Junior problems

J109. Let a, b, c be positive real numbers. Prove that

(a + b)2 c2
+ ≥ 4b.
c a
Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by Ovidiu Furdui, Cluj, Romania

We will use the following lemma. Lemma. Let a, b, α, β be positive real numbers
then
a2 b2 (a + b)2
+ ≥ .
α β α+β
Proof. The lemma can be proved by straight forward calculations. Based on
the lemma we have that
(a + b)2 c2 (a + b + c)2
+ ≥ ,
c a a+c
so it suffices to prove that

(a + b + c)2
≥ 4b.
a+c
The preceding inequality is equivalent to (a − b + c)2 ≥ 0, and the problem is
solved.

Second solution by Filip Stankovski, Skopje, Macedonia

By the Cauchy-Schwarz inequality, we have

r r !2
√ (a + b)2 √ c2
LHS · (c + a) ≥ c· + a·
c a
= (a + b + c)2 = (b + (a + c))2 and by AM-GM inequality
p
≥ (2 b · (a + c))2 = 4b · (a + c)
= RHS · (a + c).

Dividing both sides by (a + c) we obtain the desired result

Third solution by Shamil Asgarli, Howard Ko, Burnaby, Canada

Mathematical Reflections 2 (2009) 1

 After clearing denominators we need to prove that f (b) = ab2 + (2a2 − 4ac)b +
a3 + c3 ≥ 0. Considering f as a quadratic function in b we can easily check
that the discriminant (2a2 − 4ac)2 − 4a(a3 + c3 ) = −4ac(2a − c)2 ≤ 0. Since
the leading coefficient of this quadratic function is positive, we conclude that
f (b) ≥ 0 for all b ≥ 0.

Fourth solution by Ercole Suppa, Teramo, Italy

The proposed inequality follows from:

a b2 + 2(a − 2c)b + a2 + c3
 
(a + b)2 c2 a3 + 2a2 b + ab2 + c3 − 4abc
+ − 4b = =
c a ac ac
a b + 2(a − 2c)b + (a − 2c)2 − (a − 2c)2 + a2 + c3
 2
=
ac 
a (b + a − 2c)2 + 4ac − 4c2 + c3

=
ac
a(a + b − 2c)2 + 4a2 c − 4ac2 + c3
=
ac
a(a + b − 2c)2 + c 4a2 − 4ac + c2


=
ac
(a + b − 2c)2 (2a − c)2
= + ≥ 0.
c a

Also solved by Paolo Perfetti, Universita degli studi di Tor Vergata, Italy; John
T. Robinson, Yorktown Heights, NY, USA; Baleanu Andrei Razvan, Romania;
Brian Bradie, Newport News, USA; Arkady Alt, San Jose, California, USA;
Navid Safaei, Tehran, Iran; Daniel Lasaosa, Universidad Publica de Navarra,
Spain; Nguyen Manh Dung, Hanoi University of Science, Vietnam; Paolo Leonetti,
Milano, Italy.

Mathematical Reflections 2 (2009) 2

J110. Let τ (n) and φ(n) denote the number of divisors of n and the number of positive
integers less than or equal to n that are relatively prime to n, respectively. Find
all n such that τ (n) = 6 and 3φ(n) = 7!.

Proposed by Ivan Borsenco, Massachusetts Institute of Technology, USA

First solution by Brian Bradie, Newport News, USA

If τ (n) = 6, then n = p5 or n = p1 p22 for some primes p, p1 and p2 . If n = p5 ,
then φ(n) = p4 (p − 1); however, there is no prime p for which
1
p4 (p − 1) = 7! = 24 · 3 · 5 · 7.
3
If n = p1 p22 , then φ(n) = (p1 − 1)p2 (p2 − 1). We have four cases to consider:

p2 =2 p2 (p2 − 1) = 2 23 · 3 · 5 · 7 + 1 = 841 is not prime

p2 =3 p2 (p2 − 1) = 6 23 · 5 · 7 + 1 = 281 is prime
p2 =5 p2 (p2 − 1) = 20 22 · 3 · 7 + 1 = 85 is not prime
p2 =7 p2 (p2 − 1) = 42 23 · 5 + 1 = 41 is prime

Thus, there are two n such that τ (n) = 6 and 3φ(n) = 7!:

n = 281 · 32 = 2529 and n = 41 · 72 = 2009.

Second solution by John T. Robinson, Yorktown Heights, NY, USA

Suppose the prime factorization of n is

n = pa11 pa22 · · · pakk .

Then τ (n) = (a1 + 1)(a2 + 1) · · · (ak + 1) = 6. Since the only factorizations of 6

into factors greater than 1 are a) 2 · 3 and b) 6, it follows that n must be of the
form a) p1 p22 for distinct primes p1 and p2 , or b) p5 for some prime p.
In case b) we have

φ(n) = 2 · 4 · 5 · 6 · 7 = 1680 = p4 (p − 1) = 24 · 3 · 5 · 7

for some prime p, however since 1680 has only 2 as a fourth power of a prime
in its factorization, and 24 (2 − 1) 6= 1680, this case is excluded.
In case a) we have

φ(n) = 2 · 4 · 5 · 6 · 7 = 1680 = (p1 − 1)(p2 − 1)p2 = 24 · 3 · 5 · 7

Mathematical Reflections 2 (2009) 3

 for distinct primes p1 and p2 . Checking p2 = 2, 3, 5, and 7, the requirement is
1680
that (p −1)p +1 be prime; we have (for p2 = 2, 3, 5, and 7 respectively) 841 =
2 2
292 , 281 which is prime, 85 = 5 · 17, and 41 which is prime. The result is that
there are only two possibilities for n: n = 281 · 32 = 2529 and n = 41 · 72 = 2009.

Third solution by Paolo Leonetti, Milano, Italy

If n = i pαi i and the number of divisor is i αi + 1 = 6 then n is in the form
Q Q
pq 2 with p, q different primes or in the form p5 . Considering the constraint
q(q − 1)(p − 1) = 24 · 3 · 5 · 7 we have q ∈ {2, 3, 5, 7} (in the other case only p = 2
has the fourth power, but it is impossible for the other factors). So the only
numbers are 281 · 32 and 41 · 72 .

Also solved by Daniel Lasaosa, Universidad Publica de Navarra, Spain; Elmir

Sukanovic, Sarajevo, Bosnia and Herzegovina.

Mathematical Reflections 2 (2009) 4

 n
Y
J111. Prove that there is no n for which (k 4 + k 2 + 1) is a perfect square.
k=1

Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by John T. Robinson, Yorktown Heights, NY, USA

Let g(k) = k 4 + k 2 + 1. If we evaluate g(1), g(2), g(3), etc., we notice an
interesting pattern: g(1) = 1 · 3, g(2) = 3 · 7, g(3) = 7 · 13, g(4) = 13 · 21, and
so on. This is because g(k) factors as

g(k) = (k 2 − k + 1)(k 2 + k + 1), and

g(k + 1) = ((k + 1)2 − (k + 1) + 1) ((k + 1)2 + (k + 1) + 1)

= (k 2 + k + 1)(k 2 + 3k + 3) .
So multiplying g(1), g(2), . . . , g(n), we see that there are a sequence of squares
in the product (32 , 72 , 132 , . . . ), but with a final factor of n2 + n + 1 left over at
the end. It follows that since n2 + n + 1 cannot be a square (since it lies between
n2 and (n + 1)2 ), the product cannot be a square.

Second solution by Brian Bradie, Newport News, USA

First, note that
1
Y
(k 4 + k 2 + 1) = 3,
k=1
which is not a perfect square. Now, suppose n > 1. Because

k 4 + k 2 + 1 = (k 2 + k + 1)(k 2 − k + 1)

and
(k + 1)2 − (k + 1) + 1 = k 2 + k + 1,
it follows that
n
Y n−1
Y
4 2 2
(k + k + 1) = (n + n + 1) (k 2 + k + 1)2 .
k=1 k=1

Moreover, because the difference between (n + 1)2 and n2 is 2n + 1, n2 + n + 1

cannot be a perfect square for any n > 1. Thus, there is no n for which
Yn
(k 4 + k 2 + 1) is a perfect square.
k=1

Third solution by Ovidiu Furdui, Campia Turzii, Cluj, Romania

Mathematical Reflections 2 (2009) 5

 Since k 4 + k 2 + 1 = (k 2 + k + 1)(k 2 − k + 1) we obtain that
n
Y
P = (k 4 + k 2 + 1)
k=1
Yn
(k 2 + k + 1)(k 2 − k + 1)


=
k=1
n
!2
Y
2
= (k − k + 1) · (n2 + n + 1).
k=1

Thus it suffices to prove that there is no natural number n ≥ 1 such that n2 +n+1
is a square. Let n2 + n + 1 = u2 . This implies that n2 + n + 1 − u2 = 0 and this
equation has rational solutions provided that the discriminant of the quadratic
equation is a square. Thus, 5 = 4u2 + ∆2 . This implies that ±∆ = 1 = ±u. It
is easy to see that if u2 = 1 then n2 + n = 0 which does not have solutions in
positive integers.

Also solved by Arkady Alt, San Jose, California, USA; Navid Safaei, Tehran,
Iran; Daniel Lasaosa, Universidad Publica de Navarra, Spain; Elmir Sukanovic,
Sarajevo, Bosnia and Herzegovina; Paolo Leonetti, Milano, Italy.

Mathematical Reflections 2 (2009) 6

J112. Let a, b, c be integers such that gcd(a, b, c) = 1 and ab + bc + ca = 0. Prove that
|a + b + c| can be expressed in the form x2 + xy + y 2 , where x and y are integers.

Proposed by Samin Riasat, Notre Dame College, Dhaka, Bangladesh

First solution by John T. Robinson, Yorktown Heights, NY, USA

Note that a, b, and c can’t all be zero (gcd(a, b, c) is either undefined or zero,
depending on which variant of the definition for gcd is used). If two, say b and
c, are zero, then a = ±1 and |a + b + c| = 12 + 1 · 0 + 02 . Finally if one, say
c, is zero, then from ab = 0 we see that a or b or both has to be zero and we
are back to one of the earlier cases. So we can assume that a, b, and c are
non-zero. Next, for notational convenience, let gab = gcd(a, b), gac = gcd(a, c),
and gbc = gcd(b, c) (note that by the definition of gcd these are all positive).
Since gcd(a, b, c) = 1 we can write

a = gab gac a0 ,

b = gab gbc b0 ,
c = gac gbc c0 ,
where gab , gac , gbc , a0 , b0 , and c0 are pairwise relatively prime. From ab+ac+bc =
0 we have
2
gab gac gbc a0 b0 + gab gac
2
gbc a0 c0 + gab gac gbc
2 0 0
bc =0;
gab a0 b0 + gac a0 c0 + gbc b0 c0 = 0
−gab a0 b0 = c0 (gac a0 + gbc b0 )
Not that this implies a0 |c0 (gac a0 +gbc b0 ) which implies that a0 |(gac a0 +gbc b0 ) (since
a0 and c0 are relatively prime) which implies that a0 |gbc b0 which implies that
a0 = ±1 (since a0 is relatively prime to gbc and b0 ). The argument is symmetric,
so we also have b0 = ±1 and c0 = ±1.
Since ab + ac + bc = 0 we cannot have all of a, b, and c positive or all negative;
the choices are one negative and two positive, or two negative and one positive.
However if two are negative and one positive, since we are looking at how |a +
b + c| can be expressed we can substitute −a, −b, −c for a, b, c to get the one
negative and two positive case, so it suffices to consider only this case. Finally,
by symmetry it also suffices to consider only the case a0 = −1, b0 , c0 = 1.
Substituting these into

gab a0 b0 + gac a0 c0 + gbc b0 c0 = 0

we have
−gab − gac + gbc = 0 ;

Mathematical Reflections 2 (2009) 7

 gac = gbc − gab .
Therefore
a + b + c = −gab gac + gab gbc + gac gbc =
−gab (gbc − gab ) + gab gbc + (gbc − gab )gbc =
2 2
−gab gbc + gab + gab gbc + gbc − gab gbc =
2 2
gab − gab gbc + gbc .
Choosing x = −gab and y = gbc completes the proof.

