Solutions for Mathematical Reflections 4

Juniors

J19. Let a, b, c, d be positive integers such that 3(a + b) ≥ 2|ab + 1|. Prove
that
9(a3 + b3 ) ≥ |a3 b3 + 1|

Proposed by Dr. Titu Andreescu, University of Texas at Dallas

First solution by Ivan Borsenco, University of Texas at Dallas

First solution. Our inequality can can be rewritten as

3(a + b) a2 b2 − ab + 1
≥ ,
2|ab + 1| 6(a2 − ab + b2 )
because a2 b2 − ab + 1 ≥ 0 and a2 − ab + b2 ≥ 0. We will prove that
3(a + b) a2 b2 − ab + 1
≥1≥
2|ab + 1| 6(a2 − ab + b2 )
Left side of the inequality is true from the problem statement,
thus it remains to prove 6(a2 − ab + b2 ) ≥ a2 b2 − ab + 1 or
9(a + b)2
6a2 − 3ab + 6b2 ≥ (ab + 1)2 . But ≥ (ab + 1)2 and it is enough
4
to show that 4(2a2 − ab + 2b2 ) ≥ 3(a + b)2 which is equivalent to
5a2 + 5b2 ≥ 10ab or 5(a − b)2 ≥ 0 and the problem is solved.

Second solution by Ashay Burungale

Second solution. Note that a + b ≥ 0. First of all let us prove

1 9
(a2 − ab + b2 ) ≥ · | (a + b)2 − 3ab|
6 4
This is equivalent 8(a2 − ab + b2 ) ≥ |3(a + b)2 − 4ab| or

8(a2 − ab + b2 ) ≥ 8(a − b)2 + 8|ab| ≥ |3(a − b)2 + 8ab| = |3(a + b)2 − 4ab|.

1
 Using 3(a + b) ≥ 2|ab + 1| we get
1 9 1
· | (a + b)2 − 3ab| ≥ |(ab + 1)2 − 3ab)|
6 4 6
Thus
1
(a2 − ab + b2 ) ≥ |(ab + 1)2 − 3ab)|.
6
Multiplying the right side with 3(a + b) and the left side with 2|ab + 1|
our inequality becomes

9(a3 + b3 ) ≥ |a3 b3 + 1|

and we are done.

Also solved by Daniel Campos Salas, Costa Rica.

2
 J20. Let a, b, c, d be positive integers such that ab+cd = (a+b)(c+d).
a) Prove that equation has infinitely many of solutions such that
a, b, c, d are pairwise not equal.
√
3+1
b) Prove that max(a, b, c, d) ≥ √ (a + b + c + d)
4 3
Proposed by Ivan Borsenco, University of Texas at Dallas

Solution by Ivan Borsenco, University of Texas at Dallas

Solution. a) Our equation can be rewritten as

(a − (c + d))(b − (c + d)) = c2 + cd + d2
So it is enough to consider a−(c+d) = c2 +cd+d2 and b−(c+d) = 1.
Then we obtain quadruple (c2 +cd+d2 +c+d, c+d+1, c, d) which generates
infinitely number of of pairwise different solutions.

b) Using inequalities (a + b)2 ≥ 4ab, (c + d)2 ≥ 4cd we have

(a + b)2 + (c + d)2 ≥ 4(a + b)(c + d).
Without loss of generality we can suppose a + b ≥ c + d. From upper
equation we have
√ √
((a + b) − (2 + 3)(c + d))((a + b) − (2 − 3)(c + d)) ≥ 0.
√
Knowing that (a + b) − (2 − 3)(c + d) ≥ 0 we deduce that
√ 1
(a + b) − (2 + 3)(c + d) ≥ 0 or √ (a + b) ≥ c + d.
2+ 3
√
3+ 3
Thus √ (a+b) ≥ (a+b+c+d) and using 2·max(a, b, c, d) ≥ a+b
2+ 3
√ √
2 3( 3 + 1)
we get √ max(a, b, c, d) ≥ (a + b + c + d).
2+ 3
√
4+2 3
Rewriting what we got: max(a, b, c, d) ≥ √ √ (a+b+c+d) =
√ √ 4 3( 3 + 1)
( 3 + 1)2 3+1
= √ √ (a + b + c + d) = √ (a + b + c + d)
4 3( 3 + 1 4 3
and the problem is solved.
F

3
 J21. A (2m + 1) × (2n + 1) grid is colored with two colors. A 1 × 1
square is called row-dominant if there are at least n + 1 squares of its
color in its row. Define column-dominant squares in the same way. Prove
that there at least m + n + 1 both column-dominant and row-dominant
squares.
Proposed by Iurie Boreico, Moldova
Solution by Iurie Boreico, Moldova
Solution. In every row, there are at least n+1 row-dominant squares,
so at least (2m + 1)(n + 1) row-dominant squares in total. Analogously
there are at least (2n + 1)(m + 1) column-dominant squares in total. We
thus get at least (2m + 1)(n + 1) + (2n + 1)(m + 1) = (2m + 1)(2n +
1) + m + n + 1 row-dominant or column-dominant squares. There is a
total of (2m + 1)(2n + 1) squares and the exceed (of at least m + n + 1)
can come only from squares which are both row and column dominant.
Hence the conclusion follows.
Also solved by Ashay Burungale
F

J22. There are n 1’s written on a board. At each step

q we can select
a2 b2
two of the numbers in the board and replace them with 3 (a+b) . We keep
applying this operation until there is only one number left. Prove that
1
this number is not less than √ .
3
n
Proposed by Liubomir Chiriac, Princeton University
Solution by Iurie Boreico, Moldova
1
Solution. Instead of any number x on the blackboard write .
x3
Now we are working with this set of numbers. The transformation
r
3 a2 b2 p p
(a, b) → turns into (x, y) → 3 x2 y + 3 xy 2 ≤ x + y.
a+b
So the sum of the numbers does not increase. Initially we have the
sum of numbers is n so in the end we get a number which is at most n.
Thus after reversing our numbers
√ back the number left on the original
3
blackboard is then at least 1/ n.

4
 J23. Let ABCDEF be a hexagon with parallel opposite sides, and
let F C ∩ AB = X1 , F C ∩ ED = X2 , AD ∩ EF = Y1 , AD ∩ BC = Y2 ,
BE ∩ CD = Z1 , BE ∩ AF = Z2 . Prove that
a) If X1 , Y1 , Z1 are collinear then X2 , Y2 , Z2 are also collinear.
b) The lines X1 Y1 Z1 and X2 Y2 Z2 are parallel.

