Problem A1: Find the volume of the region of points (x, y, z) such that

(x2 + y 2 + z 2 + 8)2 ≤ 36(x2 + y 2 ).

Solution: Set r 2 = x2 + y 2 . The intersection of this region with the plane z = t is

the set St of points (x, y, t) such that

0 ≥ r 2 − 6r + t2 + 8 = (r − 3)2 + t2 − 1,

i.e. √ √
3− 1 − t2 ≤ r ≤ 3 + 1 − t2 .
√
In other words, radius 3 − 1 − t2 and outer circle of
√ St is a ring with inner circle of√
radius 3 + 1 − t2 . Thus the area of St is 12π 1 − t2 . It follows that the volume in
question is equal to Z 1 Z 1 √
St dt = 12π 1 − t2 dt = 6π 2 .
−1 −1

Problem A2: Alice and Bob play a game in which they take turns removing stones
from a heap that initially has n stones. The number of stones removed at each turn
must be one less than a prime number. The winner is the player who takes the last
stone. Alice plays first. Prove that there are infinitely many n such that Bob has a
winning strategy. (For example, if n = 17, then Alice might take 6 leaving 11; then
Bob might take 1 leaving 10; then Alice can take the remaining stones to win.)
Solution: Let us call n winning if Bob has a winning strategy. Call n loosing if Alice
has a winning strategy. The key observation is that each n is either winning or loosing.
In fact, 1 is loosing. If all positive integers smaller than n are either winning or loosing
then n is loosing if there is a prime p such that n + 1 − p is winning and n is winning
otherwise.
Suppose that the set W of all winning numbers is finite so there is k > 1 such that
every n ≥ k is loosing. This means that for every n ≥ k there is a prime p such that
n + 1 − p is winning. In particular, the distance of n + 1 to the nearest prime is at
most k for every n ≥ k. Note however that this is not possible since all the numbers
(k!)3 ± i are composite for i = 0, 1, 2, 3, ..., k.
Problem A3: Let 1, 2, 3, . . . , 2005, 2006, 2007, 2009, 2012, 2016, . . . be a sequence de-
fined by xk = k for k = 1, 2, . . . , 2006 and xk+1 = xk + xk−2005 for k ≥ 2006. Show that
the sequence has 2005 consecutive terms each divisible by 2006.
Solution: For k ≤ 0 define xk = xk+2006 − xk+2005 . Then xh is defined for every integer
and satisfies the recursive relation xk+1 = xk + xk−2005 . Note that x0 = x−1 = ... =
x−2004 = 1 and x−2005 = x−2006 = ...x−4009 = 0. There are integers m and k > 0
such that xm+i − xm+i+k are divisible by 2006 for i = 0, 1, 2, ..., 2005. It follows from
the recursive formula by obvious induction that xn+k − xn is divisible by 2006 for all
n. Since x−4009 , x−4008 , ..., x−2005 are all divisible by 2006, the same is true for the 2005
numbers x5000k−4009 , x5000k−4008 , ..., x5000k−2005 .

1
Problem A4: Let S = {1, 2, . . . , n} for some integer n > 1. Say a permutation π of
S has a local maximum at k ∈ S if

(i) π(k) > π(k + 1) for k = 1;

(ii) π(k − 1) < π(k) and π(k) > π(k + 1) for 1 < k < n;

(iii) π(k − 1) < π(k) for k = n.

(For example, if n = 5 and π takes values at 1, 2, 3, 4, 5 of 2, 1, 4, 5, 3, then π has a local

maximum of 2 at k = 1, and a local maximum of 5 at k = 4.) What is the average
number of local maxima of a permutation of S, averaging over all permutations of S?
Solution: Let M (π) be the number of local maxima of π. For each k let pk be the
number of permutations which have a local maximum at k. The problem asks us to
evaluate X
M (π)/n!.
π
P Pn
Note that π M (π) = k=1 pk . The computation of pk is quite easy. To get a
permutation with a local
maximum at 1 we need to choose two elements for π(1) > π(2),
n

done in 2 ways and then order the remaining elements in (n − 2)! ways,
which can be
n
so p1 = 2 (n − 2)! = n!/2 and the same argument shows that pn = n!/2. To get a
permutation with a local maximum at 1 < k < n we need
to choose three elements
for π(k − 1) < π(k) > π(k + 1), which can be done
in 2 n3 ways and then order the
remaining elements in (n − 3)! ways. Thus pk = 2 n3 (n − 3)! = n!/3. It follows that
n
X n! n! n+1
pk = 2 · + (n − 2) = n! .
k=1
2 3 3

The average number of local maxima is then equal to (n + 1)/3.