Second solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Assume without loss of generality a = 0. Then, bc = 0, and without loss of
generality b = 0, and since gcd(a, b, c) = 1, then |c| = |a + b + c| = 1, for x = 1
and y = 0 or vice versa. Note that if without loss of generality a + b = 0, then
ab = 0, leading again to a = b = 0. It suffices therefore to prove the proposed
result for nonzero integers (a, b, c) such that a + b, b + c, and c + a are nonzero.

Since simultaneously inverting the signs of a, b and c does not alter the proposed
problem, assume without loss of generality that two of a, b, c are positive, with-
ab
out loss of generality a, b > 0. Since c = − a+b , then a + b divides ab. Calling
d the greatest common divider of a, b, we may write a = du and b = dv for
some relatively prime positive integers u, v, and u + v must divide duv. If for
some prime p, pα divides u + v but does not divide d, then p divides uv, or it
divides one of them, and hence it divides both, absurd. Therefore, u + v divides
d. Let d = (u + v)q, where q is a positive integer, which leads to a = u(u + v)q,
b = v(u + v)q and c = −uvq. Clearly, q divides a, b, c, or since gcd(a, b, c) = 1,
then q = 1, and |a + b + c| = a + b + c = u2 + uv + v 2 . The conclusion follows.

Note finally that |a + b + c| can be expressed in the form x2 + xy + y 2 in at least

three different ways:

u2 + uv + v 2 = (−u − v)2 + (−u − v)v + v 2 = (−u − v)2 + (−u − v)u + u2 .

Also solved by Arkady Alt, San Jose, California, USA; Paolo Leonetti, Milano,
Italy.

Mathematical Reflections 2 (2009) 8

J113. Call penta-sequence a sequence of consecutive positive integers such that each
of them can be written as a sum of five nonzero perfect squares. Prove that
there are infinitely many penta-sequences of length 7.

Proposed by Ivan Borsenco, Massachusetts Institute of Technology, USA

First solution by John T. Robinson, Yorktown Heights, NY, USA

It is a result of number theory that every integer greater than or equal to
170 is a sum of five positive squares (for example see Niven and Zuckerman,
An Introduction to the Theory of Numbers, Theorem 5.6 – in fact 170 can be
lowered to 34 in this statement, but the integers in the range 34 to 169 need to
be checked separately); the result follows (note that “7” can be replaced by any
positive integer).

Second solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Consider, for each non-negative integer n, the numbers

(2n + 6)2 + (n + 5)2 + (n + 4)2 + (n + 2)2 + (n + 2)2 = 8n2 + 50n + 85,

(2n + 7)2 + (n + 4)2 + (n + 4)2 + (n + 2)2 + (n + 1)2 = 8n2 + 50n + 86,

(2n + 6)2 + (n + 5)2 + (n + 4)2 + (n + 3)2 + (n + 1)2 = 8n2 + 50n + 87,
(2n + 7)2 + (n + 5)2 + (n + 3)2 + (n + 2)2 + (n + 1)2 = 8n2 + 50n + 88,
(2n + 8)2 + (n + 4)2 + (n + 2)2 + (n + 2)2 + (n + 1)2 = 8n2 + 50n + 89,
(2n + 5)2 + (n + 6)2 + (n + 4)2 + (n + 3)2 + (n + 2)2 = 8n2 + 50n + 90,
(2n + 8)2 + (n + 4)2 + (n + 3)2 + (n + 1)2 + (n + 1)2 = 8n2 + 50n + 91.
Clearly, for each non-negative integer n, these numbers form different penta-
sequences of length 7. The conclusion follows.

Also solved by Paolo Leonetti, Milano, Italy

Mathematical Reflections 2 (2009) 9

J114. Let p be a prime. Find all solutions to the equation a + b − c − d = p, where
a, b, c, d are positive integers such that ab = cd.

Proposed by Iurie Boreico, Harvard University, USA

First solution by John T. Robinson, Yorktown Heights, NY, USA

Let x = ab = cd, and let y = gcd(a, c), so that a = ya0 and c = yc0 (where a0
and c0 are relatively prime). Since a|x and c|x we can write

x = za0 c0 y .

Next, b = x/a = zc0 and d = x/c = za0 . Substituting into a + b − c − d = p we

have:
ya0 + zc0 − yc0 − za0 = p ;
(y − z)(a0 − c0 ) = p
or equivalently
(z − y)(c0 − a0 ) = p .
Since p is prime, an exhaustive list of possible solutions in positve integers
y, z, a0 , c0 is as follows:

y = z + 1 and a0 = c0 + p ;

y = z + p and a0 = c0 + 1 ;
z = y + 1 and c0 = a0 + p ;
z = y + p and c0 = a0 + 1 .
These solutions all work out to be equivalent if, considering z (or y) and c0 (or
a0 ) to be free parameters, we allow the resulting expressions for a and b, and
c and d, to be switched. For example, considering z and c0 to be parameters
in the first solution above, we have a = (z + 1)(c0 + p), b = zc0 , c = (z + 1)c0 ,
d = z(c0 + p), and similarly for the other cases. All of the above solutions can
be expressed in a more concise parametric form as follows, where u and v are
any two positive integers:

{a, b} = {uv, (u + 1)(v + p)} ;

{c, d} = {(u + 1)v, u(v + p)} .

Second solution by Arkady Alt, San Jose, California, USA

We will use the notation x ⊥ y when integers x and y are relatively prime, i.e.
gcd (x, y) = 1 and x | y if x divides y. Let s = gcd (a, c) . Since a = ms,

Mathematical Reflections 2 (2009) 10

 c = sc1 then gcd (sm, sc1 ) = s ⇐⇒ gcd (m, c1 ) = 1. Let t = gcd (b, c1 ) . Since
b = nt, c1 = tc2 then gcd (tn, tc2 ) = t ⇐⇒ gcd (n, c2 ) = 1. Thus we obtain
a = ms, b = tn, c = stc2 , where c2 ⊥ n and c2 ⊥ m (because c2 | c1 and
c1 ⊥ m . Hence, c2 ⊥ mn and

gcd (ab, c) = gcd (smtn, stxc2 ) = st gcd (mn, c2 ) = st

and, since c | ab yields gcd (ab, c) = c,then st = c. Let a = ms, b = tn, c = st .

Then ab = cd yields d = mn and, therefore, a + b − c − d = p ⇐⇒
ms + nt − st − mn = p ⇐⇒ (m − t) (s − n) = p. Thus we have four types of
solutions:
 
m − t = −1 m=t−1
1. then and
s − n = −p n=s+p
a = s (t − 1) , b = t (s + p) , c = st, d = (t − 1) (s + p) ;
 
m − t = −p m=t−p
2. then and
s − n = −1 n=s+1
a = s (t − p) , b = t (s + 1) , c = st, d = (t − p) (s + 1) ;
 
m−t=1 m=t+1
3. then and
s−n=p n=s−p
a = s (t + 1) , b = t (s − p) , c = st, d = (s − p) (t + 1) ;
 
m−t=p m=t+p
4. then and
s−n=1 n=s−1
a = s (t + p) , b = t (s − 1) , c = st, d = (s − 1) (t + p) ,
where s and t are any non-zero integers.

Third solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Assume without loss of generality that a ≥ b and c ≥ d, since a, b are inter-
changeable, and so are c, d. Clearly a > c ≥ d > b, since if a = c, then b = d and
p = 0 absurd, and similarly if d = b. Moreover, since ab = cd but a + b > c + d,
then a − b > c − d, and if a ≤ c, then b < d, yielding ab < cd absurd. Clearly,
p(a + b + c + d) = (a + b)2 − (b + d)2 = (a + c)(a − c) + (b + d)(b − d), or

(b + d)(p + d − b) p(b + d)
a+c= =b+d+ ,
a−c−p d−b
pd pb
yielding a − d = d−b and c − b = d−b .

Assume first that d − b divides d and b. Hence, we may write d − b = k, and

d = (u + 1)k, b = uk for some positive integer u. Then, a = (u + 1)(k + p)
and c = u(k + p). Note that these necessary forms for a, b, c, d satisfy both
conditions, since ab = cd = u(u + 1)k(k + p). All possible solutions in this case

Mathematical Reflections 2 (2009) 11

 may then be written as (a, b, c, d) = ((u + 1)(k + p), uk, u(k + p), (u + 1)k) for
any positive integers u, k. Generality is restored by allowing permutations of
a, b, of c, d, and of both.

Also solved by Paolo Leonetti, Milano, Italy

Mathematical Reflections 2 (2009) 12

 Senior problems

S109. Solve the system of equations

√ 1 √ 1 √ 1 7
x− = y− = z− = .
y z x 4

Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by Ercole Suppa, Teramo, Italy

Adding the three equations we obtain:
√ √ √
1 1 1 21
x+ y+ z=
+ + + ⇒
x y z 4
√
√ √

x − 2 f (x) + ( y − 2) f (y) + z − 2 f (z) = 0 (1)
where √
2+
t + 4t
f (t) = > 0, ∀t > 0
4t
√ √
Now x ≥ 2. In fact if x < 2 then

x<4 ⇒
√ 1 7 1 7
z=+ > + =2 ⇒
x 4 4 4
√ 1 7 1 7
y= + < + =2 ⇒
z 4 4 4
√ 1 7 1 7
x= + > + =2
y 4 4 4
√ √
which is impossible. Similarly y < 2 or z < 2 leads to contradiction. Thus
√ √ √
x−2≥0 , y−2≥0 , z−2≥0

Therefore, since f (x) ≥ 0, f (y) ≥ 0, f (z) ≥ 0 from (1) follows that:

√ √ √
x= y= z=2 ⇒ x=y=z=4

and we are done.

Second solution by Francisco Javier Garcı́a Capitán, Spain

We have x = f (y), y = f (z), z = f (x) where f (x) = ( x1 + 74 )2 is a strictly
decreasing function over the positive real numbers. Hence we have f (f (f (x))) =
x and the same for y and z.

Mathematical Reflections 2 (2009) 13

 Suppose that f (x) > x. Then we have f (f (x)) < f (x) and x = f (f (f (x))) >
f (f (x)). And, applying f again, we have f (x) < x, contradiction. In a similar
way we discard the case f (x) < x and we have f (x) = x, and the same for y
and z.
Solving f (x) = x we get only a real solution x = 4, so the solution of our system
is x = y = z = 4.

Third solution by Magkos Athanasios, Kozani, Greece

√ √ √
Set x = a, y = b, z = c. We have a, b, c > 0 and the system becomes

1 7 1 7 1 7
a− 2
= ,b − 2 = ,c − 2 = .
b 4 c 4 a 4
Pairwise subtractions of these equations yield

(c − b)(c + b) (a − c)(a + c) (b − a)(b + a)

a−b= 2
,b − c = 2
,c − a = .
(bc) (ac) (ab)2
Multiplying these equations we obtain

(b − a)(a − c)(c − b)(a + b)(b + c)(c + a)

(a − b)(b − c)(c − a) = .
(abc)4

Now, if the system had a solution (a, b, c) with a 6= b 6= c 6= a, the last equation
would imply (a + b)(b + c)(c + a) = −(abc)4 , which is impossible since a, b, c > 0.
Hence two of a, b, c are equal, say a = b. Then from the first equation we easily
find a = b = 2 and from the third equation we find c = 2. Hence, the original
system has the unique solution (4, 4, 4).

Also solved by Baleanu Andrei Razvan, Romania; Paolo Perfetti, Universita

degli studi di Tor Vergata, Italy; John T. Robinson, Yorktown Heights, NY,
USA; Oleh Faynshteyn, Leipzig, Germany; Navid Safaei, Tehran, Iran; Daniel
Lasaosa, Universidad Publica de Navarra, Spain.