Proposed by Santiago Cuellar

Solution by Ivan Borsenco, University of Texas at Dallas

Solution. Denote AD ∩ BE = U , BE ∩ CF = V , CF ∩ AD = W .
Suppose X1 , Y1 , Z1 are collinear, then by Menelaos theorem applied to
the triangle U V W and the line through points X1 , Y1 , Z1 we have
X1 V Y1 W Z1 U
· · =1
X1 W Y1 U Z1 V
Also because ABCDEF is a hexagon with parallel opposite sides we
can use similarity of triangles and get
X1 V BV BC X1 W AW AF
= = , = = ,
X2 V VE EF X2 W WD FD
Y1 W FW AF Y1 U EU DE
= = , = = ,
Y2 W WC CD Y2 U BU AB
Z1 U DU DE Z1 V CV BC
= = , = = .
Z2 U AU AB Z2 V VF EF
Replacing the values into first relation we easily obtain
X2 V Y2 W Z2 U
· · =1
X2 W Y2 U Z2 V
Thus we proved that if X1 , Y1 , Z1 are collinear then X2 , Y2 , Z2 are also
collinear.
X1 W Y1 W
b) It enough to prove = . Using similarity we have
X2 W Y2 W
X1 W AW AF Y1 W FW AF
= = , = = .
X2 W WD CD Y2 W CW CD
and we are done.

5
 J24. Consider a triangle ABC and a point M inside the triangle.
Denote da , db , dc the distances from the point M to the triangle’s sides.
Prove that
1 1 1 1
2S( + + − ) ≥ da + db + dc
a b c R
where S, R are triangle’s area and circumradius.
Proposed by Ivan Borsenco, University of Texas at Dallas

First solution by Iurie Boreico, Moldova

First solution. Suppose a ≤ b ≤ c. Summing up areas for triangles

AM B, BM C, CM A we get 2S = ada + bdb + cdc ≥ a(da + db + dc )
2S
or ≥ da + db + dc . Thus it is enough to prove
a
1 1 1 1 1 1
2S( + − ) ≥ 0 or + ≥ . But this is easy
b c R b c R
1 1 1 1 1
+ ≥ + = and we are done.
2R sin b 2R sin c 2R 2R R

Second solution by Ivan Borsenco, University of Texas at Dallas

Second solution. First of all we recall Erdos theorem:

AM + BM + CM ≥ 2(da + db + dc )
A1 B1 A1 B1 · 2R
Denote projections as A1 , B1 , C1 , then = = CM .
sinγ c
Also using triangle inequality for the triangle A1 M B1 we get
da + db CM
da + db ≥ A1 B1 or ≥ .
c 2R
Summing for AM, BM, CM :
da + db db + dc da + dc AM + BM + CM da + db + dc
+ + ≥ ≥
c a b 2R R
1 1 1 1 da db dc
which we rewrite as (da + db + dc )( + + − ) ≥ ( + + )
a b c R a b c

6
Using ada + bdb + cdc = 2S and Cauchy-Schwartz inequality we get
da db dc
(ada + bdb + cdc )( + + ) ≥ (da + db + dc )2
a b c
Putting in the previous one
1 1 1 1 (da + db + dc )2
(da + db + dc )( + + − ) ≥
a b c R 2S

1 1 1 1
2S( + + − ) ≥ (da + db + dc )
a b c R
and our problem is solved.

7
 Seniors

S19. Let ABC be a scalene triangle. A point P is called nice if

AD, BE, CF are concurrent, where D, E, F are the projections of P on
the respective sides of ABC. Find how many nice points lie on the line
OI.
Proposed by Iurie Boreico, Moldova and Ivan Borsenco, UTD

Solution by Iurie Boreico, Moldova

Solution. Let x1 , x2 be the length of the projectiones of BP onto

BC and CP onto BC respectively; being positive iff D, C are on the
same side of B. Define analogously y1 , y2 , z1 , z2 . Then we have
x1 + x2 = a, y1 + y2 = b, z1 + z2 = c
A characteristic property of the line OI is that
x1 + y1 + z1 = x2 + y2 + z2 = u
for any point on OI (the converse is also true): this trivially holds for
O, I hence for any point on line OI satisfies this property by linearity of
projection. Also from Carnot Lemma we have x21 + y12 + z12 = x22 + y22 + z22
which since x1 + x2 + x3 = y1 + y2 + y3 is equivalent to
x1 y1 + y1 z1 + z1 x1 = x2 y2 + y2 z2 + z2 x2 = v

Now by Ceva Theorem AD, BE, CF concur if and only if

x1 y1 z1 = x2 y2 z2 = w
Both triples (x1 , y1 , z1 ) and (x2 , y2 , z2 ) are solutions for polynomial
of degree 3: x3 − ux2 + vx − w = 0. Thus x1 , y1 , z1 is a permutation
of x2 , y2 , z2 . If x1 = x2 then P is the intersection of the perpendicular
bisector of BC and OI hence P = O. Analogously if y1 = y2 or z1 = z2 .
If x1 = y2 x2 6= y1 since a = x1 + x2 6= y1 + y2 = b so x2 = z1 and
z2 = y1 . Analogously if x2 = y1 we get x1 = z2 and z1 = y2 . This gives
us two more points, which are I and J, where J is the intersection of the
perpendiculars from the centers of the exinscribed circles of ABC onto
the corresponding sides. Hence, the total number of nice points is 3.

8
 S20. Let ABC be an acute triangle and P a point inside this triangle.
Prove that if a = BC, b = CA and c = AB, then the following inequality
holds:
√
(AP + BP + CP )2 ≥ 3(aP A + bP B + cP C)
Proposed by Khoa Lu Nguyen, M.I.T

Solution by Hung Quang Tran, Hanoi National University

Solution. Let ∠BP C = x, ∠CP A = y, ∠AP B = z, x + y + z = 2π.

Using Cosine Theorem we have a2 = P B 2 + P C 2 − 2P BP C cos x

a2 P A = P A(P B 2 + P C 2 ) − 2P AP BP C cos x
Analogously we express b2 P B and c2 P C. Summing them and using
3
famous inequality cos x + cos y + cos z ≥ − for x + y + z = 2π we get
2
a2 P A + b2 P B + c2 P C = P A(P B 2 + P C 2 ) + P B(P A2 + P C 2 )+

+P C(P A2 + P B 2 ) − 2P AP BP C(cos x + cos y + cos z) ≤

P A(P B 2 + P C 2 ) + P B(P A2 + P C 2 ) + P C(P A2 + P B 2 ) + 3P AP BP C =
1
= (P A+P B +P C)(P A·P B +P B ·P C +P A·P C) ≤ (P A+P B +P C)3
3
Thus we have (P A + P B + P C) ≥ 3(a P A + b P B + c2 P C).
3 2 2

Multiplying by (P A+P B +P C) and using Cauchy-Schwartz inequal-

ity we obtain

(P A + P B + P C)4 ≥ 3(a2 P A + b2 P B + c2 P C)(P A + P B + P C) ≥

≥ 3(aP A + bP B + cP C)2
√
which is equivalent to (P A + P B + P C)2 ≥ 3(aP A + bP B + cP C)2
and the problem is solved.

