Professional Documents
Culture Documents
(a + b)2 c2
+ ≥ 4b.
c a
Proposed by Titu Andreescu, University of Texas at Dallas, USA
(a + b + c)2
≥ 4b.
a+c
The preceding inequality is equivalent to (a − b + c)2 ≥ 0, and the problem is
solved.
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).
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.
Thus, there are two n such that τ (n) = 6 and 3φ(n) = 7!:
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
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
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.
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
we have
−gab − gac + gbc = 0 ;
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.
Also solved by Arkady Alt, San Jose, California, USA; Paolo Leonetti, Milano,
Italy.
x = za0 c0 y .
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:
(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 .
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
1 7 1 7 1 7
a− 2
= ,b − 2 = ,c − 2 = .
b 4 c 4 a 4
Pairwise subtractions of these equations yield
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).
AV WD
= (2)
VC DX
so that we have from (1) and (2),
WE WD
=
EC DX
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.
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.
1 a a b b c c
a b c
⇐⇒ (s − c) (s − a) (s − b) ≤ .
2a+b+c 2 2 2
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.
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
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.
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
a2 + ac + c2 − b2 a2 + ab + b2 − c2
BV = BA1 − V A1 = , CV = a − BV = .
2a + b + c 2a + b + c
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
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,
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
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.
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
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.
(m + n)2 − 2n2 = 1 .
x2 − 1 = (x − 1)(x + 1) = 2n2 .
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 )
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.
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→∞
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.
Proof. We have
Also solved by Arkady Alt, San Jose, California, USA; Navid Safaei, Tehran,
Iran; Daniel Lasaosa, Universidad Publica de Navarra, Spain; Paolo Leonetti,
Milano, Italy.
ax (x − y) (x − z) + by (y − z) (y − x) + cz (z − x) (z − y) ≥ 0.
This rewrites as
Thus,
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
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).
and
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
References
[1] Darij Grinberg, The Vornicu-Schur inequality and its variations, version 13
August 2007.
http://de.geocities.com/darij grinberg/VornicuS.pdf
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.
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.
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.
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
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
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
Also solved by Arkady Alt, San Jose, California, USA; Daniel Lasaosa, Uni-
versidad Publica de Navarra, Spain.
O109. Let a, b, c be positive real numbers such that abc = 1. Prove that
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
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 + 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.
Similarly, we obtain
Also solved by Navid Safaei, Tehran, Iran; 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
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 )
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.
is triangular.
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
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.
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
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
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
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
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
Also solved by John T. Robinson, Yorktown Heights, NY, USA; Brian Bradie,
Newport News, USA; Daniel Lasaosa, Universidad Publica de Navarra, Spain.
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 )
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.
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.
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
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.
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
IJ 0 − IK 0 r2 OI rOI
IO0 = = 2 2
= = A0 N.
2 R − OI 2R
The claim follows.
y 2 + yz + z 2 z 2 + zx + x2 x2 + xy + y 2
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
b − a = xy 2 + yz 2 + zx2 − x2 y + y 2 z + z 2 x = (x − y) (y − z) (z − x)
), qed.
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
[ab + bc + ca − (a + b + c)]2 ≥ 0.
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
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.