), we may assume that p ≥ 0. If p = 0,

we are already done. Let p = m n , where m, n > 0. I claim that after a finite number of steps, you can
obtain a rational number with a non-negative numerator non-negative which is strictly smaller than m.
−n
Let n = qm − r where 0 ≤ r < m. We erase m n and replace it with m . By adding 1 to this q times, we get
r
a fraction m and 0 ≤ r < m. We obtain a number with a non-negative numerator which is smaller than
m. Repeating this procedure gives us fractions with strictly smaller non-negative numerators. Hence, we
can eventually obtain a rational number with numerator 0, which is the number zero. 

Source: John Conway’s Talk on Knot Theory

Comments: This is motivated by the Euclidean algorithm and using a general technique in number
theory called descent. A strategy for this type of problem generally involves solving this problem for
certain values of p with small numerator and denominator. It should be pretty natural to try to lower the
numerator and/or denominator of the number on the board. I will leave as an exercise that starting at 0,
you can make p appear after a finite number of steps.

Another note is that we used a variant of the division algorithm. Instead of writing n = qm + r with
0 ≤ r < m, which is the more traditional method, we wrote it as n = qm − r with 0 ≤ r < m to make the
problem easier. The other useful variant is writing n as n = qm + r where −m/2 < r ≤ m/2.

N3 Find all triples of positive integers (a, b, c) such that a2 + 2b+1 = 3c .

Solution: The answers are (1, 2, 2) and (7, 4, 4).

9
IMO Winter Camp 2010 Warm-Up Problems 10

Since 2b+1 is even and 3c is odd, a2 is odd. Therefore, a2 ≡ 1 mod 4. Since 2b+1 ≡ 0 mod 4, 3c ≡ 1 mod
4. This implies 3c ≡ 1 mod 4. Hence, (−1)c ≡ 1 mod 4. We conclude that c is even. Therefore,
c c
2b+1 = (3 2 − a)(3 2 + a).
c c
Hence, 3 2 − a and 3 2 + a are both powers of 2. Since a is odd, both of these terms are even, so none
c c
of these terms are 1. Hence, we can let s, t be positive integers such that 3 2 − a = 2s and 3 2 + a = 2t .
t s t−1 s−1
Subtracting the former from the latter yields 2a = 2 − 2 . Hence, a = 2 − 2 . Since a is odd, s = 1.
c c
Therefore, a = 2t−1 − 1. Substituting this into 3 2 + a = 2t yields 3 2 + 2t−1 − 1 = 2t , or equivalently,
c
3 2 − 1 = 2t−1 .
c c
If t = 1, then 3 2 = 2, which has no integer solutions. If t = 2, then 3 2 − 1 = 2. Hence, c = 2,
a = 2t−1 − 1 = 22−1 − 1 = 1. Therefore, 1 + 2b+1 = 32 , implying b = 2. Hence, (1, 2, 2) is a solution to the
c
equation and is easily verified to be a solution. Otherwise, t > 2, implying 4 | 2t−1 . Hence, 3 2 ≡ 1 mod
c
4, which implies (−1) 2 ≡ 1 mod 4. Hence, 2c is even. Therefore,
c c
(3 4 − 1)(3 4 + 1) = 2t−1 .
c c
From this, we have that both 3 4 − 1, 3 4 + 1 are powers of 2 and are consecutive even integers. This is
c
only possible when the smaller integer is 2 and the larger integer is 4. Therefore, 3 4 − 1 = 2, implying
c = 4. Therefore, 32 − 1 = 2t−1 = a + 1, or equivalently, a = 7. Finally, 72 + 2b+1 = 34 , implying b = 4.
Therefore, (7, 4, 4) is a solution to the equation and is easily verified to be a solution. 

Source: Italian Team Selection Test 2008

Comments: One technique to solve Diophantine equations is consider reduce everything modulo a cer-
tain positive integer. The existence of a square and a power of 2 suggests that mod 4 and mod 8 are
possibilities. Problems like this test your comfort level with divisibility problems. You should always try
to find as many solutions by inspection as you can as a first step. (However, I do not expect you to initially
find the solution (7, 4, 4).)