9
 S21. Let p be a prime number and let a1 , a2 , . . . , an be distinct positive
integers not exceeding p − 1. Suppose that

p|ak1 + ak2 + . . . + akn

for k = 1, 2, . . . , p − 2. Find {a1 , a2 , . . . , an }.

Proposed by Pascual Restrepo Mesa, Universidad de los Andes, Colom-
bia

First solution by David E. Narvaez, Technological University of Panama

First solution. Let g be a primitive root modulo p. Since 1 ≤ ai ≤

p − 1, p 6 |ai , for i = 1, 2, . . . , n and there is a positive integer αi such that
ai ≡ g αi ( mod p). Therefore, the condition

p|ak1 + ak2 + . . . + akn

can then be written as

ak1 + ak2 + . . . + akn ≡ (g α1 )k + (g α2 )k + . . . + (g αn )k ( mod p)

α1
α2
αn
0 ≡ gk + gk + . . . + gk ( mod p)

Notice that g k , for k = 1, 2, . . . , p−2 generates the set {2, 3, . . . , p − 1},

therefore the polynomial

P (x) = xα1 + xα2 + . . . + xαn

has p − 2 distinct roots modulo p, which are 2, 3, . . . , p − 1, thus we can

write

P (x) ≡ (x − 2) (x − 3) · · · (x − (p − 1)) ( mod p)

xp−1 − 1
x α1 + x α2 + . . . + x αn ≡ ( mod p)
x−1
x α1 + x α2 + . . . + xαn ≡ xp−2 + xp−3 + . . . + 1

From this we can easily deduce that {α1 , α2 , . . . , αn } = {0, 1, . . . , p − 2},

and {a1 , a2 , . . . , an } = {g α1 , g α2 , . . . , g αn } = {1, 2, . . . , p − 1}.

10
 Second solution by Ashay Burungale

Second solution. First of all we need the following lemma:

Lemma. Let Sk = k 1 + k 2 + ... + k p−2 , then

p|Sk + 1 if k ≡
6 0, 1( mod p) and p|Sk + 2 if k ≡ 1( mod p).

Proof. Clearly if k ≡ 1( mod p) then

Sk = k 1 + k 2 + ... + k p−2 ≡ p − 2 ≡ −2( mod p).

k p−1 − k
Suppose k 6≡ 1( mod p) then Sk = k(k p−2 − 1)/k − 1, or Sk = .
k−1
From the Fermat Theorem we have k p−1 ≡ 1( mod p), so

k p−1 − k −(k − 1)
Sk = ≡ ≡ −1( mod p).
k−1 k−1
Thus Sk ≡ −1( mod p) for k 6≡ 0, 1( mod p), lemma is proven.
p−2
X
Solution. From the problem statement we have p| ai1 +ai2 +. . .+ain
i=1
Since they all are distinct and positive integers we have at most one
of them is equal to 1 and all other ai 6≡ 0, 1( mod p), thus using our
lemma
p−2
X
ai1 + ai2 + . . . + ain ≡ (−1) · n( mod p) = −n(mod p) or
i=1
p−2
X
ai1 + ai2 + . . . + ain ≡ (−2) + (−1) · (n − 1) = −n − 1(mod p).
i=1
It follows that p| − n or p| − (n + 1) and we know from the problem
statement that n ≤ p − 1. Hence the only one possibility is n = p − 1
which gives us the solution {a1 , a2 , ..., an } ={1, 2, ..., p − 1}, which clearly
satisfies the condition.

11
 S22. Let n, k be positive integers. Eve gives to Adam k apples. How-
ever she can first give him some bitter apples, at most n. The procedure
goes as follows: Eve gives to Adam the apples one by one, and he can
either eat it (and find out whether it’s sweet or not), or throw it. Adam
knows that the bitter apples come first, and the sweet come last. Find,
in terms of n, the least value of k for which Adam can ensure he eats
more sweet apples than bitter ones.
Proposed by Iurie Boreico, Moldova

Solution by Iurie Boreico, Moldova

Solution. It’s clear that if an apple is sweet, then all the apples
coming after it are sweet, so Adam can surely eat all the apples he takes
after encountering one sweet one. If the apple is bitter, then all the k
sweet apples are yet to come.
We prove that Adam can ensure he eats more sweet apples then bitter
ones if and only if n < k(k+3) 2
. For these values of n we can show the
following strategy: let Adam eat the apple k. If it’s sweet, he won.
Otherwise let him it the apple 2k − 1. If it’s sweet, then there is at
least one more sweet apple coming next, as there are just k − 2 apples
between k and 2k − 1. So he has eaten two sweet apples and one bitter.
At the next step Adam will eat the apple k + (k − 1) + (k − 2) and so
on. Reasoning analogously, we see that if Adam eats at least one sweet
apple he won. However he will eat at least one sweet apple: if not, then
he will have eaten the apples k, k + (k − 1), . . . , k(k+1)
2
, and the last apple
k(k+3)
for sure is sweet otherwise there will be at least 2 > n apples.
Assume that Adam decides to eat the apple numbered a1 . Then
a1 ≤ k otherwise it could be possible that the first k apples were sweet
so he would miss all of them. If n > 2k − 1 ≥ a1 + k − 1 then the apple
numbered a1 could be bitter, so Adam could be forced to eat another
apple, numbered a2 . Then a2 ≤ a1 + k − 1 because otherwise it could
be possible that the apples numbered a2 − k + 1, . . . , a2 were sweet, and
Adam would eat only at most one of them (namely a2 ), and with already
one bitter apple eaten, he would fail. So a2 ≤ 2k − 1. The continuation
is now by induction: we prove that if n >≥ f rack(k + 3)2 for l < k
then Adam is forced to eat at least l possibly bitter apples a1 , a2 , . . . , al
with ai+1 − ai ≤ k − i (so al < n − k + l). The base was just proven,
and for the induction step notice that if n ≥ k(k+3) 2
≥ al + k and Adam
ate the bitter apples a1 , . . . , al then he would be forced to eat one more
apple al+1 . Then al+1 ≤ al + k − l, otherwise it could be possible that
the apples al+1 − k + l, . . . , al+1 + l − 1 were the last and sweet apples,

12
and Adam could only eat l of them, starting from a2 which is not more
than the number of bitter ones already consumed. The induction step
is complete. Now by setting l = k Adam has to eat at least k apples
which can all be bitter. Since he can eat at most k sweet apples, he fails.
Therefore n ≤ k(k+1)
2
.