Mathematical Reflections 2 (2009) 14

S110. Let X be a point on the side BC of a triangle ABC. The parallel through X
to AB meets CA at V and the parallel through X to AC meets AB at W . Let
D = BV ∩ XW and E = CW ∩ XV . Prove that DE is parallel to BC and
1 1 1
= + .
DE BX CX

Proposed by Francisco Javier Garcı́a Capitán, Spain

First solution by Miguel Amengual Covas, Mallorca, Spain

By Thales’ theorem,
AV WE
= (1)
VC EC
AV BV VC
Since W X k CA, we get WD = BD = DX , so

AV WD
= (2)
VC DX
so that we have from (1) and (2),

WE WD
=
EC DX

Thus DE k XC, that is DE k BC .

We extend DE to meet AB at F and obtain

FD WD DE
= =
BX WX XC

Therefore

DE DE DE FD
+ = +
BX XC BX BX
DE + F D
=
BX
FE
=
BX
= 1
since F E = BX follows because the quadrilateral F BXE is a parallelogram.

Mathematical Reflections 2 (2009) 15

 Hence
1 1 1
= +
DE BX CX
as desired.

Second solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Call ~u = AB ~ and ~v = AC. ~ ~ = ~v − ~u, and for any X ∈ BC, a
Clearly, BC
~ ~ ~
real ρ exists such that BX = ρBC, AX = (1 − ρ)~u + ρ~v . Since V ∈ AC, then
~ = α~v for some real κ. Moreover, since XV
AV ~ k AB,~ the coefficient of ~v is the
~ ~ ~ ~
same for AV and AX, or AV = ρ~v . Similarly, AW = (1 − ρ)~u. Since D ∈ BV ,
then AD~ = κAB ~ + (1 − κ)AV ~ for some real κ, or AD
~ = κ~u + (1 − κ)ρ~v , and
~
since D ∈ XW , then the coefficient of ~u in AD is 1 − ρ. Or, κ = 1 − ρ and
~ = (1 − ρ)~u + ρ2~v . Similarly, AE
AD ~ = (1 − ρ)2 ~u + ρ~v , or DE
~ = AE
~ − AD ~ =
~ Hence, DE
ρ(1 − ρ) (~v − ~u) = ρ(1 − ρ)BC. ~ k BC,
~ and

1 1 1 1 1 1
= = + = + .
DE ρ(1 − ρ)BC ρBC (1 − ρ)BC BX CX

Note that the result is true for any real ρ except for ρ = 0, 1, ie, we may take
X on any point of line BC except for B or C, and the result will always hold
as long as we take signed distances for BX and CX, in other words, the last
relation will be valid as long as BX ≤ 0 if and only if X is on line BC but not
on ray BC, and similarly for CX.

Also solved by Baleanu Andrei Razvan, Romania; Salem Malikic, Bosnia and
Herzegovina.

Mathematical Reflections 2 (2009) 16

S111. Prove that there are infinitely many positive integers n that can be expressed
as a4 + b4 + c4 + d4 − 4abcd, where a, b, c, d are positive integers, such that n is
divisible by the sum of its digits.

Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by John T. Robinson, Yorktown Heights, NY, USA

For j ≥ 1, let
n = 196 · 104j−2 + 2 · 104j−4 , then
n = (10j )4 + (10j )4 + (10j−1 )4 + (10j−1 )4 − 4(10j )(10j )(10j−1 )(10j−1 ),
and n is divisible by the sum of its digits (18) since n is even and also divisible
by 9 by the decimal integer divisibility test for 9 (i.e. the sum of the digits is
divisible by 9).

Second solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Note that it suffices to find one, since if n = a4 + b4 + c4 + d4 − 4abcd, then
n0 = (10m a)4 + (10m b)4 + (10m c)4 + (10b d)4 − 4(10m a)(10m b)(10m b)(10m c) =
104m n has the same sum of digits as n, and if n is divisible by the sum of its
digits, so is n0 . Note now that (105 )4 + (104 )4 + (2 · 103 )4 + 14 − 4 · 2 · 105+4+3 =
1020 + 1016 + 8 · 1012 + 1 has sum of digits 11, and is clearly divisible by 11 since
the sum of its digits in odd positions is 11, and the sum of its digits in even
positions is 0. The conclusion follows.

Mathematical Reflections 2 (2009) 17

S112. Let a, b, c be the side lengths and let s be the semiperimeter of a triangle ABC.
Prove that
 a  a  b b  c  c
(s − c)a (s − a)b (s − b)c ≤ .
2 2 2

Proposed by Johan Gunardi, Jakarta, Indonesia

First solution by Arkady Alt, San Jose, California, USA

By the weighted AM-GM inequality we have
a+b+c
s−c s−a s−b


s−c
a 
s−a
b 
s−b
c ·a+ ·b+ ·c
≤ a b c =
 
a b c a+b+c


1  a  a  b b  c  c
a b c
⇐⇒ (s − c) (s − a) (s − b) ≤ .
2a+b+c 2 2 2

Second solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

If the triangle is degenerate, then the LHS is identically zero, while the RHS
is non-negative, the inequality being trivially true. We need to consider thus
only non-degenerate triangles. Dividing by the RHS and taking logarithm, the
inequality may be rewritten in the following equivalent form:
     
b−c c−a a−b
a log 1 + + b log 1 + + c log 1 + ≤ 0.
a b c

Clearly, g(x) = x−log(1+x) is zero for x = 0, while dg(x) 1

dx = 1− 1+x . Note that if
x > 0, then g(x) strictly increases, while if x < 0, then g(x) strictly decreases, or
the maximum of g(x) is 0, occurring iff x = 0. As a consequence, x ≥ log(1 + x)
for all x > −1, with equality iff x = 0. The triangular inequality
guarantees
that −1 < b−c a < 1 for non-degenerate triangles, or a log 1 + b−c
a ≤ b − c with
equality iff b = c. Adding this inequality to its cyclic permutations we obtain
the proposed inequality. Equality holds if and only if, either a = b = c, or the
triangle is degenerate with two equal sides and one side of length 0.

Third solution by Nguyen Manh Dung, Hanoi University of Science, Vietnam

The inequality is equivalent to

(a + b − c)a (b + c − a)b (c + a − b)c ≤ aa bb cc

It suffices to shows that

a+b−c b+c−a c+a−b
a ln + b ln + c ln ≤0
a b c

Mathematical Reflections 2 (2009) 18

 Since the function f (x) = ln x is concave, by Jensen’s Inequality we get

a+b−c b+c−a c+a−b

a ln + b ln + c ln
a b  c 
a+b−c+b+c−a+c+a−b
≤ (a + b + c) ln
a+b+c
= (a + b + c) ln 1 = 0

Hence we are done.

The equality holds when a = b = c.

Fourth solution by Magkos Athanasios, Kozani, Greece

Observe that the inequality we have to prove is equivalent to

 a  b  c
b−c c−a a−b
1+ 1+ 1+ ≤ 1.
a b c
b−c
Employing the well known inequality ex ≥ 1 + x and setting x = a , we get
 a
b−c b−c b−c
e a ≥1+ ⇒ 1+ ≤ eb−c .
a a

Analogously, we have
 b  c
c−a c−a a−b
1+ ≤e , 1+ ≤ ea−b .
b c

Multiplying these three inequalities we get the desired result. Since ex = 1+x ⇔
x = 0, it is clear that the equality holds iff a = b = c.

Mathematical Reflections 2 (2009) 19

S113. Prove that for different choices of signs + and − the expression
±1 ± 2 ± 3 ± · · · ± (4n + 1),
yields all odd positive integers less than or equal to (2n + 1)(4n + 1).

Proposed by Dorin Andrica, “Babes-Bolyai” University, Romania

First solution by John T. Robinson, Yorktown Heights, NY, USA

If we take all signs positive we have
1 + 2 + 3 + · · · + 4n + (4n + 1) = (2n + 1)(4n + 1) = S .
Consider now what happens if on the left hand side we change the sign of exactly
one summand: if we replace 1 by -1 this subtracts 2 from S; if we replace 2 by
-2 this subtracts 4 from S; and so on, up to replacing (4n + 1) by −(4n + 1)
which subtracts (8n + 2) from S. Clearly this process can be repeated, leaving
the rightmost terms with the signs changed alone on the next iteration (that is
for the first iteration all terms are initially positive, for the second iteration the
(4n+1) term starts out negative, for the third iteration the 4n and (4n+1) terms
start out negative, and so on). This gives a decreasing sequence of consecutive
odd positive integers starting with S and ending with −S, and this sequence
includes 1, 3, 5, . . . , S as a sub-sequence.

Second solution by Brian Bradie, Newport News, USA

If we take all of the signs to be +, we generate (2n + 1)(4n + 1) because
4n+1
X (4n + 1)(4n + 2)
j= = (2n + 1)(4n + 1).
2
j=1

Now, let x be an odd integer less than (2n + 1)(4n + 1). Suppose
x ≥ (2n + 1)(4n + 1) − 2(4n + 1) = (2n − 1)(4n + 1).
Because (2n+1)(4n+1) and x are both odd, it follows that there exists a natural
number k with 1 ≤ k ≤ 4n + 1 such that (2n + 1)(4n + 1) − x = 2k. Thus,
4n+1
X 4n+1
X
x = (2n + 1)(4n + 1) − 2k = j − 2k = j − k;
j=1 j=1,j6=k

that is, take the + sign for all integers except k for which we take the − sign.
Next, suppose x < (2n − 1)(4n + 1) and let ` be the smallest natural number
for which
X ` 4n+1
X
j− j > x.
j=1 j=`+1

Mathematical Reflections 2 (2009) 20

 Because
`
X 4n+1
X 4n+1
X 4n+1
X
j− j= j−2 j
j=1 j=`+1 j=1 j=`+1

is odd and x is also odd, there exists a natural number k with 1 ≤ k ≤ ` such
that
X` 4n+1
X
j− j − x = 2k.
j=1 j=`+1

Thus,
`
X 4n+1
X `
X 4n+1
X
x= j− j − 2k = j−k− j;
j=1 j=`+1 j=1,j6=k j=`+1

that is, take the + sign on 1 through ` except k and the − sign on k and ` + 1
through 4n + 1.

Third solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Clearly, the result is true for n = 1 since 15 = 5+4+3+2+1, 13 = 5+4+3+2−1,
11 = 5+4+3−2+1, 9 = 5+4−3+2−1, 7 = 5−4+3+2+1, 5 = 5−4+3+2−1,
3 = 5 − 4 + 3 − 2 + 1 and 1 = 5 − 4 − 3 + 2 + 1. Assume now that the result
is true for n − 1. Then, since (4n + 1) + 4n + (4n − 1) + (4n − 2) = 16n − 2,
adding the sum of the remaining 4n − 3 elements with all combinations of signs
that produce all positive integers between 1 and (2n − 1)(4n − 3), we obtain all
odd numbers between 16n − 1 and 8n2 + 6n + 1 = (4n + 1)(2n + 1). Moreover,
subtracting from 16n − 2 the same sums of the remaining 4n − 3 elements with
the same combinations of signs, we obtain all odd integers between 16n − 3 and
−8n2 + 10n − 3 = −(2n − 1)(4n − 3). Clearly, this last integer is negative for
any positive integer n, so all odd numbers between 1 and (2n + 1)(4n + 1) have
thus been generated. The conclusion follows.

Mathematical Reflections 2 (2009) 21

S114. Consider triangle ABC with angle bisectors AA1 , BB1 , CC1 . Denote by U the
intersection of AA1 and B1 C1 . Let V be the projection from U to BC. Let W
be the intersection of the angle bisectors of ∠BC1 V and ∠CB1 V. Prove that
A, V, W are collinear.