Problem A5: Let n be a positive odd integer and let θ be a real number such that
θ/π is irrational. Set ak = tan(θ + kπ/n), k = 1, 2, . . . , n. Prove that
a1 + a 2 + · · · + a n
a1 a2 · · · an
is an integer, and determine its value.
Solution: Recall that tan α = (e2iα − 1)/i(e2iα + 1). It follows that

e2i(θ+kπ/n) − 1 e−2iθ − e2πik/n

ak = = i
i(e2i(θ+kπ/n) + 1) e−2iθ + e2πik/n

Set z = e−2iθ . Recall that ni=1 (z −e2πik/n ) = z n −1 and ni=1 (z +e2πik/n ) = z n −(−1)n .
Q Q
It follows that
zn − 1
a1 a2 ...an = in n .
z +1

2
Now n
X z − e2πik/n
a1 + ... + an = i
z + e2πik/n
k=1

Note that
z − e2πik/n
(z n + 1)
z + e2πik/n
is a monic polynomial of degree n which vanishes at −e2πil/n for all 1 ≤ l ≤ n, l 6= k
and assumes value −2n at −e2πik/n (in general, if a is a root of a polynomial f (z) then
g(z) = f (z)/(z − a) is a polynomial and g(a) = f 0 (a)). It follows that
n
n
X z − e2πik/n
(z + 1)
z + e2πik/n
k=1

is polynomial of degree n, with leading coefficient n and it assumes value −2n at

−e2πil/n for all 1 ≤ l ≤ n. The same property has the polynomial n(z n − 1) so
n
n
X z − e2πik/n
(z + 1) = n(z n − 1).
k=1
z + e2πik/n

We see that
zn − 1
a1 + ... + an = in n
z +1
and
a1 + a 2 + · · · + a n in
= n = (−1)(n−1)/2 n.
a1 a2 · · · an i

Remark: From the equality

e2i(θ+kπ/n) − 1
ak =
i(e2i(θ+kπ/n) + 1)

we get that
i − ak
e2i(θ+kπ/n) =
i + ak
so
(i − ak )n
= e2in(θ+kπ/n) = e2inθ
(i + ak )n
It follows that a1 , ..., an are roots of the polynomial e2inθ (i + x)n − (i − x)n , i.e.

e2inθ (i + x)n − (i − x)n = (e2inθ + 1)(x − a1 )...(x − an ).

Comparing the coefficient at xn−k we see that

n
ik (e2inθ − (−1)n−k )


X
k k
ai1 ai2 ...aik = (−1) .
i1 <i2 ..<ik
e2inθ + 1

3
Problem A6: Four points are chosen uniformly and independently at random in the
interior of a given circle. Find the probability that they are the vertices of a convex
quadrilateral.
Solution: Let A1 , A2 , A3 , A4 be the four random points in in a unit circle (clearly we
may assume that the circle has radius 1). The lines A1 A2 , A2 A3 and A3 A1 divide the
circle into 7 parts. Note that A1 A2 A3 A4 is convex iff A4 is in one of the three parts
which contain exactly one of the sides of the triangle A1 A2 A3 . So the probability that
A1 A2 A3 A4 is convex is equal to the expected area of these three parts divided by π (the
area of the circle). Note that the expected area E of the triangle A1 A2 A3 divided by
π is the probability that Ar belongs to the triangle A1 A2 A3 . The expected area Ei of
the part which intersects A1 A2 A3 only at the vertex Ai divided by π is the probability
that Ai is in the triangle Aj Ak A4 (here {1, 2, 3} = {i, j, k}). Thus E1 = E2 = E3 = E
and therefore the probability that A1 A2 A3 A4 is convex is (π − 4E)/π = 1 − 4E/π. So
it suffices to compute E.
Consider the set
√ √
{S = (α, r, a, b, x, y) : α ∈ [0, 2π), 0 ≤ r ≤ 1, − 1 − r 2 ≤ a, b ≤ 1 − r 2 , x2 +y 2 ≤ 1}.