13
 S23. Let a, b, c, d be positive numbers, prove the following inequality

3(a2 − ab + b2 )(c2 − cd + d2 ) ≥ (a2 c2 − abcd + b2 d2 )

Proposed by Dr. Titu Andreescu, University of Texas at Dallas

Solution by Bin Zhao

Solution. We observe that

(a2 c2 − abcd + b2 d2 ) − (a2 − ab + b2 )(c2 − cd + d2 ) = (ad + bc)(a − b)(c − d)

Thus the inequality is equivalent to:

(a2 − ab + b2 )(c2 − cd + d2 ) ≥ 2(ad + bc)(a − b)(c − d)

If a = b = 0 or c = d = 0 the inequality is obviously true. Now

let us ignore this case. Without loss of generality suppose a 6= 0, c 6= 0
(analogously for a 6= 0, d 6= 0). Dividing the above inequality by a2 c2
b d
both sides and substituting x = , y = , we get the following inequality
a c
(x2 − x + 1)(y 2 − y + 1) ≥ 2(x + y)(x − 1)(y − 1)

which is equivalent to

(y 2 − 3y + 3)x2 − (3y 2 − 5y + 3)x + 3y 2 − 3y + 1 ≥ 0

The left side is a binomial of x,we have the discriminant:

∆ = (3y 2 − 5y + 3)2 − 4(y 2 − 3y + 3)(3y 2 − 3y + 1) =

= −3y 4 + 18y 3 − 33y 2 + 18y − 3 = −3(y 2 − 3y + 1)2 ≤ 0

So we have:

(y 2 − 3y + 3)x2 − (3y 2 − 5y + 3)x + 3y 2 − 3y + 1 ≥ 0

Thus we prove the original inequality. And the inequality hold when:
√ √ √ √
3+ 5 3+ 5 3+ 5 3+ 5
a= b, c = d or b = a,d= c.
2 2 2 2

14
 S24. Let ABC be an acute-angled triangle inscribed in a circle C. One
considers all equilateral triangles DEF with vertices on C. The Simpson
lines of D, E, F with respect to the triangle ABC form a triangle T . Find
the greatest possible area of the triangle T .
Proposed by Iurie Boreico, Moldova and Ivan Borsenco, UTD

Solution by Iurie Boreico, Moldova

Solution. We will firstly prove that T is equilateral. Indeed, if

D ∈ BC let DA , DB , DC be its projections on the sides of ABC. WLOG
DB ∈ (AC), B ∈ (ADC ). Then the angle between the Simpson line of D
and BC is m(∠DC DA B) = m(DC DB) = 12 m(DBA) − 90 = 12 m(DC) +
B − 90. Proceeding analogously for E, F we can conclude that the angle
between the Simpson lines of two points is actually half of the measure
of arc between these two points. Particularly, since DEF is equilateral,
so is T .
Another fact that we will use the well-known fact that the Simpson
line lD of the point D passes through the midpoint M od the segment
[DH], where H is the orthocenter of ABC. Indeed, using the notations
and the situation analyzed above let H 0 be the symmetric of H wrt BC
0
that lies on C from Hamilton Theorem and let DA be the symmetric
of D wrt BC. Since lD passes through DA , to prove it passes through
M is equivalent to proving that the angle between lD and DDA equals
0
the angle ∠M DA D ≡ ∠HDA D ≡ ∠HH 0 D as HH 0 DDA 0
is an isosceles
trapezoid. However this results immediately from angle chasing, since
we know the angle between lD and BC.
Now, let’s pass to the solution of the problem. Since T is equilateral,
we must find the largest possible value of the inradius of T. However the
Euler circle of ABC clearly contains the midpoints M, N, P of segments
[HD], [HE], [HF ] and so intersects the sides of the triangle T. It easily
follows now that its radius is not less than the inradius of T. Indeed, if
T = U V W and J is the midpoint of [OH] then we get d(J, U V ) ≤ R2 ),
where R is the radius of C so [JU V ] ≤ U V4 R . Summing this with the
analogous relations for V W, U W we get

pU V W rU V W = S[U V W ] ≤ S[JU V ] + S[JU W ] + S[JV W ] ≤

R(U V + V W + U W ) R
= pU V W
4 2
R
so rU V W ≤ 2
.

15
 Now we shall find a triangle U V W with rU V W = R2 . To do this, we
must ensure that the Euler circle of ABC touches the Simpson lines of
lD , lE , lF . Let’s effectively compute the positions of D, E, F that satisfy
this condition. Suppose D belongs to the small arc BC. As lD passes
through M , it is tangent iff lD ⊥M J or lD ⊥DO. The angle between DO
and BC is 180 − m(∠ODC) − m(∠BCD) = 180 − (90 − 21 m(CD)) −
1
2
m(BD) = 90+ 12 (m(CD)−m(BD)). However the angle between lD and
BC is, as computed above 12 m(DC) + B − 90. So DO⊥lD iff m(CD) −
1
2
m(BD) + B = 90. As m(BD) = 2A − m(CD), this is possible if and
only if m(CD) = 2(90+A−B) 3
and so the position of D on the arc BC is
uniquely determined (and since the triangle is acute-angled, we can check
that D indeed lies on the arc). Analogously if we let E ∈ AC we get
m(CE) = 2(90+B−A) 3
, thus m(DE) = m(arcCD) + m(CE = 120). So we
conclude that the triangle DEF √ that we seek is indeed equilateral. Its
R 3 3R2
inradius is so its area is , and this is the maximal possible area.
2 4

16
 Undergraduate

U19. Let f0 be a real-valued function, Zcontinuous on the interval [0, 1]

x
and for each integer n ≥ 0 let fn+1 (x) = fn (t)dt. Suppose that there
0
1
is a positive integer k with the property that fk (1) = . Prove
(k + 1)!
that there exists x0 such that f0 (x0 ) = x0 .

Proposed by Dr. Titu Andreescu, University of Texas at Dallas

No solutions proposed.

U20. Prove that one cannot find an entire function f such that
f (f (x)) = ex for all real number x, but can find an infinitely many
times differentiable function with this property.
Proposed by Gabriel Dospinescu, Ecole Normale Superieure,Paris

No solutions proposed.

17
 U21. Evaluate Z 1   
1 1
dx
0 x 1−x
where {x} is the fractional part of x.
Proposed by Ovidiu Furdui, Western Michigan University

Solution by Bin Zhao

Solution. The answer is 2γ − 1,where γ is the Euler’s Constant.