Proposed by Ivan Borsenco, Massachusetts Institute of Technology, USA

First solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

If ABC is isosceles at A, then AA1 = AU = AV is an axis of symmetry,
and the internal bisectors of ∠BC1 V and ∠CB1 V are parallel to each other
and symmetric around AV . In this sense, W is a point ”at infinity” on AV .
Otherwise, assume withoit loss of generality that b > c and thus ∠B > ∠C. We
begin by calculating BV and CV . The theorem of the transversals guarantees
that
CA1 · BC1 BA1 · CB1 BC · U A1
+ = ,
AC1 AB1 AU
or since, as it is well known, BA BA
CA1 = CA , and similarly for its cyclic permutations,
1

2a
then b+c = UAU A1
= AAU1 −U
A1 2aAA1
A1 , and U A1 = 2a+b+c . Now, applying Stewart’s
theorem to the internal bisector of ∠BAC, we find the well-known result AA1 =
A
2bc cos
b+c
2
, or

B−C 4abc sin B+C

2 sin 2
B−C
2abc(cos C − cos B)
V A1 = U A1 sin = = ,
2 (2a + b + c)(b + c) (2a + b + c)(b + c)

where we have used that U V is perpendicular to BC and V A1 is the internal

bisector of ∠BAC to deduce that ∠V U A1 = B−C2 . Finally,

a2 + ac + c2 − b2 a2 + ab + b2 − c2
BV = BA1 − V A1 = , CV = a − BV = .
2a + b + c 2a + b + c

Let now the internal bisector of ∠BC1 V intersect BC at X and AV at W1 , and

let the internal bisector of ∠CB1 V intersect BC at Y and AV at W2 . Clearly,
XB C1 B YC B1 C
XV = C1 V , and Y V = B1 V , while Menelaus’ theorem guarantees that

V W1 AC1 BX V W2 AB1 CY
1= , 1= .
W1 A C1 B XV W2 A B1 C Y V
The proposed result is clearly equivalent to W1 = W2 , and it suffices to show
that VWW
1A
1
= VWW
2A
2
, or equivalently,

B1 V AB1 a+b
= = , (a + c)2 B1 V 2 = (a + b)2 C1 V 2 .
C1 V AC1 a+c

Mathematical Reflections 2 (2009) 22

 a2 +b2 −c2
Now, the Cosine Law allows us to write cos C = 2ab and

BV · BC1 2 2 2
C1 V 2 = BV 2 + BC12 − 2BV · BC1 cos C = BV 2 + BC12 − (a + b − c ).
a
After some algebra,

a2 (a + b)2 (a + c)2 (a + b)(a + c)(b2 − c2 )2

(a + b)2 C1 V 2 = − −
(2a + b + c)2 (2a + b + c)2

a(a + b)(a + c)(b + c)(b − c)2 (b2 + c2 ) − a(a + b)(a + c) + (b3 + c3 )

− 2
+ a2 .
(2a + b + c) 2a + b + c
Note that the RHS is invariant under exchange of b and c, or (a + c)2 B1 V 2 =
(a + b)2 C1 V 2 . The result follows.

Second solution by Ercole Suppa, Teramo, Italy

Denote as usual by a, b, c, R the sides BC, CA, AB and the circumradius of
4ABC, respectively. Let X, W1 be the points where the the angle bisector of
∠BC1 V intersects BC, AV and let Y , W2 be the points where the the angle
bisector of ∠V B1 C intersects BC, AV (refere to Figure 1). In order to com-
plete the proof is enough to show that W1 = W2 .

Applying Menelaus’s theorem to triangle 4ABV and the line XC1 W1 , and
taking into account that BX : XV = C1 B : C1 V (internal bisector theorem),
we have:
AC1 C1 B V W1
· · =1 (1)
C1 B C1 V W1 A

In similar way applying Menelaus’s theorem to triangle 4AV C and the line
Y B1 W2 , and taking into account that V Y : Y C = B1 V : B1 C, we have:
AB1 B1 C V W2
· · =1 (2)
B1 C B1 V W2 A

From (1), (2) follows that:

AC1 V W1 AB1 V W2
· = · (3)
C1 V W1 A B1 V W2 A

C1 V
In order to compute the ratio B 1V
, we draw the perpendiculars from C1 , B1 to
BC and denote with C2 , B2 the projections of C1 , B1 , respectively.

Mathematical Reflections 2 (2009) 23

 W1=W2=W

A
B1
U
C1
I

B X V A1 Y C

Figure 1

B1
U
C1
I

B C2 V A1 B2 C

Figure 2

Mathematical Reflections 2 (2009) 24

 bc ac
Using the Thales theorem and the well known relations AC1 = a+b , BC1 = a+b ,
bc ab
AB1 = a+c , CB1 = a+c we have:

ac 2R
C1 C2 BC1 sin B a+b a a+c
= = ab 2R
= (4)
B1 B2 CB1 sin C a+c b
a+b
bc
C2 V C1 U AC1 a+b a+c
= = = bc
= (5)
V B2 U B1 AB1 a+c
a+b
From (4), (5) follows that:
C1 C2 C2 V
= (6)
B1 B2 V B2
Thus the right triangles 4C1 C2 V , 4B1 B2 V are similar, so:
C1 V C2 V C1 U AC1
= = = (7)
B1 V V B2 U B1 AB1
From (3), (7) follows that

V W1 V W2
= =⇒ W1 = W2
W1 A W2 A
and the proof is completed.

Mathematical Reflections 2 (2009) 25

 Undergraduate problems

U109. Find all pairs (m, n) of integers such that m2 + 2mn − n2 = 1.

Proposed by Titu Andreescu, University of Texas at Dallas, USA and Dorin

Andrica, Babes-Bolyai University, Romania

Solution by John T. Robinson, Yorktown Heights, NY, USA

The equation can be written as

(m + n)2 − 2n2 = 1 .

Let x = m + n, then we have

x2 − 1 = (x − 1)(x + 1) = 2n2 .

We see from this that x must be odd, so substituting x = 2y + 1:

2y(2y + 2) = 2n2 ;

2y(y + 1) = n2 .
We now see that n must be even, so substituting n = 2z:

2y(y + 1) = 4z 2 ;
y(y + 1)
= z2 .
2
From this we see that z 2 , if non-zero, must be a square triangular number.
There are various formulas for these numbers; one is the following:
√ let pk /qk
be the kth convergent in the continued fraction expansion for 2, then the kth
square triangular number is p2k qk2 . Therefore, letting z = ±pk qk and working
backwards, we have:
n = ±2pk qk ;
(m + n)2 = 1 + 2n2 = 1 + 8p2k qk2 ;
q
m + n = ± 1 + 8p2k qk2 ;
q
m = ∓2pk qk ± 1 + 8p2k qk2 .
So in addition to (m, n) = (1, 0), a general solution is
q
(m, n) = (∓2pk qk ± 1 + 8p2k qk2 , ±2pk qk )

Mathematical Reflections 2 (2009) 26

 √
where pk /qk is the kth convergent in the continued fraction expansion for 2,
and where the signs in this expression are chosen as + + −, − + +, + − −, or
− − + (that is, the first and last signs are opposite, and the middle sign can be
chosen independently).

Also solved by Arkady Alt, San Jose, California, USA; Brian Bradie, Newport
News, USA; Daniel Lasaosa, Universidad Publica de Navarra, Spain; Minghua
Lin, China, Paolo Leonetti, Milano, Italy.

Mathematical Reflections 2 (2009) 27

U110. Let a1 , a2 , . . . , an be real numbers with an , a0 6= 0 such that the polynomial
P (X) = (−1)n an X n + (−1)n−1 an−1 X n−1 + · · · + a2 X 2 − a1 X + a0 has all of
its roots in the interval (0, ∞), and let f : R → R be an n-time differentiable
function. Prove that if
 00 0

lim an f (n) (x) + an−1 f (n−1) (x) + · · · + a2 f (x) + a1 f (x) + a0 f (x) = L ∈ R,
x→∞
L
then lim f (x) exists and lim f (x) = a0 .
x→∞ x→∞

Proposed by Radu Ţiţiu, Targu-Mures, Romania

Solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

The proposed result will be first proved by induction over n in the particular
case L = 0. For n = 0, the result is trivially true, since the condition is actually
lim a0 f (x) = 0 with a0 6= 0.
x→∞

For n = 1, P (X) = −a1 X + a0 has a positive real root, ie, a0 and a1 have the
same sign, and we may assume without loss of generality that a0 , a1 > 0. The
condition limx→∞ (a1 f 0 (x) + a0 f (x)) = 0 is equivalent to stating that, for any
positive  > 0, there exists one real value M such that, for all x > M , then
|a1 f 0 (x) + a0 f (x)| < . Define now the (possibly empty) set X0 such that any
x0 ∈ X0 satisfies x0 > M and f 0 (x0 ) = 0. Clearly, for all x0 ∈ X0 , |f (x0 )| < a0 .
If X0 does not have an upper bound, then for any x which is larger than some
x0 we may find x− < x+ ∈ X0 such that x ∈ (x− , x+ ). If |f (x)| > a0 for some x
larger than some x0 , then clearly there must be a local maximum or minimum
in (x− , x+ ) where |f (x)| > a0 . In this local maximum or minimum, f 0 (x) = 0,
reaching a contradiction. Hence if f (x) does not converge to 0, the set X0 has
an upper bound, and some real N exists such that f 0 (x) does not change signs
for x > N , ie, f (x) is monotonous for x > N . Since we may exchange f (x)
by −f (x) without altering the problem, assume without loss of generality that
f (x) is increasing, hence f 0 (x) > 0, for all x > N . Therefore, for all x > N ,
0
either f (x) > 0 and f (x) < −aa1 f0 (x) < a0 , or f (x) is negative. In either case,
f (x) is an increasing function with an upper bound , hence it has a limit.
Calling ` this limit, clearly limx→∞ f 0 (x) = − `a a1 = 0, since otherwise f (x)
0

would not have a limit. Therefore, ` = 0, and the result is proved for n = 1.
Assume now that the result is proved for 1, 2, . . . , n − 1. Choose any root
r ∈ R+ of P (X). Clearly, P (X) = (X − r)Q(X), where Q(x) is an n − 1-degree
d
polynomial with all roots real and positive. Defining the operator ∆ = dx ,
clearly the condition is written as
lim (P (−∆)f (x)) = 0, or equivalently lim (Q(−∆)(∆ + r)f (x)) = 0.
x→∞ x→∞

Mathematical Reflections 2 (2009) 28

 Note therefore that (∆ + r)f (x) = f 0 (x) + rf (x) is a real function, n − 1-
times differentiable, and such that limx→∞ (Q(−∆)f (x)) = 0 for some poly-
nomial Q(X) with all its roots real and positive. Therefore, (∆ + r)f (x) sat-
isfies the condition of the problem for n − 1, and by hypothesis of induction
limx→∞ ((∆ + r)f (x)) = 0, where clearly −X + r is a linear polynomial with
real and positive root r, and f (x) is differentiable at least once. Hence f (x)
satisfies the condition of the problem for n = 1. The result for L = 0 follows.

If L 6= 0, define g(x) = f (x) − aL0 . Clearly, g(x) satisfies the conditions of the
problem with L = 0, and limx→∞ g(x) = 0. The conclusion follows.