The map

F : (α, r, a, b, x, y) 7→ (a cos α+r sin α, −a sin α+r cos α, b cos α+r sin α, −b sin α+r cos α,

x cos α + y sin α, −x sin α + y cos α)

is a parametrization of the product T of three unit circles (the inverse map corresponds
to rotating the triangle with vertices A1 , A2 , A3 inside the unit circle to a triangle
0 0 0 0 0 0
A1 A2 A3 such that the line A1 A2 is parallel to the x-axis and above it; here A1 = (a, r),
0 0
A2 = (b, r), A3 = (x, y); α is the angle of rotation). For a point t = (A1 , A2 , A3 ) of T ,
let A(t) be the area of A1 A2 A3 . Thus the expected area of a random triangle in the
unit circle is Z Z
−3 −3
E=π A=π A ◦ F |JF |
T S

where JF is the Jacobian of F . Note that A(F (α, r, a, b, x, y)) = |a−b||y −r|/2. Also it
is not hard to see that |JF | = |a − b| (this requires computation of a 6 × 6 determinant,
but it is not hard). Thus
√ √
2π 1 1−r 2 1−r 2
1
Z Z Z Z Z
E = π −3 (a − b)2 |r − y|da db dx dy dr dα
2 0 0 D
√
− 1−r 2
√
− 1−r 2

where D = {x2 + y 2 ≤ 1} is the unit circle. Since the angle is not involved in the
function under integral, after evaluation of the first two most-right integrals we get

8 −2 1
Z Z
E= π |r − y|(1 − r 2 )2 dx dy dr.
3 0 D

4
Now
Z Z 1 Z √1−y2 Z 1 p
|r − y|dx dy = |r − y| √ dx dy = 2 |r − y| 1 − y 2 dy =
D −1 − 1−y 2 −1

r Z 1 Z rp
4
Z p p
2[ 2
(r − y) 1 − y dy + 2
(y − r) 1 − y dy] = 4r 1 − y 2 dy + (1 − r 2 )3/2 .
−1 r 0 3
Thus
8 −2 1
Z rp
4
Z
2 2
E= π (1 − r ) [4r 1 − y 2 dy + (1 − r 2 )3/2 ] dr.
3 0 0 3
Rrp √ d
Since the derivative of 0 1 − y 2 dy is 1 − r 2 and dr (1 − r 2 )3 = −6r(1 − r 2 )2 , inte-
gration by parts yields
Z 1 Z rp
2 1
Z
2 2
(1 − r ) 4r 2
1 − y dy dr = (1 − r 2 )7/2 dr.
0 0 3 0

Thus
1
16 35
Z
E = π −2 (1 − r 2 )7/2 dr = .
3 0 48π
The probability that four random points in a circle form a convex quadrilateral equals
then 1 − 35/12π 2 .
Problem B1: Show that the curve x3 + 3xy + y 3 = 1 contains only one set of three
distinct points, A, B, and C, which are vertices of an equilateral triangle, and find its
area.
Solution: The key here is to realize that the curve is just a union of a line and a point.
Indeed, note that

x3 +3xy+y 3−1 = (x+y−1)(x2 +y 2 −xy+x+y+1) = (x+y−1)((x+1)2 +(y+1)2 +(x−y)2 )/2.