First we will prove:

Z 1   
2 1 1 1
dx = γ −
0 x 1−x 2
This is because
Z 1    +∞ Z 1    
2 1 1 X n 1 1
dx = dx
0 x 1−x 1
n=2 n+1
x 1−x

Set Z 1   
n 1 1
an = dx
1
n+1
x 1−x
because for n ≥ 2, we have
   
1 1 1 1 x
= − n, = −1= ,
x x 1−x 1−x 1−x
1 1
when < x < . From here
n+1 n
1    Z 1 Z 1  
n−1
Z
n 1 1 n 1 x n
an = dx = ( −n) dx = n− dx
1
n+1
x 1−x 1
n+1
x 1−x 1
n+1
1−x
 
1 1 1
= − (n − 1) ln(1 − ) − ln(1 − ) =
n+1 n+1 n
n2
 
1
= − (n − 1) ln .
n+1 (n − 1)(n + 1)

18
Thus
Z 1   +∞ 
n2
  
2 1 1 X 1
dx = − (n − 1) ln
0 x 1−x n=2
n+1 (n − 1)(n + 1)

but
N
n2
  
X 1
SN = − (n − 1) ln =
n=2
n+1 (n − 1)(n + 1)
N N
n2
 
X 1 X
= − (n − 1) ln =
n=2
n + 1 n=2 (n − 1)(n + 1)
N N
n2n−2
 
X 1 X
= − ln =
n=2
n + 1 n=2 (n − 1)n−1 (n + 1)n−1

N
X 1 22 · 34 · · · N 2N −2
= − ln
n=2
n+1 (1 · 3)(22 · 42 ) · · · ((N − 1)N −1 (N + 1)N −1 )

N N N −1
NN

X 1 X 1 N
= − ln = − ln − ln N
n=2
n+1 (N + 1)N −1 n=2
n+1 N +1

With the well-known formula

 
1 n 1 1
lim (1 + ) = e, lim 1 + + · · · + − ln n = γ
n→+∞ n n→+∞ 2 n
We have  Z 1 
1 2 1 1
dx = lim SN = γ − .
0 x 1−x N →+∞ 2
But set t = 1 − x,we have
Z 1   Z 1   
1 1 2 1 1
dx = dx
1
2
x 1−x 0 x 1−x
Finally Z 1   
1 1
dx = 2γ − 1
0 x 1−x

19
 U22. Let ||.|| be a norm in Cn and define
|||A||| = sup ||Ax||
||x||≤1

for any complex matrix A in Mn (C). Let a < 2 and G be a subgroup of

GLn (C) such that for A ∈ G one has |||A − In ||| ≤ a. Prove that G is
finite.
Proposed by Gabriel Dospinescu and Alexander Thiery

Solution by Gabriel Dospinescu, Ecole Normale Superieure, Paris

Solution. Consider A ∈ G. Then for all integers k we know that

|||Ak −In ||| ≤ a. Let λ be an eigenvalue of A and let x be a proper vector.
Then (Ak − In )x = (λk − 1)x and so |λk − 1| ≤ a < 2 for all integers k.
Thus |λ|k < 3 for all integers k, which implies that |λ| = 1. Thus λ = eiπr
for some real number r. It easily follows from the previous inequality that
2
cos(πkr) ≥ 1 − a2 > −1 for all integers k. Using Kronecker’s theorem,
we deduce that r must be rational and then a simple argument allows to
prove that the denominator of r is uniformly bounded in terms of a only.
This shows the existence of a number N not depending on A ∈ G such
that λN = 1 for all A ∈ G and all λ an eigenvalue of A.

Fix now A ∈ G and write AN = In + B, where B is nilpotent(by

the previous arguments, all eigenvalues of B are zero). We know that
|||AN p − I
n ||| < 2
for all p. But
using the binomial formula, this im-
plies ||| p1 B + p2 B 2 + ... + n−1 p
B n−1 ||| < 2 for all p (we used here
n
the fact that B = On because B is nilpotent). Suppose that B is not
zero and take j ≥ 1 the least positive integer for which B j 6= O. Then

|
2 > pj ||B j ||| − j−1 p
|||B j−1 ||| − ... − p1 |||B|||, which is impossible for

all positive integers p ( just divide by pj and take the limit for p → ∞).
Thus B = 0.

We have established the existence of a positive integer N such that

N
A = In for all A ∈ G. By Burnside’s theorem, any subgroup of GLn (C)
of finite exponent is finite. Therefore the group is finite. However, we
will present a proof of Burnside’s theorem: consider A1 , A2 , ..., Ar ∈ G
a basis of the linear space spanned by the elements of G, with r ≤ n2 .
We claim that the function f (X) = (tr(A1 X), tr(A2 X), ..., tr(Ar X)) is
injective when restricted to G. Supposing the contrary, let A, B ∈ G
such that f (A) = f (B). Because A1 , ..., Ar generate the linear space
spanned by G, it follows that tr(AX) = tr(BX) for all X ∈ G. Thus if

20
U = AB −1 we have tr(U X) = tr(X) for all X ∈ G. Thus tr(U ) = n and
also U N = In . This implies that U = In , because tr(U ) is the sum of the
eigenvalues of U , which are roots of unity. If their sum if n, it means that
all of them are 1, so U is unipotent. Because U is also diagonalisable in
Mn (C), it follows that U = In , which proves the claim. Now, the proof
is finished, because clearly the image of f is finite: all tr(X) with X ∈ G
are among the collection of sums of n roots of unity of order N , thus in
a finite set. This shows that G is finite.

Remark
If a = 2, the statement is false, because for the euclidean norm we can
take the set of unitary matrices, which is an infinite group with all de-
sired properties. However, even in this case some strong restrictions are
imposed. First of all, it is clear that such subgroups must be bounded.
Thus their closure is also bounded. It follows that the closure of any such
subgroup is a compact subgroup of GLn (C). It can be shown (the proof
is not trivial at all) that such subgroups are conjugated to subgroups of
the unitary group.

U23. Find the following sum

n−1
X 1
k=0
1 + 8sin2 ( kπ
n
)

Proposed by Dorin Andrica and Mihai Piticari

No solutions proposed.