Mathematical Reflections 2 (2009) 29

 π
U111. Let n be a given positive integer and let ak = 2 cos 2n−k , k = 0, 1, . . . , n − 1.
Prove that
n−1
Y (−1)n−1
(1 − ak ) = .
1 + a0
k=0

Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by Brian Bradie, Newport News, USA

This problem is very similar to Problem 1097 which appeared in the Pi Mu
Epsilon Journal (volume 12, number 2, Spring 2005, pp. 113-114). Note that
π π
 π  1 + 2 cos 2n−k 1 − 4 cos2 2n−k
1 − ak = 1 − 2 cos π = π
2n−k 1 + 2 cos 2n−k 1 + 2 cos 2n−k
π π


1 − 2 1 + cos 2n−k−1 1 + 2 cos 2n−k−1
= π =− π
1 + 2 cos 2n−k 1 + 2 cos 2n−k
1 + ak+1
= − .
1 + ak
Thus, the indicated product telescopes, and
n−1
Y 1 + 2 cos π (−1)n+1 (−1)n−1
(1 − ak ) = (−1)n = = .
1 + a0 1 + a0 1 + a0
k=0

Second solution by Darij Grinberg, Germany

Lemma 1. For every t ∈ R, we have

(2 cos t − 1) (2 cos t + 1) = 2 cos (2t) + 1.

Proof. We have

(2 cos t − 1) (2 cos t + 1) = 4 cos2 t − 1 = 2 2 cos2 t − 1 +1 = 2 cos (2t) + 1,

| {z }
=cos(2t)

and Lemma 1 is proved.

Lemma 2. For every k ∈ {0, 1, ..., n − 1} , we have
ak+1 + 1
ak − 1 = ,
ak + 1
π
where we set an = −2 (so that ak = 2 cos holds for all k ∈ {0, 1, ..., n}).
2n−k

Mathematical Reflections 2 (2009) 30

 π
Proof. We have ak + 1 6= 0 (since ak = 2 cos n−k
> 0) and
2
| {z }
∈[0,π/2]
 π  π   π 
(ak − 1) (ak + 1) = 2 cos n−k − 1 2 cos n−k + 1 = 2 cos 2 n−k + 1 (by Lemma 1)
2 2 2
π
= 2 cos n−(k+1) + 1 = ak+1 + 1,
2
ak+1 + 1
so that ak − 1 = . Lemma 2 is proved.
ak + 1
Now,
n−1 n−1 n−1 n−1
Y Y n
Y n
Y ak+1 + 1
(1 − ak ) = (− (ak − 1)) = (−1) (ak − 1) = (−1) (by Lemma 2)
ak + 1
k=0 k=0 k=0 k=0
n−1
Q n
Q
(ak+1 + 1) (ak + 1)
an + 1
= (−1)n k=0
n−1
= k=1
(−1)n n−1 = (−1)n
Q Q a0 + 1
(ak + 1) (ak + 1)
k=0 k=0
−2 + 1 −1 − (−1)n (−1)n−1
= (−1)n = (−1)n = = .
a0 + 1 a0 + 1 1 + a0 1 + a0

Also solved by Arkady Alt, San Jose, California, USA; Navid Safaei, Tehran,
Iran; Daniel Lasaosa, Universidad Publica de Navarra, Spain; Paolo Leonetti,
Milano, Italy.

Mathematical Reflections 2 (2009) 31

U112. Let x, y, z be real numbers greater than 1. Prove that
3 +2xyz 3 +2xyz 3 +2xyz
xx · yy · zz ≥ (xx y y z z )xy+yz+zx .

Proposed by Cezar Lupu, University of Bucharest, Romania and Valentin

Vornicu, San Diego, USA

First solution by Darij Grinberg, Germany

Lemma 1. Let x, y, z, a, b, c be nonnegative reals such that x ≥ y ≥ z and
ax ≥ by. Then,

x3 + 2xyz a + y 3 + 2xyz b + z 3 + 2xyz c ≥ (yz + zx + xy) (xa + yb + zc) .

Proof. Since x ≥ y ≥ z and ax ≥ by, the Vornicu-Schur inequality1 , applied to

A = x, B = y, C = z, X = ax, Y = by, Z = cz, yields

ax (x − y) (x − z) + by (y − z) (y − x) + cz (z − x) (z − y) ≥ 0.

This rewrites as

x3 + 2xyz a+ y 3 + 2xyz b+ z 3 + 2xyz c−(yz + zx + xy) (xa + yb + zc) ≥ 0.

Thus,

x3 + 2xyz a + y 3 + 2xyz b + z 3 + 2xyz c ≥ (yz + zx + xy) (xa + yb + zc) ,

proving Lemma 1.
Let a = ln x, b = ln y, c = ln z. Then, a, b, c are nonnegative (since x, y, z are
greater or equal to 1). Without loss of generality assume that x ≥ y ≥ z (we
can assume this since the inequality is symmetric). Then, ax ≥ by (since a, b,
x, y are nonnegative and a ≥ b and x ≥ y, where a ≥ b is because x ≥ y yields
ln x ≥ ln y ). Hence, Lemma 1 yields
|{z} |{z}
=a =b

x3 + 2xyz a + y 3 + 2xyz b + z 3 + 2xyz c ≥ (yz + zx + xy) (xa + yb + zc) .

1
The ”Vornicu-Schur inequality” that we use here is the following fact:
Let A, B, C be three reals, and let X, Y, Z be three nonnegative reals. If A ≥ B ≥ C and X ≥ Y,
then
X (A − B) (A − C) + Y (B − C) (B − A) + Z (C − A) (C − B) ≥ 0.
This is Theorem 1 a) in [1]. The proof is fairly easy (just show that X (A − B) (A − C) +
Y (B − C) (B − A) ≥ 0 and Z (C − A) (C − B) ≥ 0).

Mathematical Reflections 2 (2009) 32

 Since

x3 + 2xyz a + y 3 + 2xyz b + z 3 + 2xyz c

= x3 + 2xyz ln x + y 3 + 2xyz ln y + z 3 + 2xyz ln z

 3 3 3

= ln xx +2xyz y y +2xyz z z +2xyz

and

(yz + zx + xy) (xa + yb + zc) = (yz + zx + xy) (x ln x + y ln y + z ln z)

= (yz + zx + xy) ln (xx y y z z ) = ln (xx y y z z )yz+zx+xy ,

 3 3 3

this becomes ln xx +2xyz y y +2xyz z z +2xyz ≥ ln (xx y y z z )yz+zx+xy . Since the

ln function is strictly increasing, this yields

3 +2xyz 3 +2xyz 3 +2xyz
xx yy zz ≥ (xx y y z z )yz+zx+xy .

References

[1] Darij Grinberg, The Vornicu-Schur inequality and its variations, version 13
August 2007.
http://de.geocities.com/darij grinberg/VornicuS.pdf

Second solution by Minghua Lin, China

It suffices to show

(x3 + 2xyz) ln x + (y 3 + 2xyz) ln y + (z 3 + 2xyz) ln z

≥ x(xy + yz + xz) ln x + y(xy + yz + xz) ln y + z(xy + yz + xz) ln z,

i.e.

x(x − y)(x − z) ln x + y(y − x)(y − z) ln y + z(z − x)(z − y) ln z. (1)

without loss of generality, let x ≥ y ≥ z, then (x−y)(x−z) ln x+(y−x)(y−z ln y.

Therefore, (1) is valid.
The equality holds if and only if x = y = z.

Also solved by Paolo Perfetti, Universita degli studi di Tor Vergata, Italy;
Arkady Alt, San Jose, California, USA; Navid Safaei, Tehran, Iran; Daniel
Lasaosa, Universidad Publica de Navarra, Spain.

Mathematical Reflections 2 (2009) 33

U113. Find all continuous functions f : R → R such that f is a periodic function but
it does not have a least period.

Proposed by Radu Ţiţiu, Targu-Mures, Romania

First solution by Paolo Perfetti, Universita degli studi di Tor Vergata, Italy
Not having a least period means that for all r > 0 there exists Tr < r such
that f (x + Tr ) = f (x) for all x ∈ R. Let’s suppose that f is not constant. This
means that there exists two points, x1 > x0 such that f (x1 ) > f (x0 ). Without
loss of generality we may suppose f (x0 ) = 0 and f (x1 ) > 0. Let d be x1 − x0 .
The continuity of f implies that in an open neighborhood of x1 , say (a, b),
f (x) > f (x1 )/2 holds. Starting by x0 we move toward right doing jumps of
lengths Td/2 < d/2. The value assumed by f after every jump is 0 by periodicity
but after a while, inevitably we enter (a, b) and we fall in contradiction because
f should assume the value 0 and a value greater than f (x1 )/2.

Second solution by Arkady Alt, San Jose, California, USA

Let Ω+ be set of all positive periods of function f and let ω∗ = inf Ω+ .
Since by definition ”least period” mean least positive period then by problem’s
condition ω∗ ∈
/ Ω+ .Consider two cases:
1
1. ω∗ = 0. Then for any natural n there is ωn ∈ Ω+ such that 0 < ωn < .
n
 
x
Let x by any real number and let xn := x − ωn then 0 ≤ x < ωn and,
ωn
  
x
therefore, lim xn = 0. Since f (x) = f xn + ωn = f (xn ) and f is
n→∞ ωn
continuous
 
function then f (x) = lim f (x) = lim f (xn ) = f lim xn = f (0) .
n→∞ n→∞ n→∞
1
2. ω∗ > 0. Then for any natural n there is ωn ∈ Ω+ such that ω∗ < ωn < ω∗ + .
n
Let x be any real number. Since lim ωn = ω∗ and f is continuous function
n→∞
then
 
f (x) = lim f (x) = lim f (x + ωn ) = f lim (x + ωn ) = f (ω∗ ) .
n→∞ n→∞ n→∞
Thus we obtain that ω∗ is positive period of f , that is contradiction with
ω∗ ∈
/ Ω+
and this prove that only constant functions is solution of problem.

Third solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Mathematical Reflections 2 (2009) 34

 It is well known, that, given any two periods p1 , p2 of a periodic function f , then
for any pair of integers m, n, not both zero, |mp1 + np2 | is also a period. The
fact that |mp1 | is a period for any m may be trivially shown by induction, while
since |np2 | and |mp1 | are periods, then f (x + mp1 + np2 ) = f (x + mp1 ) = f (x).
1
Claim: For any positive integer N , there is a period p of f such that p ≤ N.

Proof: If any period of f is a rational multiple of a given period p0 , then every

period may be written as p0vu , where u, v are relatively prime positive integers.
If the v’s have a least common multiple V , then every period is a multiple
of pV0 , and a least period (which is one of the multiples of pV0 ) may be found.
Contradiction, hence the v’s do not have a least common multiple, and one may
p0
always find an integer W ≥ N p0 such that W ≤ N1 is a period. If periods p1
and p2 of f are not rational multiples of one another, then |mp1 + np2 | does not
have a lower bound; indeed, for any m, take n as the closest integer to − mp 1
p2 .
This results in 0 < |mp1 + np2 | < p2 . When at least N p2 such pairs (m, n) have
been chosen, by Dirichlet’s principle there are two, (m1 , n1 ) and (m2 , n2 ), such
that |m1 p1 + n1 p2 | − |m2 p1 + n2 p2 | < N1 , and in one of the cases where (m, n)
equals (m1 − m2 , n1 − n2 ) or (m1 + m2 , n1 + n2 ), then p = |mp1 + np2 | < N1
will be a period of f .

Consider now any real x0 6= 0. Since f is continuous, then given x0 , and for any
 > 0, there is a δ > 0 such that |f (x) − f (x0 )| <  whenever |x − x0 | < δ. For
given x0 and , and its corresponding δ, take N as the smallest integer which
is at least 1δ , and take any period p < N1 . Clearly, there is an integer value m
such that |x0 − mp| < δ, since otherwise it would be 2δ ≤ p < N1 . Since p is a
period, then |f (x0 ) − f (0)| = |f (x0 ) − f (mp)| < . Hence f (x0 ) = f (0) for all
x0 , and the only such functions f are constant functions.