Try to use modular arithmetic to solve the following problem.

Exercise: Let m, n be positive integers. Find the minimum possible value of |12m − 5n |.

N4 Find all positive integers that can be written in the form

m2 + n2 + 1
,
mn
for all positive integers m, n.

Solution: The answer is 3.

Let S = {(m, n) | mn divides m2 + n2 + 1, m, n ∈ N} and

m2 + n2 + 1
f (m, n) = .
mn

10
IMO Winter Camp 2010 Warm-Up Problems 11

Note that (m, n) ∈ S if and only if (n, m) ∈ S.

If m = 1, then n divides n2 + 2. This implies n | 2. Therefore, n = 1 or n = 2. Both yield the property

2 2
that m +n
mn
+1
= 3. This holds similar for n = 1. Therefore, if m = 1 or n = 1, then f (m, n) = 3.

We next consider the case m = n. If (m, n) ∈ S and m = n, then m2 divides 2m2 + 1. Therefore, m2 | 1.
Therefore, m = 1. Hence, if (m, m) ∈ S, then m = 1. This also yields that f (m, n) = 3.

m2 +n2 +1
Now, suppose (m, n) ∈ S with 1 < m < n. Suppose k = mn . I claim there exists 0 < n0 < m such
that (m, n0 ) ∈ S. Rewriting the equation gives us
n2 − kmn + (m2 + 1) = 0.
Hence, n is an integer root of the equation x2 − kmx + (m2 + 1) = 0. Let n0 be the other root. Since
n + n0 = km, which is an integer, n0 is an integer. Since nn0 = m2 + 1 > 0 and n > 0, n0 > 0. Note that
f (m, n) = f (m, n0 ), by rewriting the quadratic equation back to the form f (m, n0 ). Therefore, (m, n0 ) ∈ S.
I claim that n0 < m. Suppose on the contrary that n0 ≥ m. Since (m, n0 ) ∈ S and n, m > 1, n0 6= m by
the argument in the case m = n. Suppose n0 > m. Let n = m + a, n0 = m + b for some positive integers
a, b. Then nn0 = (m + a)(m + b) = m2 + 1, since the product of the roots of x2 − kmx + (m2 + 1) is m2 + 1.
Hence, m(a + b) + ab = 1, which is impossible since a, b ≥ 1. Therefore, n0 < m. This proves the claim.

Hence, (m, n0 ) ∈ S, which implies (n0 , m) ∈ S. Note that k = f (m, n) = f (m, n0 ) = f (n0 , m). If n0 = 1,
then from an earlier argument, k = 3. Otherwise, we repeat this process for 1 < n0 < m. Every time we
perform the previous paragraph, min{m, n} strictly decreases and remains positive and f (m, n) remains
constant. Hence, eventually, at least one of m, n = 1. By an earlier argument, this results in k = 3. Since
f (m, n) remains constant at every step, we conclude that f (m, n) = 3. This completes the problem. 

Source: Arthur Engel’s Problem-Solving Strategies

Comments: The technique used in this problem is called Vieta-jumping or descent. It is extremely
helpful in solving Diophantine equations with quadratic terms. By letting S be a solution set, you can
claim statements of the form if (a, b) ∈ S, then (c, d) ∈ S for certain choices of a, b, c, d. By doing so, you
can create a series of solutions for which min{m, n} is decreasing. Try to solve the following problem from
the 1988 International Mathematical Olympiad using Vieta-jumping.

Exercise: Suppose m, n are positive integers such that mn + 1 divides m2 + n2 . Prove that
m 2 + n2
mn + 1
is a perfect square.

N5 Find all prime numbers p such that the following statement is true: there are exactly p ordered pairs of
integers (x, y) such that 0 ≤ x, y < p and y 2 ≡ x3 − x mod p.

Solution: The answers are p = 2 and all primes p such that p ≡ 3 mod 4.

11
IMO Winter Camp 2010 Warm-Up Problems 12

If p = 2, then we can check that the only pairs that work are (0, 0) and (1, 0). There are exactly two
solutions. Hence p = 2 is an answer.