Thus our curve consists of the point (−1, −1) and the line x+y = 1. It is now clear that
any equilateral triangle with vertices on the curve must have vertex (−1, −1) and the
side opposite this √
vertex has midpoint (1/2,√1/2).√Thus, there is unique such√triangle,
its height is h = 3 2/2, its side is a = 2h/ 3 = 6 and its area is ah/2 = 3 3/2.
Problem B2: Prove that, for every set X = {x1 , x2 , . . . , xn } of n real numbers, there
exists a non-empty subset S of X and an integer m such that
 
 X  1
m + s ≤ .
 
  n+1
s∈S

Solution: Consider the intervals [i/(n + 1), (iP+ 1)/(n + 1)], i = 0, 1, ..., n. The problem
asks as to show that the fractional part of s∈S s belongs to either [0, 1/(n + 1)] or
[n/(n + 1), 1] for some non-empty S. If the fractional part of any of the numbers x1 ,

5
x1 +x2 , ..., x1 +x2 +...+xn is in one of the intervals [0, 1/(n+1)] or [n/(n+1), 1] we are
done. Otherwise, each of these n fractional parts belongs to one of the remaining n − 1
intervals. By pigeon-hole principle, there are 1 ≤ k < m ≤ n such that the fractional
parts of both x1 + ... + xk and x1 + ... + xm belong to the same interval. It follows then
that the fractional part of the difference (x1 + ... + xm ) − (x1 + ... + xk ) = xk+1 + ... + xm
is in one of the intervals [0, 1/(n + 1)] or [n/(n + 1), 1] .
Problem B3: Let S be a finite set of points in the plane. A linear partition of S is
an unordered pair {A, B} of subsets of S such that A ∪ B = S, A ∩ B = ∅, and A and
B lie on opposite sides of some straight line disjoint from S (A or B may be empty).
Let LS be the number of linear partitions of S. For each positive integer n, find the
maximum of LS over all sets S of n points.
Solution: For a point x let LxS be the number of linear partitions of S obtained by
lines through x. Consider a set S 0 of n + 1 points and let S 0 = S ∪ {x} where S consists
of n points. Each liner partition of S 0 restricts to a linear partition of S. This is
surjective map from linear partitions of S 0 to linear partitions of S. If a linear partition
of S can be obtained from 2 different linear partitions of S 0 then it can be realized by
line through x and there are exactly 2 partitions of S 0 restricting to it. It follows that
LS 0 = LS + LxS . Now for LxS rotate a line through x and note that you get at most n
partitions of S this way (you change a partition when you pass through a point of S),
and the equality holds if no three points of S 0 are collinear. Thus LS 0 ≤ LS + n and
the equality holds if no three
points of S are collinear. A straightforward induction
0
n
shows now that LS ≤ 1 + 2 and equality holds if on three points of S are collinear.
Problem B4: Let Zn denote the set of points in Rn whose coordinates are 0 or 1.
(Thus Zn has 2n elements, which are the vertices of a unit hypercube in Rn .) Given
a vector subspace V of Rn , let Zn (V ) denote the number of members of Zn that lie in
V . Let k be given, 0 ≤ k ≤ n. Find the maximum, over all vector subspaces V ⊆ Rn
of dimension k, of the number of points in V ∩ Zn .)
Solution: Clearly the 2k points whose first n − k coordinates are 0 are in the k
dimensional subspace of all points whose first n − k coordinates are 0. Suppose we
have 2k + 1 of our points in a k dimensional subspace V . Consider the projection
Rn −→ Rn−1 given by (x1 , ..., xn ) 7→ (x1 , ..., xn−1 ). Clearly it maps Zn onto Zn−1
and for each point of Zn−1 there are exactly 2 points of Zn mapped to it. If this
projection is injective on V then the image of V is a k-dimensional subspace of Rn−1
which contains 2k + 1 points of Zn−1 . If the projection is not injective then the image
of V has dimension k − 1 and the image of Zn ∩ V contains at least 2k−1 + 1 different
points of Zn−1 . Continuing this process we get an l dimensional subspace of R which
contains at lest 2l + 1 elements of Z1 . This is clearly false, so a k dimensional subspace
of Rn contains at most 2k elements of Zn .
R1
Problem B5: For each continuous function f : [0, 1] → R, let I(f ) = 0 x2 f (x) dx
R1
and J(x) = 0 x (f (x))2 dx. Find the maximum value of I(f ) − J(f ) over all such
functions f .