Mathematical Reflections 2 (2009) 35

U114. Let a, b, c be nonnegative real numbers. Evaluate
n
1 X 1
lim p .
n→∞ n i2 + j 2 + ai + bj + c
i,j=1

Proposed by Ovidiu Furdui, Campia Turzii, Cluj, Romania

First solution by Paolo Perfetti, Universita degli studi di Tor Vergata, Italy
We show that the limit is independent on a, b, c allowing us to set a = b = c = 0
for evaluating it. If Q = [0, 1] × [0, 1], the limit becomes
n n Z Z
1 X 1 1 X 1 1
lim p = lim 2 q = p dxdy
n→∞ n 2
i +j 2 n→∞ n i2 j2 Q x2 + y2
i,j=1 i,j=1 + n2 n2
 
R1 Rx
By writing the integral as 2 0 0
√ 1 dy dx and passing to polar coordi-
x2 +y 2
nates we have
!
Z π/2 Z 1/ sin θ
ρ
Z π/2
1 θ π/2 √
2 dρ dθ = 2 dθ = 2 ln tan  = 2 ln( 2 + 1)
π/4 0 ρ π/4 sin θ 2 π/4

To show that the limit is independent by a, b, c, we prove

n n
1 X 1 1 X 1
lim p = lim p
n→∞ n i2 + j 2 + ai + bj + c n→∞ n i,j=1 i2 + j 2 + a0 i + b0 j + c0
i,j=1

for any a, b, c, a0 , b0 , c0 . We introduce a number of positive constants Ck , k =

0, 1, . . . . Since i|a0 − a| + j|b0 − b| + |c0 − c| ≤ C0 (i + j) and i2 + j 2 + ai + bj + c ≤
C1 (i2 + j 2 ) we have the bound
 
1 1
 
√ √
 i2 +j 2 +ai+bj+c − i2 +j 2 +a0 i+b0 j+c0  =

  −1
i(a0 −a)+j(b0 −b)+c0 −c  √ 1 1

 (i2 +j 2 +ai+bj+c)(i2 +j 2 +a0 i+b0 j+c0 )  √
+ 2 2 0 0 0 ≤
i2 +j 2 +ai+bj+c i +j +a i+b j+c
√
C0 (i+j) i2 +j 2 i+j
(i2 +j 2 )2 C1 = C2 (i2 +j 2 )3/2

Thus
n ∞ n ∞ n
1 X i+j 1 XX i 1 XX j √
2 2 3/2
≤ 3/2
+ 3/2
≤ C3 / n
n (i + j ) n (2ij) n (2ij)
i,j=1 j=1 i=1 i=1 j=1

Mathematical Reflections 2 (2009) 36

 and it follows that for any a, b, c, a0 , b0 , c0
n
!
1 X 1 1
lim p −p =0
n→∞ n i2 + j 2 + ai + bj + c i2 + j 2 + a0 i + b0 j + c0
i,j=1

In particular we can take a0 = b0 = c0 = 0 and write

!
1 1 1 1
p = p −p +p
i2 + j2 + ai + bj + c i2 + j2 + ai + bj + c + j2 i2 + j2 i2
√
The conclusion is that for any a, b, c the limit assumes the same value 2 ln( 2 +
1)

Second solution by John T. Robinson, Yorktown Heights, NY, USA

Let X 1
F (n) = p ,
1≤i,j≤n i2 + j2 + ai + bj + c
X 1
f1 (n) = √ ,
1≤i≤n−1
n2 + i2 + ai + bn + c
X 1
f2 (n) = p , and
1≤j≤n−1 n2 + j2 + an + bj + c
1
g(n) = √ .
2n2 + an + bn + c
Then
F (n) = F (n − 1) + f1 (n) + f2 (n) + g(n) .
√
Claim: lim (f1 (n) + f2 (n) + g(n)) = 2 ln(1 + 2). Since g(n) goes to zero, we
n→∞
need only consider f1 (n) and f2 (n). We can rewrite f1 (n) as follows, where
x = i/n and ∆x = 1/n:
X 1
f1 (n) = √ ∆x .
1≤i≤n−1
1 + x2 + ax∆x + b∆x + c∆x2

Therefore Z 1
1
lim f1 (n) = dx √
n→∞ 0 1 + x2
h p i1 √
= ln(x + 1 + x2 ) = ln(1 + 2) .
0
The argument for f2 is the same, establishing the claim. Finally, expressing
F (n) as X
F (n) = (f1 (k) + f2 (k) + g(k)),
1≤k≤n

Mathematical Reflections 2 (2009) 37

 √
we see that lim (1/n)F (n) = 2 ln(1 + 2).
n→∞

Also solved by Arkady Alt, San Jose, California, USA; Daniel Lasaosa, Uni-
versidad Publica de Navarra, Spain.

Mathematical Reflections 2 (2009) 38

 Olympiad problems

O109. Let a, b, c be positive real numbers such that abc = 1. Prove that

a+b+1 b+c+1 c+a+1 (a + 1)(b + 1)(c + 1) + 1

2 3
+ 2 3
+ 2 3
≤ .
a+b +c b+c +a c+a +b a+b+c

Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by Gheorghe Pupazan, Chisinau, Moldova

From the Cauchy-Schwartz inequality and because abc = 1 we obtain that
1 1 + a + ab
(a + b2 + c3 )(a + 1 + ab) ≥ (a + b + c)2 ⇐⇒ 2 3
≤ .
a+b +c (a + b + c)2

So we get that
a+b+1 (a + b + 1)(1 + a + ab)
2 3
≤ .
a+b +c (a + b + c)2
In a similar way we obtain the following inequalities

b+c+1 (b + c + 1)(1 + b + bc)

2 3
≤
b+c +a (a + b + c)2

and
c+a+1 (c + a + 1)(1 + c + ca)
2 3
≤ .
c+a +b (a + b + c)2
So it is enough only to prove that

(a + b + 1)(1 + a + ab) + (b + c + 1)(1 + b + bc) + (c + a + 1)(1 + c + ca)

(a + b + c)2

(a + 1)(b + 1)(c + 1) + 1
≤ .
a+b+c
The last one is equivalent to
X
(a + b + 1)(1 + a + ab) ≤ (a + b + c)(a + 1)(b + 1)(c + 1) + a + b + c
cyc
X X X X
⇐⇒ 3 a+3+ a2 + 2 ab + ab(a + b)
X X X X X
≤ abc · a + 3abc + ab(a + b) + a2 + 2 ab + 2 a
, which is true.

Mathematical Reflections 2 (2009) 39

 Second solution by Manh Dung Nguyen, Vietnam
By Cauchy-Schwarz Inequality, we have:

(a + b2 + c3 )(a + 1 + ab) ≥ (a + b + c)2

Similarly, we obtain

(a + b + 1)(a + 1 + ab) + (b + c + 1)(b + 1 + bc) + (c + a + 1)(c + 1 + ca)

LHS ≤
(a + b + c)2
(a + b + c)(ab + bc + ca + a + b + c + 3)
=
(a + b + c)2
(a + 1)(b + 1)(c + 1) + 1
= = RHS
a+b+c
Equality holds when a = b = c = 1.

Also solved by Navid Safaei, Tehran, Iran; Daniel Lasaosa, Universidad Publica
de Navarra, Spain.

Mathematical Reflections 2 (2009) 40

O110. Hexagon A1 A2 A3 A4 A5 A6 is inscribed in a circle C(O, R) and at the same time
circumscribed about a circle ω(I, r). Prove that if
1 1 1 1 1 1
+ + = + +
A1 A2 A3 A4 A5 A6 A2 A3 A4 A5 A6 A1
then one of its diagonals coincides with OI.

Proposed by Ivan Borsenco, Massachusetts Institute of Technology, USA

Solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

We will prove that, either I = O and the hexagon is regular, its three diagonals
meeting at I = O, or I 6= O and IO is one of the diagonals of the hexagon
which is also an axis of symmetry. We will use cyclic notation throughout the
problem, ie, Ai+6 = Ai for all i, and begin by proving the following

Lemma: Let A1 A2 A3 A4 A5 A6 be a circuminscribed hexagon, such that the

incircle touches side Ai Ai+1 at point Ti,i+1 . Denote di = Ai Ti,i+1 = Ai Ti−1,i ,
where clearly Ai Ai+1 = di + di+2 . Then, it is equivalent that (1) the hexagon is
regular and consequently its three diagonals meet at I = O, (2) I = O, (3) at
least three of the di are equal, and (4) each two opposite sides have the same
length.

Proof: If the hexagon is regular, by symmetry its three diagonals clearly meet at
I = O. If (1) is true, clearly so are (2), (3) and (4). Since IA2i = d2i +r2 , if (3) is
true, then three of the distances IAi are equal, and I = O, or (2) is true. If (2)
is true, then d√2i = IA2i − r2 = R2 − r2 for all i, and the hexagon is regular with
sidelengths 2 R2 − r2 . Since the hexagon admits an inscribed circle, Brian-
chon’s theorem guarantees that its diagonals Ai Ai+3 meet at a point P , where
clearly ∠Ai Ai+1 P = ∠P Ai+3 Ai+4 and ∠Ai+1 Ai P = ∠P Ai+4 Ai+3 because the
hexagon is cyclic, and triangles Ai P Ai+1 and Ai+3 P Ai+4 are similar. Therefore,
if (4) is true, triangles Ai P Ai+1 and Ai+3 P Ai+4 are equal, and Ai Ai+1 Ai+4 Ai+3
is a rectangle with center P = O and whose diagonals have thus length 2R, and
such that the incircle touches its sides Ai Ai+1√ and Ai+3 Ai+4 . Thus Ai+1 Ai+4 =
Ai Ai+3 = 2r, and Ai Ai+1 = Ai+3 Ai+4 = 2 R2 − r2 for all i. The lemma fol-
lows.
di +di+1 P Ai
Since triangles Ai P Ai+1 and Ai+3 P Ai+4 are similar, then di+3 +di+4 = P Ai+1 =
P Ai+1
P Ai+3 , or
P A1 P A5 P A3 (d1 + d2 )(d3 + d4 )(d5 + d6 )
1= = ,
P A5 P A3 P A1 (d2 + d3 )(d4 + d5 )(d6 + d1 )

Mathematical Reflections 2 (2009) 41

 and we may call p = (d1 + d2 )(d3 + d4 )(d5 + d6 ) = (d2 + d3 )(d4 + d5 )(d6 + d1 ).
Now,
d6 − d2 1 1 1 1 1 1
= − = + − − =
(d1 + d2 )(d1 + d6 ) A1 A2 A6 A1 A2 A3 A4 A5 A3 A4 A5 A6

d2 + d3 + d4 + d5 d3 + d4 + d5 + d6 (d3 + d4 + d5 − d1 )(d6 − d2 )
= − = .
(d2 + d3 )(d4 + d5 ) (d3 + d4 )(d5 + d6 ) p
If d2 6= d6 , then 0 = (d6 + d1 )(d1 + d2 )(d3 + d4 + d5 − d1 ) − p = (d3 − d1 )(A1 A2 −
A4 A5 ), and similarly for its cyclic permutations. Hence, if the hexagon is not
regular, wlog A1 A2 6= A4 A5 , resulting in d2 = d6 6= d4 ; similarly d1 , d3 , d5
cannot be all distinct. If d1 = d3 , then (d2 − d4 )(d5 − d1 ) = 0 and the hexagon
is regular, or d1 6= d3 , and similarly d1 6= d5 . Hence d3 = d5 6= d1 , leading to
A2 A3 = A5 A6 , A1 A2 = A1 A6 , and A3 A4 = A4 A5 , or triangles A6 A1 A2 and
A3 A4 A5 are isosceles at A1 and A4 respectively, while A3 A5 A6 A2 is an isosceles
trapezoid with A3 A5 k A6 A2 . The circumcenters of these three polygons lie
then on A1 A4 , the common perpendicular bisector of A2 A6 and A3 A5 . The
hexagon is clearly symmetric with respect to A1 A4 , and I and O lie on it. The
result follows.

Mathematical Reflections 2 (2009) 42

O111. Prove that for each positive integer n the number
      2       2
n n 2 n n n 2 n
+2 +2 + ··· +2 +2 + ···
0 2 4 1 3 5

is triangular.