21
 U24. Find all linear maps f : Mn (C) → Mn (C) such that for any
A ∈ Mn (C) we have f (Ak ) = f k (A) for some integer k > 1
Proposed by Gabriel Dospinescu, Ecole Normale Superieure,Paris
Solution by Gabriel Dospinescu, Ecole Normale Superieure, Paris
Solution. First of all, let us prove that f (Ar ) = f r (A) for all positive
integers r and all A. Fix an A and observe that f ((A + xIn )k ) = f (A +
xIn )k = (f (A) + xIn )k for all complex x. Using the binomial formula and
the linearity of f , we obtain
   
k k k−1 k
f (A ) + xf (A ) + ... + xk−1 f (A) + xk In
1 k−1
 
k
= f (A)k + xf (A)k−1 + ... + xk In .
1
By identifying coefficients in the last equality we deduce that f (Ar ) =
f r (A) for all r ≤ k. In particular, f (A2 ) = f (A)2 . Thus f (A2 + AB +
BA + B 2 ) = f (A)2 + f (A)f (B) + f (B)f (A) + f (B)2 . This implies
f (AB) + f (BA) = f (A)f (B) + f (B)f (A).
By taking B = Ap in this last relation, we find 2f (Ap+1 ) = f (A)f (Ap )+
f (Ap )f (A), which allows an immediate proof by induction of the equality
f (Ar ) = f (A)r for all r. Observe that f is continuous (being linear in a
finite dimensional space), so we can write
A2 f (A)2
f (eA ) = f (In + A + + ...) = In + f (A) + + ... = ef (A)
2! 2!
by the previous result. Now, take A any invertible matrix with complex
entries. A classical result asserts that there exists a matrix B such that
A = eB . Thus f (A) = ef (B) , that is f (A) is also invertible.

Despite all our efforts, we could not find an elementary continua-

tion of these arguments, so we are obliged to use a deep result of Mar-
cus and Purves (1959) saying that any linear map preserving invert-
ibility and such that f (In ) = In is of the form f (X) = BXB −1 or
f (X) = BX T B −1 for some constant invertible matrix B. For a proof
we recommend the excellent book Problems and theorems in linear
algebra by V.Prasolov. Using this result, it is easy to conclude: the
above maps are all solutions of the problem.

22
 Olympiad
O19. Let a, b, c be positive numbers, prove the following inequalities
a) (a3 + b3 + c3 )2 ≥ (a4 + b4 + c4 )(ab + bc + ac)
b) 9(a4 + b4 + c4 )2 ≥ (a5 + b5 + c5 )(a + b + c)3
Proposed by Ivan Borsenco, University of Texas at Dallas

First solution by Daniel Campos Salas, Costa Rica.

First solution.From Lagrange’s identity we have that
X
(a3 + b3 + c3 )2 = (a4 + b4 + c4 )(a2 + b2 + c2 ) − a2 b2 (a − b)2 .
cyc

Therefore,

(a3 + b3 + c3 )2 − (a4 + b4 + c4 )(ab + bc + ac)

X
= (a4 + b4 + c4 )(a2 + b2 + c2 − ab − bc − ac) − a2 b2 (a − b)2
cyc
X1
= (a4 + b4 + c4 )(a − b)2 − a2 b2 (a − b)2
cyc
2
1X
(a2 − b2 )2 + c4 (a − b)2 ≥ 0,


=
2 cyc

and we have proved the first inequality.

Again, from Lagrange’s identity we have that

X
(a4 + b4 + c4 )2 = (a5 + b5 + c5 )(a3 + b3 + c3 ) − a3 b3 (a − b)2 .
cyc

Note also the identity

X
9(a3 + b3 + c3 ) − (a + b + c)3 = (4a + 4b + c)(a − b)2 .
cyc

23
 Then,

9(a4 + b4 + c4 )2 − (a5 + b5 + c5 )(a + b + c)3

X
= (a5 + b5 + c5 )(9(a3 + b3 + c3 ) − (a + b + c)3 ) − 9 a3 b3 (a − b)2
cyc
X
5 5 5 3 3 2


= (a + b + c )(4a + 4b + c) − 9a b (a − b)
cyc
X
4(a5 + b5 )(a + b) − 9a3 b3 (a − b)2


≥
cyc
X
≥ (16a3 b3 − 9a3 b3 )(a − b)2 ≥ 0,
cyc

and we’re done.

Second solution by Tung Nguyen Trong, Natural Science, Hanoi, Viet-

nam
Second solution. a) We consider the following inequality:
X
(a3 + b3 − ab2 − a2 b)2 ≥ 0 ↔
X
a6 + b6 + (a2 b4 + a4 b2 ) + 4a3 b3 − 2(a4 b2 + a2 b4 ) − 2(a5 b + b5 a) ≥ 0 ↔
X X X
a6 + b6 + 4a3 b3 ≥ (a4 b2 + a2 b4 ) − 2 (a5 b + b5 a)

Dividing both two sides by 2:

1X 4 2 X
(a3 + b3 + c3 )2 ≥ (a b + a2 b4 ) + (a5 b + b5 a)
2
By AM-GM we have:
X X X
(a4 b2 + a2 b4 ) = (a4 b2 + a4 c2 ) ≥ 2 a4 bc

So (a3 +b3 +c3 )2 ≥ a4 bc+

P 5
(a b+b5 a) = (a4 +b4 +c4 )(ab+bc+ca)
P

and we are done.

b) This time let us look at this inequality:

X
(2a4 + 2b4 − 2a2 b2 − ab3 − a3 b)≥ 0 ↔
X
4(a8 + b8 ) + 14a4 b4 + (a2 b6 + a6 b2 ) − 8(a6 b2 + a2 b6 ) − 4(a5 b3 + a3 b5 )
−4(a7 b + ab7 ) + 4(a5 b3 + a3 b5 ) ≥ 0 ↔
X X X
4(a8 + b8 ) + 14a4 b4 ≥ 7 (a6 b2 + a2 b6 ) + 4(a7 b + ab7 )

24
 It’s easy to see that:
X
4(a8 + b8 ) + 18a4 b4 = 9(a4 + b4 + c4 )2 − (a8 + b8 + c8 )
P
From here we will use the notation (.) inspire of cyclic for short! It
suffices to show that:
7(a6 b2 + a2 b6 ) + 4(a7 b + ab7 ) + 4(a4 b4 ) ≥ (a5 + b5 + c5 )(a + b + c)3 −
(a8 + b8 + c8 )(∗)

By some calculations we get:

RHS = (a5 b3 + a3 b5 ) + 3(a7 b + ab7 ) + 3(a6 b2 + a2 b6 ) + 3a5 (b2 c + c2 b) +
5
6(a abc)

So (*) is equivalent to:

4(a6 b2 + a2 b6 ) + (a7 b + ab7 ) + 4(a4 b4 ) ≥ (a5 b3 + a3 b5 ) + 3a5 (b2 c + c2 b) +
6
6(a bc) ↔ 8(a6 b2 + a2 b6 ) + 2(a7 b + ab7 ) + 8(a4 b4 ) ≥ 2(a5 b3 + a3 b5 ) +
6a5 (b2 c + c2 b) + 12(a6 bc)

By AM-GM we obtain:
X X
a6 b2 +a2 b6 +2a4 b4 ≥ 2 (a5 b3 +a3 b5 ) ↔ 4(a6 b2 +a2 b6 )+8(a4 b4 ) ≥
8(a5 b3 + a3 b5 ) ≥ 2(a5 b3 + a3 b5 ) + 6a5 (b2 c + c2 b)

According Muirhead’s Inequality: sym a5 b3 ≥ 5 2

P P
sym a b c. It re-
mains to prove that:
4(a6 b2 + a2 b6 ) + 2(a7 b + ab7 ) ≥ 12a6 bc

But it’s also true by Muirhead:

X X X X
a6 b 2 ≥ a6 bc and a7 b ≥ a6 bc
sym sym sym sym
and we are done.