6
Solution: Recall that ab ≤ (a+b)2 /4 for any real numbers a, b. In particular, f (x)(x−
f (x)) ≤ x2 /4 for all x. Thus
Z 1 Z 1 Z 1
2 2
I(f )−J(f ) = (x f (x)−x (f (x)) ) dx = xf (x)(x−f (x)) dx ≤ x3 /4 dx = 1/16.
0 0 0

and the equality holds for f (x) = x/2.

Problem B6: Let k be an integer greater than 1. Suppose that a0 > 0 and define
1
an+1 = an + √
k a
n

for n ≥ 0. Evaluate
ak+1
n
lim .
n→∞ nk

Solution: Let f (x) = x + x−1/k . Clearly f (x) > 1, f is increasing on (1, ∞) and
f (x) > 2 for x > 1. Let c = (k + 1/k)k/k+1 . We claim that f (cxk/k+1 ) ≥ c(x + 1)k/k+1 .
In fact, this is equivalent to
k+1 k+1 k k+1 1 k
c k x+1= (x + )≥ x k+1 (x + 1) k+1 ,
k k+1 k
i.e. to
k 1 k
≥ x k+1 (x + 1) k+1 .
x+
k+1
This follows from the generalized AMGM inequality:
1 k 1 k k
x k+1 (x + 1) k+1 ≤ x+ (x + 1) = x + .
k+1 k+1 k+1
It is clear that an increases and hence tends to infinity (pass to the limit in the
recursion formula for an ). There is an m such that am > c and then easy induction
and the inequality above yield that

am+n ≥ c(n + 1)k/k+1 (∗).

On the other hand, using (∗) we get

n−1 n n
1 1 X 1 1 dx
X Z
am+n = am+1 + 1/k
≤ am+1 + ≤ am+1 + ≤
i=1 am+i
c1/k i=2
i1/(k+1) c1/k 1 x1/(k+1)

1 k + 1 k/k+1
≤ am+1 + (n − 1) ≤ am+1 + cnk/k+1
c1/k
k
1 k+1
(since c1/k k
= c). Thus we proved that

am+1 + cnk/k+1 ≥ am+n ≥ c(n + 1)k/k+1 .

7
Pinching theorem yields now that
an
lim k/k+1
=c
n→∞ n

and therefore
ak+1
n k+1 k+1 k
lim = c = ( ) .
n→∞ nk k

Remark 1 Set bn = am+n /nk/k+1 . We claim that bn is decreasing. In fact,

−(k+1)/k n k/(k+1)
bn+1 /bn = (1 + am+n )( ) ≤1
n+1
is equivalent to
(k+1)/k nk/k+1
am+n ≥ .
(n + 1)k/k+1 − nk/k+1
Since the mean value theorem yields (n + 1)k/k+1 − nk/k+1 = (k/(k + 1))u−1/(k+1) for
some n ≤ u ≤ n + 1 and nk/k+1 u1/(k+1) ≤ n + 1, it suffices to show that

(k+1)/k k+1
am+n ≥ (n + 1),
k
which is equivalent to (∗).
Remark 2 The limit can be computed easily using the following Stolz Theorem (an
analog of L’Hospitals rule for sequences)
Stolz Theorem If an , bn are sequences such that bn is increasing and unbounded and
an+1 − an
lim =g
n→∞ bn+1 − bn

exists, then
an
lim = g.
n→∞ bn

k+1/k
Let us apply it to our sequence an and bn = n so the problem reduces to finding
k+1/k k+1/k
the limit of an+1 − an . The Mean Value theorem for the function xk+1/k implies
that
k+1/k k + 1 1/k k + 1 1/k −1/k
an+1 − ak+1/k
n = un (an+1 − an ) = un a n
k k
for some an ≤ un ≤ an+1 . Since an+1 /an converges to 1, we see that un /an converges
to 1 and therefore
k+1/k k+1
lim ak+1/k
n /n = lim (an+1 − ak+1/k
n )= .
n→∞ n→∞ k