We may assume now that p is odd. For a ∈ {0, 1, · · · , p − 1}, we call a a quadratic residue modulo p if
and only if n2 ≡ a mod p for some integer n. Otherwise, we call a a quadratic non-residue. We recall
that the product of two quadratic residues is a quadratic residue, and the product of two quadratic non-
residues is a quadratic residue. Also, the product of a quadratic residue and a quadratic non-residue, is a
quadratic non-residue. Also recall that −1 is a quadratic residue of p if and only if p ≡ 1 mod 4 or p = 2. (*)

Note that (−1, 0), (0, 0), (1, 0) are the only solutions involving y = 0 and x ∈ {−1, 0, 1}. There are three
such solutions. We will henceforth only consider the cases when x ∈ {2, · · · , p − 2} and y 6= 0.

We will call x ∈ {2, · · · , p − 2} solvable if there exists y such that y 2 ≡ x3 − x mod p.

If p ≡ 1 mod 4, then note that x3 − x = −1 · ((−x)3 − (−x)). By the arguments made in (*), since −1 is
a quadratic residue mod 4, x is solvable if and only if −x is solvable. For each x ∈ {2, · · · , p − 2} which is
solvable, there are exactly two values of y such that (x, y) is a solution to the original equation. This is
because (x, y) is a solution if and only if (x, −y) is a solution. Note that x 6≡ −x and y 6≡ −y mod p since
p is odd and y 6= 0. Hence the number of values of x ∈ {2, · · · , p − 2} which is solvable is even and each
solvable x yields two values of y. Hence, the number of ordered pair solutions with x ∈ {2, · · · , p − 2} is
divisible by 4. Including the three aforementioned solutions, the number of solutions (x, y) is congruent
to 3 mod 4. But p ≡ 1 mod 4. Therefore, the number of solutions (x, y) cannot equal to p.

If p ≡ 3 mod 4, again, note that x3 − x = −1 · ((−x)3 − (−x)). Since −1 is a quadratic non-residue

mod p and the product of a quadratic residue and a quadratic non-residue is a quadratic non-residue, x is
solvable if and only if −x is not solvable. Hence, exactly half of the value of x ∈ {2, · · · , p − 2} are solvable
and using a similar argument as in the case p ≡ 1 mod 4, each solvable x yields two solutions. Therefore,
there are p−3
2 · 2 = p − 3 solutions. Adding the aforementioned three solutions, there are p solutions total.
Hence, p ≡ 3 mod 4 is a solution to the problem. 

Source: Turkish Team Selection Test 2005

Comments: The facts in the paragraph marked (*) are important topics to learn for number theory,
especially the fact that n2 ≡ −1 mod p has a solution if and only if p = 2 or p ≡ 1 mod 4. Please read
Theorem 4.9 and Chapter 7 of [3] for more details.

Another important point is that the very least you should be able to do for this problem, even without
knowledge of quadratic residues and of fact (*), is to conjecture your answer for p by solving this problem
for small primes p, say up to p = 17. The motivation behind this given solution arises from the work you do
for small values of p. Only then would you be able to consider the differences between primes 1 and 3 mod
4. The least that you should be able to write, is something along the lines of ”I conjecture that the solution
to this problem are p = 2 and all primes p ≡ 3 mod 4”. Of course, now that you have read and under-
stood this solution, I will expect much more than this. :) Using fact (*), you can try the following exercise.

Exercise: Prove that there are no pairs of positive integers a, b such that 4ab − a − b is a perfect square.

12
IMO Winter Camp 2010 Warm-Up Problems 13

Geometry

G1 A circle C and a point P are given on the same plane. Given any point Q on the circumference of C, let
M be the midpoint of P Q. Find the locus of point M , i.e. find all possible locations for point M .

Solution 1: The locus of M is exactly the dilation of C about point P with ratio 12 . The image is therefore
the circle with radius half of that of C whose centre is the midpoint of P and O, where O is the centre of C. 

|P M | |P N |
Solution 2: Let O be the centre of C, r the radius of C and N the midpoint of P O. Since |M Q| = |N O| = 1,
|M N | 1 r r
we have |QO| = 2. Hence, |M N | = 2. Therefore, M lies on the circle with radius 2 centered at N .