Proposed by Titu Andreescu, University of Texas at Dallas, USA

First solution by Ercole Suppa, Teramo, Italy

We denote
      2       2
n n 2 n n n 2 n
f (n) = +2 +2 + ··· · +2 +2 + ···
0 2 4 1 3 5

For each real number x the the Binomial Theorem yields:

     
n n n n 2
(x + 1) = + x+ x + ··· (1)
0 1 2
     
n n n n 2
(x − 1) = − x+ x − ··· (2)
0 1 2
By summing (1) and (2) we have
     
1 n n 2 n 4
[(x + 1)n + (x − 1)n ] = + x + x + ··· (3)
2 0 2 4
     
1 n n n n 2 n 4
[(x + 1) − (x − 1) ] = + x + x ··· (4)
2x 1 3 5
√
From (3) and (4), taking x = 2, we obtain:

1h √ √ i n  n   
n n n 2
( 2 + 1) + ( 2 − 1) = + 2+ 2 + ··· (5)
2 0 2 4

1 h√ √ i n  n   
n 2
√ ( 2 + 1)n − ( 2 − 1)n = + 2+ 2 + ··· (6)
2 2 1 3 5
Thus
2  n
an + bn a − bn 2 1 an + bn 2 a − bn 2
     n 
f (n) = · √ = · (7)
2 2 2 2 2 2

Mathematical Reflections 2 (2009) 43

 √ √ n n
2
where a = 2 + 1 and b = 2 − 1. Let k = a −b 2 . Since ab = 1 we have:

a − bn 2
 n
a2n + b2n − 2an bn

k+1= +1= +1=
2 4
a2n + b2n − 2 + 4 a2n + b2n + 2an bn a + bn 2
 n 
= = =
4 4 2
Therefore  
k(k + 1) k+1
f (n) = =
2 2
and we are done.

Second solution by Darij Grinberg, Germany

First, we consider a more general setting:

Theorem  1. Let q be a real. Define a sequence (f0 , f1 , f2 , ...) by

P n k
fn = q for every n ∈ N, and define a sequence (g0 , g1 , g2 , ...)
k∈N 2k  
n
q k for every n ∈ N. 2 Then, for every n ∈ N,
P
by gn =
k∈N 2k + 1
we have the matrix equality
   n
fn qgn 1 q
= (3)
gn fn 1 1
and the equalities

fn2 − qgn2 = (1 − q)n ; (4)

2fn gn = g2n . (5)

For any a ∈ N and b ∈ N, we have

fa+b = fa fb + qga gb ; (6)

ga+b = fa gb + ga fb . (7)
2 P
A sum of the form a (k) (where a : N → R is some map) only makes sense if all but finitely
k∈N !
P n
many k ∈ N satisfy a (k) = 0. But this condition is easily verified for our sum q k (in fact,
k∈N 2k
! !
n n
all k ∈ N \ {0, 1, ..., n} satisfy k > n, thus 2k ≥ k > n, thus = 0, thus q k = 0; thus, all
2k 2k
! !
n k n
q k . Similar
P
but finitely many k ∈ N satisfy q = 0) and (similarly) for our sum
2k k∈N 2k + 1
P
arguments can show that all other sums of the form a (k) that we will encounter in our solution
k∈N
will be well-defined.

Mathematical Reflections 2 (2009) 44

 Remark. Here, N means the set {0, 1, 2, ...} .
Proof of Theorem 1. We will prove (1) by induction:
Induction base. We have
X 0  X  1, if k = 0;
k
f0 = q = · qk
2k 0, if k 6= 0
k∈N k∈N
= 1 · q0 = 1 · 1 = 1
     
0 1, if 2k = 0; 1, if k = 0;
since = =
2k 0, if 2k 6= 0 0, if k 6= 0
and
 
  
X  0 1, if 2k + 1 = 0;
0 X
 since 2k + 1 = = 0, 
g0 = qk = 0 · qk 0, if 2k + 1 6= 0
2k + 1
k∈N k∈N because k ∈ N yields 2k + 1 6= 0
= 0,

so that
       0
f0 qg0 1 q·0 1 0 1 q
= = = .
g0 f0 0 1 0 1 1 1

In other words, (1) holds for n = 0. This completes the induction base.
Induction step. Let N ∈ N. Assume that (1) holds for n = N. We have to show
that (1) holds for n = N + 1 as well.
Since (1) holds for n = N, we have
   N
fN qgN 1 q
= .
gN fN 1 1

Mathematical Reflections 2 (2009) 45

 But
X  N + 1 X  N   N 
k
fN +1 = q = + qk
2k 2k 2k − 1
k∈N k∈N
       
N +1 N N
as = + by the recurrence of the binomial coefficients
2k 2k 2k − 1
X N  X N  X N  X N 
= qk + qk = qk + qk
2k 2k − 1 2k 2k − 1
k∈N k∈N k∈N k∈N;
k≥1
X N  X N

= qk + q k+1
2k 2 (k + 1) − 1 |{z}
k∈N k∈N |
 {z  } =q·qk
N
=
2k + 1
(here we substituted k + 1 for k in the second sum)
X N X N 
= q k +q q k = 1 · fN + q · gN
2k 2k + 1
k∈N k∈N
| {z } | {z }
=fN =gN

P P
here we replaced the sign by an sign, since the addend for k = 0 is zero
 
 k∈N k∈N; 
    k≥1   
 N N N 
(as = = = 0 for k = 0)
2k − 1 2·0−1 −1
and
X N + 1 X  N   N 
k
gN +1 = q = + qk
2k + 1 2k 2k + 1
k∈N k∈N
       
N +1 N N
as = + by the recurrence of the binomial coefficients
2k + 1 2k 2k + 1
X N  X N 
= qk + q k = 1 · fN + 1 · gN ,
2k 2k + 1
k∈N k∈N
| {z } | {z }
=fN =gN

Mathematical Reflections 2 (2009) 46

 so that
   
fN +1 qgN +1 1 · fN + q · gN q (1 · fN + 1 · gN )
=
gN +1 fN +1 1 · fN + 1 · gN 1 · fN + q · gN
 
1 · fN + q · gN 1 · qgN + q · fN
=
1 · fN + 1 · gN 1 · qgN + 1 · fN
     N
1 q fN qgN 1 q 1 q
= =
1 1 gN fN 1 1 1 1
 N +1
1 q
= .
1 1

In other words, (1) holds for n = N + 1. This completes the induction step.
Thus, the induction proof is complete, so that (1) is proven for all n ∈ N.
Now, (2) follows from
   n 
fn qgn 1 q
fn2 − qgn2 = fn · fn − qgn · gn = det = det (by (1))
gn fn 1 1
  n
1 q
= det = (1 · 1 − q · 1)n = (1 − q)n .
1 1

For any a ∈ N and b ∈ N, we have

   a+b
fa+b qga+b 1 q
= (by (1), applied to n = a + b)
ga+b fa+b 1 1
 a  b
1 q 1 q
=
1 1 1 1
| {z }

| {z }

fa qga  fb qgb 
= =
ga fa gb fb
(by (1), applied to n=a) (by (1), applied to n=b)
    
fa qga fb qgb fa · fb + qga · gb fa · qgb + qga · fb
= =
ga fa gb fb ga · fb + fa · gb ga · qgb + fa · fb
 
fa fb + qga gb q (fa gb + ga fb )
= .
fa gb + ga fb fa fb + qga gb

Thus, fa+b = fa fb + qga gb and ga+b = fa gb + ga fb , so that (4) and (5) are
proven. For every n ∈ N, we have

g2n = gn+n = fn gn + gn fn (by (5))

= 2fn gn ,

and (3) follows.

Mathematical Reflections 2 (2009) 47

 Altogether, we have now proven Theorem 1.
From now on, we set q = 2. Then,
1 1 2
fn2 gn2 = (fn gn )2 = (2fn gn )2 = g2n (by (3))
4 4
1 1 2 1 1 2 2 2



= · · qg2n = · · f2n − f2n − qg2n
4 q 4 q
1 1  2 
= · · f2n − (1 − q)2n
4 q
 
2
since f2n − qg2n2
= (1 − q)2n , what results if we substitute 2n for n in (2)
 

1 1   1
1
= · · f 2 − (1 − 2)2n = 2
f2n − 1 = (f2n − 1) (f2n + 1)
 
4 2  2n | {z }  8 8
=(−1)2n =1,
since 2n is even
 
1 f2n − 1 f2n + 1 1 f2n − 1 f2n − 1
= · · = · · +1
2 2 2 2 2 2
f2n − 1
for every n ∈ N. Since ∈ Z for every n ∈ N (since
2
 
   
 2n P 2n k 
20 + 2 −1
   
P 2n k P 2n k
2·0 k∈N; 2k

q −1 2 −1
f2n − 1 k∈N 2k k∈N 2k k≥1
= = =
2  2  2 2
 
P 2n k 
1 + 2 −1

k∈N; 2k      
k≥1 2n 0 2n 0
= since 2 = 2 =1·1=1
2 2·0 0
 
P 2n k
2
k∈N; 2k
k≥1 X 2n
= = 2k−1 ∈ Z
2 2k
k∈N;
k≥1

), this yields that fn2 gn2 is a triangular number for every n ∈ N. This is exactly
what the problem asked us to prove.

Remark. Theorem
1 could be proved more quickly using the binomial formula
√ n √
n
applied to 1 + q and 1 − q . However, such a proof would fail if we
replace R by a field of characteristic 2 and q is a square in that field. The proof
given above works over any field and for any q. (Then again, from a deeper

Mathematical Reflections 2 (2009) 48

 viewpoint, it is just a straightforward elementarization of the proof using the
binomial formula.)

Also solved by John T. Robinson, Yorktown Heights, NY, USA; Brian Bradie,
Newport News, USA; Daniel Lasaosa, Universidad Publica de Navarra, Spain.

Mathematical Reflections 2 (2009) 49

O112. Let a, b, c be real positive numbers. Prove that

a3 + abc b3 + abc c3 + abc 3 a3 + b3 + c3

+ + ≥ · .
(b + c)2 (c + a)2 (a + b)2 2 a2 + b2 + c2

Proposed by Cezar Lupu, University of Bucharest, Romania and Pham Huu

Duc, Ballajura, Australia

Solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Call u = (a − b)(a − c), v = (b − c)(b − a), w = (c − a)(c − b). After some

algebra,
(2a + b + c)u + (2b + c + a)v + (2c + a + b)w =
= 2(a3 + b3 + c3 ) − (a2 b + a2 c + b2 c + b2 a + c2 a + c2 b) =
= 3(a3 + b3 + c3 ) − (a + b + c)(a2 + b2 + c2 ),
while at the same time
a3 + abc a a(2a − b − c) au
2
− = + .
(b + c) 2 2(b + c) (b + c)2

Now,
a(2a − b − c) b(2b − c − a) c(2c − a − b)
+ + =
2(b + c) 2(c + a) 2(a + b)
3(a3 + b3 + c3 ) − (a + b + c)(a2 + b2 + c2 )
= (a + b + c) =
2(a + b)(b + c)(c + a)
(2a + b + c)u + (2b + c + a)v + (2c + a + b)w
= (a + b + c) .
2(a + b)(b + c)(c + a)
Since
a+b+c 1 a3 + b3 + c3 − 2abc
− 2 =
(a + b)(b + c)(c + a) a + b2 + c2 (a + b)(b + c)(c + a)(a2 + b2 + c2 )

is clearly positive because a3 + b3 + c3 ≥ 3abc, and

(2a + b + c)u + (2b + c + a)v + (2c + a + b)w = (a + b + c)(u + v + w) + au + bv + cw

is clearly non-negative because am u + bm v + cm w ≥ 0 for any m ≥ 0 as a result

of Schur’s inequality, then it suffices to show that
au bv cw
2
+ 2
+ ≥ 0.
(b + c) (c + a) (a + b)2

Mathematical Reflections 2 (2009) 50

 Finally,

a(a + b)2 (a + c)2 u = a5 + 2a3 (ab + bc + ca) + a(ab + bc + ca)2 ,

and it suffices to show that

(a5 u+b5 v+c5 w)+2(ab+bc+ca)(a3 u+b3 v+c3 w)+(ab+bc+ca)2 (au+bv+cw) ≥ 0,

clearly true again by Schur’s inequality. The conclusion follows, equality being
reached if and only if a = b = c.