25
 Third solution by Ashay Burungale

Third solution. According to EV -Theorem, it suffices to consider

the case b = c = 1.
First inequality becomes (a3 + 2)2 ≥ (a4 + 2)(2a + 1) or

a6 − 2a5 − a4 + 4a3 − 4a + 2 ≥ 0

(a − 1)2 (a4 − 2a2 + 2) ≥ 0

Second inequality becomes 9(a4 + 2)2 ≥ (a5 + 2)(a + 2)3 which trans-
forms after some work into

(a − 1)2 (4a6 + 5a5 − 9a3 + 8a + 10) ≥ 0

and we are done.

Finally we want to say that EV -Theorem is a powerful method which
we advice to find and use in proving symmetric inequalities.

26
 O20. In triangle ABC the incircle touches AC at E and BC at D.
The excircle(corresponding to A) touches the side BC at A1 and the
extensions of sides AB, AC at C1 and B1 , respectively. If DE ∩ A1 B1 =
{L}, prove that L lies on the circumcircle of ∆A1 BC1 .
Proposed by Liubomir Chiriac, Princeton University

First solution by David E. Narvaez, Technological University of Panama

First solution. If L = A1 , then the result is trivial. Otherwise, let a,

b and c be the lenghts of sides BC, AC and AB respectively, γ = ∠ACB,
β = ∠ABC, s = a+b+c 2
, IA the center of the excircle corresponding to A.
Finally, let l be the line perpendicular to BC passing through B and let
A0 = DE ∩ l. We claim that A1 A0 is a diameter of the circumcircle of
triangle A1 BC1 .

Since CA1 = CB1 = s − b, ∠CB1 A1 = π−∠A21 CB1 = π−(π−γ)2

= γ2 , and
B1 A1 is then parallel to the bisector of angle ∠ACB, which is perpendic-
ular to line DE. Therefore ∠A1 LA0 = 90, and since AA0 is a diameter
of the circumcircle of A1 BC1 , L belongs to the circumference.

To prove our claim, consider that l is parallel to line IA A1 , and

π−γ s−b CA1
BA0 = BD · tan = γ = = IA A1
2 tan 2 tan γ2

BA0 IA A1 is a rectangle, which proves our claim.

Second solution by Daniel Campos Salas, Costa Rica.

Second solution.

Let α, β, θ be the angles A, B, C, respectively, and let F = A1 B1 ∩AB.

Note that L can lie inside or outside ABC. In the first case, for proving
β
that C1 BLA1 is cyclic, we should prove that ∠BLF = ∠A1 C1 B = .
2
For the second case, proving that C1 BA1 L is cyclic is equivalent to prove
β
that ∠BLF = ∠A1 C1 B = . In both cases we have to prove that
2
β θ
∠BLF = . Since 4AB1 C1 is isosceles, and ∠A1 B1 C = it follows
2 2
β
that ∠C1 B1 F = . Then, the problem reduces to prove that BL k B1 C1 .
2

27
 θ
Note that ∠BA1 F = , then by the sine law applied to 4A1 BF
2
θ
FB sin
implies that =  2  . From the Mollweide’s formulas we
BA1 θ
sin +α
2
θ
sin c
have that  2  = , and since BA1 = BC1 it follows that
θ a+b
sin +α
2
FB c
= , from where we conclude that
BC1 a+b
F C1 FB 2s
=1+ = . (1)
BC1 BC1 a+b
Since AB1 = s, from the sine law on 4AB1 F it follows that
s sin α s sin α
F B1 =    =  . (2)
θ θ
sin 180 − α + sin α +
2 2
 
θ θ
Since ∠B1 EL + ∠EB1 L = 90 − + = 90, then DE is
2 2
θ
perpendicular to B1 L. Since B1 E = a, it follows that LB1 = a cos .
2
(3)
From (2) and (3), the Mollweide’s formulas, the double-angle formula,
and the sine law it follows that
F B1 s sin α 2sc sin α 2s
=  = = . (4)
LB1 θ θ a(a + b) sin θ a+b
a cos sin α +
2 2

From (1) and (4) it follows that BL is parallel to B1 C1 , and the

conclusion follows.

Also solved by Ivan Borsenco

28
 O21. Let p be a prime number. Find the smallest degree of a polyno-
mial f with integer coefficients such that the numbers f (0), f (1)...f (p−1)
are perfect (p − 1)th powers.
Proposed by Pascual Restrepo Mesa

Solution by Iurie Boreico, Moldova

Solution. Assume that the degree of the polynomial is less than

p − 1. Let’s work in Z[p]. We know that 0k + 1k + . . . + (p − 1)k = 0,
therefore f (0) + . . . + f (p − 1) = 0. However f (x) = 0, 1 in Z[p], and
this implies either f (x) = 0 for all x = 0, 1, . . . , p − 1 or f (x) = 1 for all
x = 0, 1, . . . , p − 1. So either f (x) or f (x) − 1 is zero in Zp impossible.
Therefore the degree of f is at least p − 1 and it’s clear it can be p − 1,
so this is the answer.

29
 O22. Consider a triangle ABC and two points P, Q in it’s plane. Let
A1 , B1 , C1 and A2 , B2 , C2 be the cevians in our triangle. Denote U, V, W
as second intersections of circles (AA1 A2 ), (BB1 B2 ), (CC1 C2 ) with the
circumcircle of the triangle ABC. Let X be the intersection of AU with
BC, similarly define Y for BV and AC and Z for CW and AB. Prove
that X, Y, Z are collinear.
Proposed by Khoa Lu Nguyen, MIT and Ivan Borsenco, University
of Texas at Dallas

Solution by Khoa Lu Nguyen, MIT and Ivan Borsenco, University of

Texas at Dallas

Solution. From the Power of a Point Theorem applied to the circles

(ABC) and (AA1 A2 ) and point X we get

XB · XC = XA · XU = XA1 · XA2

Suppose without loss of generality that B is between X and C, A1 is

between B and A2 . Let us denote XB = x, BA1 = u, A1 A2 = v, A2 C =
w then the upper relationship is equivalent to

x(x + u + v + w) = (x + u)(x + u + v).

u(u + v) XB
From where we get x = . Now let us calculate .
w−u XC
XB x u(u + v) BA1 · BA2
= = =
XC x+u+v+w w(v + w) BA2 · CA2
Thus we can use Menelaos Theorem for points X, Y, Z with respect
to the triangle ABC to prove that they are collinear.
Y XB Y BA1 · BA2
= =1
XC BA2 · CA2
Last equality is due Ceva theorem applied for P and Q, and we are
done.