Conversely, let M be any point on the circle centered at N with radius 2r . Let Q be on the circumference
of C such that QO is parallel to M N . Then |M N| |P N | 1
|QO| = |P O| = 2 , which implies P, M, Q are collinear and
|P M |
|P Q| = 12 . Therefore, M is the midpoint of P Q. 

Source: Original

G2 Let A be a point outside of a circle C. Two lines pass through A, one intersecting C at B, C, with B closer
to A than C, and the other intersecting C at D, E, with D closer to A than E. The line passing through
D parallel to AC intersects C a second time at F and the line AF intersects C a second time at G. Let
M = EG ∩ AC. Prove that
1 1 1
= + .
|AM | |AB| |AC|

Solution: Note that ∠GAM = ∠F AM = ∠AF D = ∠GF D = ∠GED, the latter assertion following
|M A| |M E|
from that DEF G is cyclic. Therefore, ∆GAM ∼ ∆AEM . Hence, we have |M G| = |M A| , which implies
2
|M A| = |M G| · |M E| = |M B| · |M C|. The assertion follows from that DEF G is cyclic. We shall call
this property (*).

Note that M lies on the segment AB. Therefore, |AB| = |M A| + |M B| and |AC| = |M A| + |M C|. Hence,

1 1 1 |AB| + |AC| 2|M A| + |M B| + |M C|

= + = = .
|AM | |AB| |AC| |AB| · |AC| (|M A| + |M B|)(|M A| + |M C|)

This equality holds if and only if (|M A| + |M B|)(|M A| + |M C|) = 2|M A|2 + |M A||M B| + |M A||M C| ⇔
|M B| · |M C| = |M A|2 , which is true by property (*). 

Source: Romanian Team Selection Test 2006

Comments: If you are not familiar with Power of a Point, please read Chapter 6 of [1]. It is an important
and powerful technique to solve certain geometry problems, especially those involving circles and tangent
lines. Power of a Point is the motivation for this solution.

13
IMO Winter Camp 2010 Warm-Up Problems 14

G3 Let ABC be a triangle and D be a point on side BC. The internal angle bisector of ∠ADB and that of
∠ACB intersect at P . The internal angle bisector of ∠ADC and that of ∠ABC intersect at Q. Let M
be the midpoint of P Q. Prove that |M A| < |M D|.

Solution: Note that P is the excentre of 4ACD opposite C, and Q is the excentre of 4ABD opposite
B. Therefore, ∠P AD = 180−∠CAD
2 and ∠QAD = 180−∠BAD
2 .

Note also that ∠QDP = ∠QDA + ∠ADP = 12 (∠BDA + ∠CDA) = 90o . Therefore, |M P | = |M Q| =
|M D|. Consider the circle centred at M passing through P, Q, D. To prove that |M A| < |M D|, it suffices
to show that A is in the interior of the circle, i.e. ∠QAP is obtuse. But
180 − ∠BAD 180 − ∠CAD ∠BAD + ∠CAD
∠QAP = ∠QAD + ∠P AD = + = 180 − = 180 − ∠BAC
2 2 2
and this angle is greater than 90o (and less than 180o ). Therefore, |M A| < |M D|. 

Source: Original and similar to one on Irish Mathematical Olympiad 2005

Comments: It is important to observe excentres in a diagram. Their existence, along with incentres,
often simplifies problems involving angle bisectors.

G4 A convex quadrilateral ABCD has |AD| = |CD| and ∠DAB = ∠ABC < 90◦ . The line through D and
the midpoint of BC intersects line AB in point E. Prove that ∠BEC = ∠DAC.

Solution: Let F be on side AB such that |DA| = |DF |, which is also equal to |DC|. Since ∠AF D =
∠DAF = ∠DAB = ∠ABC, DF ||BC. Extend DF to meet EC at G. Since DF ||BC and ED passes
through the midpoint of BC, D is the midpoint of F G. Hence, |DA| = |DC| = |DF | = |DG|.
Hence, ACGF is cyclic. Then ∠F AC = ∠F GC = ∠DGC = ∠BCE. Since ∠DAB = ∠ABC,
∠DAC = ∠DAB − ∠F AC = ∠ABC − ∠BCE = ∠BEC, as desired. 