Also solved by Paolo Perfetti, Universita degli studi di Tor Vergata, Italy; Gheo-
rghe Pupazan, Chisinau, Moldova; Oleh Faynshteyn, Leipzig, Germany; Ercole
Suppa, Teramo, Italy; Darij Grinberg, Germany.

Mathematical Reflections 2 (2009) 51

O113. Let P be a point on the circumcircle Γ of a triangle ABC. The tangents from
P to the incircle of ABC meet again the circumcircle at X and Y , respectively.
Prove that the line XY is parallel to a side of triangle ABC if and only if P is
the tangency point of Γ with some mixtilinear incircle of triangle ABC.

Proposed by Cosmin Pohoata, Bucharest, Romania

Solution by Daniel Lasaosa, Universidad Publica de Navarra, Spain

Claim 1: If XY is a chord tangent to γ, and P is a point on Γ such that

IP 0 ⊥ X 0 Y 0 , where P 0 , X 0 , Y 0 are the respective inverses of P, X, Y with respect
to γ, then γ is the inscribed circle to P XY .

Proof: Let M be the midpoint of the chord XY that does not contain P . It
is well known that M is the circumcenter of IXY , hence the inverse of the
circumcircle of IXY with respect to γ is line X 0 Y 0 . Moreover, line IP is its
own inverse IP 0 with respect to γ. If IP 0 ⊥ X 0 Y 0 , then IP is perpendicular
to the circumcircle of IXY , or IP passes through M , ie, IP is the internal
bisector of ∠XP Y . By Poncelet’s porism, there is a triangle P U V that admits
I as its incircle, hence M is the midpoint of chord U V that does not contain
P . We conclude that U V and XY are tangent to γ, such that γ is inside
triangles P U V and P XY , and with XY k U V , hence XY = U V , and P XY
has inscribed circle γ. The claim follows.

Claim 2: Let A0 be the inverse of A with respect to γ. Then, circle γA 0 (A0 , r)

is the inverse with respect to γ of the mixtilinear incircle γA (IA , rA ) which is

tangent to Γ, AB and AC. Moreover, calling Q the point where γA and Γ
0 0 0 r


touch, and Q its inverse with respect to γ, then circle Γ O , 2 with diameter
A0 Q0 passes through A0 , is internally tangent to γA
0 , and is the inverse of Γ with

respect to γ.

Proof: It is well known that A, A0 , I, IA are collinear and they lie on the internal
bisector of ∠BAC. It is also well known (or easily provable) that r = rA cos2 A2 .
Denote by S, T the two points where the mixtilinear incircle γA intersects AI.
Then, IS = IIA − rA , IT = IIA + rA , where

r r r sin A2
IIA = AIA − AI = A A
− = .
sin 2 cos2 2 sin A2 cos2 A2
Moreover, denoting S 0 , T 0 the respective inverses of S, T with respect to γ,
IS 0 · IS = r2 and IT 0 · IT = r2 , and since IA0 · IA = r2 , we have
A
r cos2 A
A0 S 0 = IS 0 − IA0 = 2
− r sin = r,
1− sin A2 2

Mathematical Reflections 2 (2009) 52

 and similarly A0 T 0 = IT 0 + IA0 = r. We conclude that circle γA with diameter
ST transforms upon inversion into circle γA 0 with diameter S 0 T 0 . Now, since

γA is internally tangent to Γ at Q, its inverse Γ0 is internally tangent to γA 0 at

Q , or line A Q contains a diameter of Γ . But Γ passes through A , or A0 Q0

0 0 0 0 0 0

is a diameter of Γ0 , and the length of the radius of Γ0 is 2r . The claim follows.

Call now D the point where BC touches γ, and D0 the diametrally opposite
point to D in γ. Let the parallel to BC through D0 intersect Γ at
X and Y . The
inverse of line XY with respect to γ is clearly circle ΓXY N, 2r with diameter
ID0 , where N is the midpoint of ID0 .

Claim 3: A0 O0 IN is a parallelogram.

Proof: By the second claim, A0 O0 = 2r is a radius of Γ0 , while N I = 2r is a radius

of ΓXY . Now, since IA0 is the internal bisector of ∠BAC and IN = ID ⊥ BC,
then ∠A0 ID0 = B−C2 , or the Cosine Law yields

B−C A r2 A B−C
A0 N 2 = IA02 +IN 02 −2IA0 ·IN 0 cos = r2 sin2 + −r2 sin cos2 =
2 2 4 2 2
r2 r2 (R − 2r)
 
2 A B−C B+C
− r sin cos − cos = ,
4 2 2 2 4R
or A0 N = rOI A B C
2R , where we have used that r = 4R sin 2 sin 2 sin 2 and OI =
2

R2 − 2Rr. Finally, points J, K where the diameter OI intersects Γ are at

distances R − OI, R + OI from I, or their inverses J 0 , K 0 are at distances
r2 r2 0
R−OI , R+OI from I. Their midpoint is clearly O , or

IJ 0 − IK 0 r2 OI rOI
IO0 = = 2 2
= = A0 N.
2 R − OI 2R
The claim follows.

Consider now an arbitrary point P on Γ, and construct points X, Y as described

in the proposed problem. By Poncelet’s porism, XY is tangent to γ. Hence,
if XY k BC, XY must be the parallel to BC through D0 , which defines P
univocally. Moreover, the hypothesis of the third claim holds, and A0 O0 IN is
a parallelogram, or A0 , O0 , Q0 are collinear, and so are I, N, D0 , with A0 O0 =
O0 Q0 = IN = N D0 = 2r . Hence IQ0 k O0 N k A0 D0 . Now, X 0 Y 0 is a common
chord of Γ0 and ΓXY , or X 0 Y 0 ⊥ O0 N , and IQ0 ⊥ X 0 Y 0 . We may then apply the
first claim, and QXY is a triangle inscribed into Γ and circumscribed around
γ, and such that XY k BC. By Poncelet’s porism, there cannot be another
such point. By cyclic permutations, XY k CA and XY k AB iff P is the
point where Γ touches respectively the mixtilinear incircles touching AB, BC
and BC, CA. The result follows.

Mathematical Reflections 2 (2009) 53

O114. Prove that for all real numbers x, y, z the following inequality holds

(x2 + xy + y 2 )(y 2 + yz + z 2 )(z 2 + zx + x2 ) ≥ 3(x2 y + y 2 z + z 2 x)(xy 2 + yz 2 + zx2 )

Proposed by Gabriel Dospinescu, Ecole Normale Superieure, France

First solution by Darij Grinberg, Germany

The inequality in question immediately follows from the identity

y 2 + yz + z 2 z 2 + zx + x2 x2 + xy + y 2

= 3 x2 y + y 2 z + z 2 x xy 2 + yz 2 + zx2 + ((x − y) (y − z) (z − x))2 .

What remains is to prove this identity. Of course, we can prove it by expanding,

but here is a more conceptual proof:
Denote a = x2 y + y 2 z + z 2 x and b = xy 2 + yz 2 + zx2 .
√
1 + 3i
We work in C. Let ζ = . Then, ζ 3 = −1 and thus
2
(x + ζy) (y + ζz) (z + ζx)
   

=  ζ 3 + 1  xyz + ζ x2 y + y 2 z + z 2 x + ζ 2 xy 2 + yz 2 + zx2 = ζa + ζ 2 b

| {z } | {z } | {z }
=−1+1=0 =a =b
= ζ (a + ζb) .
 3
1 1
The same computation with ζ replaced by everywhere (and using = −1
ζ ζ
instead of ζ 3 = −1) proves
     
1 1 1 1 1
x+ y y+ z z+ x = a+ b .
ζ ζ ζ ζ ζ

But any two complex numbers u and v satisfy

 
1
(u + ζv) u + v = u2 + uv + v 2 (8)
ζ
   
1 2 1 1
(since (u + ζv) u + v = u + ζ + uv + v 2 and ζ + = 1 as we can
ζ ζ ζ
easily see).

Mathematical Reflections 2 (2009) 54

 Hence,

y 2 + yz + z 2 z 2 + zx + x2 x2 + xy + y 2




     
1 1 1
= (y + ζz) y + z (z + ζx) z + x (x + ζy) x + y
ζ ζ ζ
   
1
since (2) yields (y + ζz) y + z = y 2 + yz + z 2 ,

   ζ  


1 1
(z + ζx) z + x = z + zx + x and (x + ζy) x + y = x2 + xy + y 2
2 2
 
ζ ζ
   
1 1 1
= (x + ζy) (y + ζz) (z + ζx) · x + y y+ z z+ x
ζ ζ ζ
   
1 1 1
= ζ (a + ζb) · a + b = (a + ζb) a + b = a2 + ab + b2 (by (2))
ζ ζ ζ
= 3ab + (b − a)2
= 3 x2 y + y 2 z + z 2 x xy 2 + yz 2 + zx2 + ((x − y) (y − z) (z − x))2

(since a = x2 y + y 2 z + z 2 x, b = xy 2 + yz 2 + zx2 , and a quick computation shows

that

b − a = xy 2 + yz 2 + zx2 − x2 y + y 2 z + z 2 x = (x − y) (y − z) (z − x)

), qed.

Second solution by Magkos Athanasios, Kozani, Greece

If xyz = 0, say z = 0 the inequality is equivalent to x2 y 2 (x − y)2 ≥ 0, which
clearly holds. Assume now that xyz 6= 0. Dividing both sides of the inequality
by (xyz)2 we have to prove that

 2  2  2    
x x y y z z x y z x y z
+ +1 + +1 + +1 ≥ 3 + + + + .
y y z z x x z x y y z x

Set
x y z
= a, = b, = c.
y z x
Then abc = 1 and we have to prove that

(a2 + a + 1)(b2 + b + 1)(c2 + c + 1) ≥ 3(a + b + c)(ab + bc + ca).

multiplying out we arrive at

[ab + bc + ca − (a + b + c)]2 ≥ 0.

Mathematical Reflections 2 (2009) 55

 Third solution by Paolo Perfetti, Universita degli studi di Tor Vergata, Italy
It is trivial to observe that the l.h.s. and the r.h.s. are equal for x = y, for
y = z and for z = x. The ratio

(x2 + xy + y 2 )(y 2 + yz + z 2 )(z 2 + zx + x2 ) − 3(x2 y + y 2 z + z 2 x)(xy 2 + yz 2 + zx2 )

(x − y)(y − z)(z − x)

yields
−xy 2 + zx2 − z 2 x + xy 2 − y 2 z + yz 2
which annihilates for x = y, for y = z and for z = x as well. We have

−xy 2 + zx2 − z 2 x + xy 2 − y 2 z + yz 2
=1
(x − y)(y − z)(z − x)

and then
(x2 + xy + y 2 )(y 2 + yz + z 2 )(z 2 + zx + x2 )+
−3(x2 y + y 2 z + z 2 x)(xy 2 + yz 2 + zx2 ) =
= (x − y)2 (y − z)2 (z − x)2

and we are done.

Also solved by Hoang Quoc Viet; Gheorghe Pupazan, Chisinau, Moldova; Arkady
Alt, San Jose, California, USA; Navid Safaei, Tehran, Iran; Daniel Lasaosa,
Universidad Publica de Navarra, Spain; Nguyen Manh Dung, Hanoi University
of Science, Vietnam.

Mathematical Reflections 2 (2009) 56