30
 O23. Let ABC be a triangle and let A1 , B1 , C1 be the points where
the angle bisectors of A, B and C meet the circumcircle of triangle ABC
respectively. Let Ma be the midpoint of the segment connecting the
intersections of segments A1 B1 and A1 C1 with segment BC. Define Mb
and Mc analogously. Prove that AMa , BMb , and CMc are concurrent if
and only if ABC is isosceles.
Proposed by Dr.Zuming Feng, Philips Exeter Academy

Solution by Daniel Campos Salas, Costa Rica.

Solution. Let I be the incenter of triangle ABC and let P = BC ∩
A1 C1 , Q = BC ∩ A1 B1 , R = AB ∩ A1 C1 , and S = CC1 ∩ AB. Note that
∠A + ∠C ∠A + ∠C
∠RP B = ∠CP A1 = and ∠P RB = ∠ARC1 = .
2 2
Then, 4BP R is isosceles and since BI is the angle bisector it follows
that B1 I is perpendicular to A1 C1 and the segment P R is bisected by
BI. Analogously, A1 I and C1 I are perpendicular to B1 C1 and A1 B1
respectively. Then, I is the orthocenter of A1 B1 C1 . Using the well-known
fact that the reflection of the orthocenter on the sides of the triangles
lies on its circumcircle and that BI is perpendicular to A0 C 0 , it follows
that P R bisects BI. Since P R and BI bisect each other, it follows that
BP IR is a parallelogram, then IP is parallel to AB.
CS 2s
It is well-known that = , and since IP is parallel to AB it
IS c
BC a 2s ac
follows that = = . It follows that P B = . Analogously,
PB PB c 2s
2
ab a ac a2 a(a + 2c)
QC = . Then, P Q = . It follows that BMa = + = .
2s 2s 2s 4s 4s
a(a + 2b) BMa a + 2c
Analogously, CMa = , then = . Analogously,
4s CMa a + 2b
CMb b + 2a AMc c + 2b
= and = .
AMb b + 2c BMc c + 2a
BMa · CMb · AMc (a + 2c)(b + 2a)(c + 2b)
It follows that = . By
CMa · AMb · BMc (a + 2b)(b + 2c)(c + 2a)
Ceva’s theorem we have that AMa , BMb , and CMc are concurrent if
(a + 2c)(b + 2a)(c + 2b)
and only if = 1, which holds if and only if
(a + 2b)(b + 2c)(c + 2a)
a2 b + b2 c + c2 a − ab2 + bc2 + ca2 = (a − b)(a − c)(b − c) = 0, which implies
the desired result.

31
 O24. Find all integers a, b, c such that

2n a + b|cn + 1

for every positive integer n.

Proposed by Gabriel Dospinescu, Ecole Normale Superieure, Paris

Solution by Iurie Boreico, Moldova

Solution. We need the following lemma:

Lemma: Let u1 , u2 , . . . un be different non-zero reals. If zm = ni=1 ci um

P
i =
0 for all sufficiently big m then ci = 0.
Proof: Consider the generating function Z(X) = ∞ i
P
i=0 zi x . If zm =
0 for all sufficient big n then Q Z is a polynomial.
P Q However Z(X) =
P n ci
i=0 X−ui therefore Z(X) (X − u i ) = ci j6=i (X − uj ). And if
ck 6= 0 then the left-hand side, as a Q polynomial, is divisible by X − uk
while all terms in the RHS except ck i6=k (X − ui ) not. Contradiction.
cn + 1
Firstly we prove that c is a power of 2. Let xn = n .
a2 + b
Assume that 2m−1 < c < 2m . Set
a 1 1 1
yn = (cn + 1) (1 − n a + 2 ± ... ± m ).
b 2 b 2n
2 b2a 2 abm
mn

m
(−1)
1
By using the identity 1+x − ( x1 − x12 ± . . . ± xm−1 1
) = xm (1+x)
we deduce
n
c +1
that |xn − yn | = mn am (2m a + b) → 0 when n → ∞.
2 bm
Now yn satisfy a polynomial recurrence with P rational coefficient, so
there are rational constants c1 , c2 , . . . , cr with ci yn+i = 0. We may as-
sume
P that c are
iP integers or elese multiply by a suitable integer. Therefore
ci xn+i = ci (xn+i −yn+i ) → 0. As this number is certainly an integer,
we conclude that it’s eventually zero, so xi startingP from some point
Pr satify
r n
the same recurrence. We deduce then that xn = i=0 ki 2ni + i=0 li 2cni 1

for all sufficiently big n. Hence cn + 1 = xn (2n a + b), so cn + 1 − xn (2n a +

2i
b) = −al0 (2c)n + cn (1 − bl0 − al1 ) − ri=1 (bli + ali+1 )( 1c )n + (1 − ak0 ) −
P
Pr 1 n c 1
i=0 (bki + aki+1 )( 2i ) = 0. As 2c, c, 2 , . . . , 1, 2 , . . . are distinct, applying
the lemma we deduce that all the coefficients computed above are zero.
Thus blr = blr−1 + alr = . . . = bl1 + al2 = bl0 + al1 − 10 thus we deduce
that l1 = l2 = . . . = lr = 0. However al0 = 0 so l0 = 1. But then
1 − bl0 − al1 6= 0 contradiction.

32
 This shows that c is a power of 2. Let c = 2m . Then

2n a + b|am (2mn + 1) = (2n a)m − (−b)m + (am + (−b)m ).

Since 2n a + b|(2n a)m − (−b)m , we conclude that 2n a + b|(am − (−b)m ).

This is possible only when a = 0 or am − (−b)m = 0. As a > 0 we have
am + (−b)m = 0 so m is odd and a = b. Then a|2mn + 1, so a is odd. If
n = φ(a) a|2mn − 1 so a|2 hence a = 1. Therefore a = b = 1, c = 2m for
some odd integer m, and it’s obvious that they satisfy the requirements.

33