Source: Bulgarian Mathematical Olympiad 1998

Comments: Construction problems in geometry are fun. Of course, there are solutions that does not
involve extra constructions given in this solution. However, the construction of point F on AB is quite nat-
ural given that |AD| = |CD| and ∠DAB = ∠ABC. You will be given three equal lengths in AD, CD, F D
and that F D is parallel to AB. Many new properties of the diagram arise from this one point. Wow! :)

G5 Let ABC be a triangle with |AB| > |AC|. Let its incircle touch side BC at E. Let AE intersect this
incircle again at D. Let F be the second point on AE such that |CE| = |CF |. Let CF intersect BD at
G. Prove that |CF | = |F G|.

Solution: Let the line tangent to the incircle at D intersect BC (extended) at P . Since |P E| = |P D|
and |CE| = |CF |, CF ||P D.

14
IMO Winter Camp 2010 Warm-Up Problems 15

By Menelaus’ Theorem on ∆BCG, we have that

|GD| |BE| |CF |

· · = 1.
|DB| |EC| |F G|
|GD| |CP |
Since CF ||P D, |DB| = |P B| . Therefore, to prove that |CF | = |F G|, it suffices to prove that

|CP | |CE|
= .
|P B| |EB|

Note that the polar of P with respect to the incircle is line DE, which also contains A. Therefore, the
polar of A contains P . Hence, the polar of A is XY , where X, Y are points where the incircle touches
AC, AB, respectively. Hence, P lies on XY . Since |AX| = |AY |, |BY | = |BE|, |CE| = |CX|, we have
that

AY BE CX
· · = 1. (13)
Y B EC XA
Hence, AE, BX, CY are concurrent by Ceva’s Theorem. Since X, Y, P are collinear, by Menelaus’ Theorem
on ∆ABC, we have

AY BP CX
· · = −1. (14)
Y B P C XA
Comparing equations (13) and (14), yields

|CP | |CE|
= ,
|P B| |EB|

as desired. 

Source: China Team Selection Test 2008

Comments: Ceva’s and Menealus’ Theorem are essential pieces of knowledge for proving concurrency
and collinearity involving triangles. Please learn these theorems if you haven’t done so already.

Poles and polars have come to the rescue. Those familiar with poles and polars and harmonic conjugates
will realize that (E, P ) divides (B, C) harmonically. To learn more about these powerful techniques, please
consult Chapter 10 and 11 of [1].

15
IMO Winter Camp 2010 Warm-Up Problems 16

Combinatorics

C1 Find the number of subsets of {1, 2, · · · , 10} that contain its own size. For example, the set (1, 3, 6) has 3
elements and contains 3.

Solution: The answer is 512.

For each k ∈ {0, 1, · · · , 10}, we want to find how many elements there are of size k. Note that k 6= 0
since 0 6∈ {1, · · · , 10}. Every

subset of size k must contain k. The other k − 1 elements can be anything.
9
Therefore, there are k−1 subsets of size k that contains k. Therefore, the number of subsets that contain
its own size is
10   X 9  
X 9 9
= = 29 = 512.
k−1 k
k=1 k=0

Source: Original

C2 A sequence of non-negative integers is defined by G0 = 0, G1 = 0 and Gn = Gn−1 + Gn−2 + 1 for every

n ≥ 2. Prove that for every positive integer m, there exists a positive integer a such that Ga , Ga+1 are
both divisible by m.

Solution: Consider each pair (Gt , Gt+1 ) modulo m. Since there are at m2 pairs of integers (x, y) mod-
ulo m, the sequence (Gt , Gt+1 ) modulo m eventually repeats, by Pigeonhole Principle. It suffices to
prove that (Ga , Ga+1 ) = (0, 0) = (G0 , G1 ) modulo m, for some positive integer a. Suppose the sequence
{(Gt , Gt+1 } repeats the first time at index t = a. Then (Ga , Ga+1 ) = (Gj , Gj+1 ) modulo m for some
0 ≤ j < t. If j = 0, then (Ga , Ga+1 ) = (0, 0), implying Ga , Ga+1 are both divisible by m and we are
done. Otherwise, note that the pair that appears before (x, y) is (y, x + y − 1). In other words, given a
pair of consecutive terms in the sequence, the previous pair of consecutive terms is determined. Hence,
(Ga , Ga+1 ) = (Gj , Gj+1 ) implies (Gj−1 , Gj ) = (Ga−1 , Ga ), contradicting that the sequence repeats the
first time at index a. Therefore, Ga , Ga+1 are both divisible by m. 

Source: Estonian Mathematical Olympiad 2007

Comments: Pigeonhole Principle is a powerful tool to solve certain existence type of combinatorics prob-
lems. It is an art to determine what are the pigeons and what are the holes.

C3 A set S of ≥ 3 points in a plane has the property that no three points are collinear, and if A, B, C are
three distinct points in S, then the circumcentre of ∆ABC is also in S. Prove that S is infinite.

Solution: Suppose on the contrary that S is finite. Consider the convex hull P of S, i.e. the smallest poly-
gon containing all of S (on the polygon’s boundary and interior). Since |S| ≥ 3, P contains at least three
vertices. Let AB be a side of P and C be any other vertex of P. Note that P is contained in the halfplane
of AB containing C and there are no points in S in the halfplane not containing C. If ∠ACB > 90◦ , then
the circumcircle of ∆ACB lies in the halfplane of AB not containing C, which does not contain any point

16
IMO Winter Camp 2010 Warm-Up Problems 17

in S. Hence, the circumcircle of ∆ACB is not in S. This contradicts the assumption given in the problem.
Otherwise, ∠ACB is acute. We let O1 be the circumcentre of ∆ACB and recursively, define Oi+1 to be
the circumcentre of ∆AOi B. Hence, Oi ∈ S for all i ≥ 1. Hence, ∆AO1 B = 2∠ACB. We choose the
smallest index k such that ∠AOk B ≥ 90◦ . Therefore, ∠AOk B = 2∠AOk−1 B. Since ∠AOk−1 B ≤ 90◦ by
our choice of k, ∠AOk−1 B < 180◦ . If ∠AOk B = 90◦ , then Ok+1 lies on AB, contradicting the fact that
no three points in S are collinear. (Note that Oi 6= A, B for any i, since the circumcentre of a triangle
cannot be any of the vertices of the triangle.) Otherwise, 90 < ∠AOk B < 180◦ , which implies Ok+1 lies
on the halfplane of AB not containing Ok . Hence, Ok+1 cannot lie in S, but this contradicts the fact that
Oi ∈ S for all i ≥ 1. Hence, S cannot be finite, implying S is infinite. 

Source: Unknown

Comments: When you are given a finite set of points in the plane in a problem, you should always
consider the convex hull of the set. It is likely to simplify the problem. Try the following problem from
the 1999 International Mathematical Olympiad by considering the convex hull.

Exercise: Find all finite set of points S such that for every pair of points A, B ∈ S, the perpendicular
bisector of A and B is an axis of symmetry for S.

C4 Let n, k be positive even integers. A survey was done on n people where on each of k days, each person was
asked whether he/she was happy on that day and answered either ”yes” or ”no”. It turned out that on any
two distinct days, exactly half of the people gave different answers on the two days. Prove that there were
at most n− nk people who answered ”yes” the same number of times he/she answered ”no” over the k days.

Solution: Consider a grid with k rows and n columns. Label the days from 1 to k and the people from
1 to n. We place a 1 in the ith row and j th column if person j is happy on day i and a 0 otherwise. Since
on any two days, exactly half of the people gave different answers on the two days, we have that for any
two rows, exactly
half of the columns differ. Hence, the number of pairs of squares in the same column
that differ, is k2 · n2 . For each person that answered yes the same number of times as he/she answered no,
the number of pairs of squares in the person’s column that differ is ( k2 )2 . Hence, the maximum number of
people that could have answered yes the same number of times as he/she answered no, is
k n

nk(k−1)
2 2 4 n(k − 1) n
= k2
= =n− ,
( k2 )2 4
k k

as desired. 

Source: Iran Mathematical Olympiad 2006

Comments: An important key to solving combinatorics problem is find out what you are suppose to be
counting. If you can count a certain object in two different ways, you can yield some extremely neat and
powerful identities and relationships. If you take the time to solve enough combinatorics problems, it will
become more natural as to what objects you should be counting. Use this technique to solve the following
problem from the 1998 International Mathematical Olympiad

17
IMO Winter Camp 2010 Warm-Up Problems 18

Exercise: In a contest, there are m candidates and n judges, where n ≥ 3 is an odd integer. Each
candidate is evaluated by each judge as either pass or fail. Suppose that each pair of judges agrees on at
most k candidates. Prove that
k n−1
≥ .
m 2n

C5 There are n ≥ 5 people in a room, where each pair is classified as friends or strangers. No three people are
mutually friends. There also exist an odd number of people P1 , · · · , Pm such that Pi is friends with Pi+1
for all i ∈ {1, · · · , m}, where the indices are taken modulo m. Prove that there exists one person who is
friends with at most 2n/5 people.

Solution: Given a positive integer m ≥ 3, we will call an ordered set of m people P1 , · · · , Pm a cycle if
Pi is friends with Pi+1 for all i ∈ {1, · · · , m}, where the indices are taken modulo m. If m is odd, we call
a cycle an odd cycle. Given the condition in the problem, an odd cycle exists. Consider the odd cycle in
the room P1 , · · · , Pg with the smallest number of people. Since no three people who are mutually friends,
the odd cycle must contain at least five people, i.e. g ≥ 5

I claim that Pi is not friends with Pj if |i − j| =

6 1 mod g. Without loss of generality, suppose P1 is friends
with Pj for some j 6= 2, j 6= g. Consider the set of vertices (P1 , · · · , Pj ) and (P1 , Pj , Pj+1 , · · · , Pg ). These
are both cycles and the sum of the the two sets is j + (g − j + 2) = g + 2, which is odd. Therefore, one of
these cycles is odd, and smaller than g. This contradicts the minimality of g.

I claim that every person outside of the cycle is friends with at most two people inside the cycle. Suppose
there exists a person A outside of the cycle who is friends with three people in the cycle, say Pi , Pj , Pk
with i < j < k. Note that no two of Pi , Pj , Pk are friends, since otherwise, A and these two people are
three mutual friends, which is not allowed. Consider the cycle drawn as a graph with Pi , Pj , Pk appearing
in clockwise order. Let x be the distance of the path between Pi and Pj on the cycle not containing
Pk . Define y, z analogously. Therefore, x + y + z = g. Since no two of Pi , Pj , Pk are friends, x, y, z > 1.
Therefore, max{x, y, z} < g−2. Since g is odd, at least one of x, y, z is odd. Without loss of generality, sup-
pose x is odd. Then APi Pi+1 · · · Pj A is an odd cycle of length x + 2 < g, contradicting the minimality of g.

Therefore, every person outside of the cycle is friends with at most two people inside the cycle. Hence,
the number of pairs of friends with one person in the pair in the cycle and the other person in the
pair outside of the cycle, is at most 2(n − g). Therefore, there exists a person inside the cycle who
is friends with at most 2(n−g)
g friends outside of the cycle. Since this person is friends with exactly two
2(n−g) 2n 2n
people inside the cycle, this person is friends with at most g +2 = g ≤ 5 , since g ≥ 5, as desired. 

Source: Unknown

Comments: Knowledge in graph theory is of utmost importance to have to solve this problem. Please
read [4] to learn what graphs are and their important properties.

18
IMO Winter Camp 2010 Warm-Up Problems 19

References
[1] Kedlaya, Kiran. Geometry Unbound
http://www-math.mit.edu/∼kedlaya/geometryunbound/gu-060118.pdf

[2] Lee, Hojoo. Topics in Inequalities - Theorems and Techniques.

http://www.eleves.ens.fr/home/kortchem/olympiades/Cours/Inegalites/tin2006.pdf
[3] Sato, Naoki. Number Theory.
http://www.artofproblemsolving.com/Resources/Papers/SatoNT.pdf

[4] Tang, Adrian. Graph Theory.

http://math.ucalgary.ca/∼atang/mathcircles/AT-GraphTheory.pdf

19