Junior problems

J535. Solve in positive integers the equation

13
x3 − xy − y 3 = 2020.
2

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

Solution by Polyahedra, Polk State College, USA

Clearly, (x, y) = (13, 2) is a solution. We show that it is the only one. Taking the equation modulo 13
we get x3 − y 3 ≡ 5. Since {0, ±1, ±5} is the complete set of cubic residues modulo 13, we must have either
(x, y) = (13m, 13n − r) or (13m + r, 13n), with m ≥ n ≥ 1 and r ∈ {7, 8, 11}. If x = 13m + r and y = 13n, then
the left side of the equation equals
13 13
(x − y)3 + (3(x − y) − ) xy ≥ 73 + (21 − ) (20)(13) > 2020.
2 2
Consider then x = 13m and y = 13n − r. If n ≥ 2 then we similarly have
13
73 + (21 − ) (26)(15) > 2020.
2

Therefore, n = 1. If m ≥ 2, then (x − y)3 ≥ 203 > 2020. Hence m = 1, and it is easy to check that only r = 11
works.

Also solved by Corneliu Mănescu-Avram, Ploieşti, Romania; Nicuşor Zlota, Traian Vuia Technical Col-
lege, Focşani, Romania; Taes Padhihary, Disha Delphi Public School, India.

Mathematical Reflections 6 (2020) 1

J536. Let ABC be an acute triangle. Prove that

a+b 2 b+c 2 c + a 2 16 (a + b + c)2

( ) +( ) +( ) ≥ .
cos C cos A cos B 3
Proposed by Florin Rotaru, Focşani, Romănia

Solution by Nguyen Viet Hung, Hanoi University of Science, Vietnam

By the Cauchy-Schwarz inequality we have

b+c 2 c+a 2 a+b 2 1 b+c c+a a+b 2

( ) +( ) +( ) ≥ ( + + )
cos A cos B cos C 3 cos A cos B cos C
2
1 (b + c)2 (c + a)2 (a + b)2
= ( + + )
3 (b + c) cos A (c + a) cos B (a + b) cos C
2
1 (b + c + c + a + a + b)2
≥ ( )
3 (b + c) cos A + (c + a) cos B + (a + b) cos C
2
1 4(a + b + c)2
= ( )
3 (b cos C + c cos B) + (c cos A + a cos C) + (a cos B + b cos A)
2
1 4(a + b + c)2
= ( )
3 a+b+c
16(a + b + c)2
=
3
as desired.

Also solved by Daniel Văcaru, Piteşti, Romania; Taes Padhihary, Disha Delphi Public School, India;
Prodromos Fotiadis, Nikiforos High School, Drama, Greece; Corneliu Mănescu-Avram, Ploieşti, Romania;
Marin Chirciu, Colegiul Nat, ional Zinca Golescu, Pites, ti, Romania; Nicuşor Zlota, Traian Vuia Technical
College, Focşani, Romania; Arkady Alt, San Jose, CA, USA.

Mathematical Reflections 6 (2020) 2

J537. Solve in rational numbers the equation
x⌊x⌋{x} = 58,
where ⌊x⌋ and {x} are the greatest integer less than or equal to x and the fractional part of x,
respectively.

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

Solution by HyunBin Yoo, South Korea

Let x = α + β where α is an integer and β is a rational number which satisfies 0 < β < 1. Let β = n
m where
αn(αm+n)
gcd(n, m) = 1. Then x⌊x⌋{x} = (α + m) ⋅ α ⋅ m
n n
= m2
= 58.
Since gcd(m, n) = 1, gcd(m , n) = 1 and gcd(m , αm + n) = 1. Therefore, α must be a multiple of m2 in
2 2

order for the fraction to be an integer. Let α = m2 k, where k is a nonzero integer.

Then the above equation becomes (m3 k + n) ⋅ k ⋅ n = 58. If k > 0, (m3 k + n) is the biggest out of the three
terms so the possible cases are 58 ⋅ 1 ⋅ 1, 29 ⋅ 2 ⋅ 1 and 29 ⋅ 1 ⋅ 2. Out of these three, only 29 ⋅ 1 ⋅ 2 results in an
integral m = 3.
On the other hand, when k < 0, none of the possible combinations results in m being an integer. So, the
only solution is when m = 3, k = 1 and n = 2. In conclusion, the only rational solution x = m2 k+ m n
= 9+ 23 = 29
3 .

Also solved by Polyahedra, Polk State College, USA; Alina Craciun, Miron Costin Theoretical High School,
Pas, cani, România.

Mathematical Reflections 6 (2020) 3

J538. Prove that in any triangle ABC

(a + b)(b + c)(c + a) R
≤1+ .
4abc 2r

Proposed by Marius Stănean, Zalău, Romania

Solution by Arkady Alt, San Jose, CA, USA

Let s be semiperimeter of a triangle. Since ab + bc + ca = s2 + 4Rr + r2 , abc = 4Rrs then (a + b)(b + c)(c + a) =
(a + b + c) (ab + bc + ca) − abc = 2s (s2 + 4Rr + r2 ) − 4Rrs = 2s (s2 + 2Rr + r2 ) and

(a + b)(b + c)(c + a) R
≤1+ ⇐⇒
4abc 2r
2abcR
(a + b)(b + c)(c + a) ≤ 4abc + ⇐⇒
r
2s (s2 + 2Rr + r2 ) ≤ 16Rrs + 8R2 s ⇐⇒
s2 + 2Rr + r2 ≤ 4R2 + 8Rr ⇐⇒
s2 ≤ 4R2 + 6Rr − r2 .

Since s2 ≤ 4R2 + 4Rr + 3r2 (Gerretsen’s Inequality) and 2r ≤ R (Euler’s Inequality)

we get 4R2 + 6Rr − r2 − s2 = 4R2 + 6Rr − r2 − (4R2 + 4Rr + 3r2 ) + (4R2 + 4Rr + 3r2 − s2 ) =
2r (R − 2r) + (4R2 + 4Rr + 3r2 − s2 ) ≥ 0.

Also solved by Polyahedra, Polk State College, USA; Taes Padhihary, Disha Delphi Public School, India;
Prodromos Fotiadis, Nikiforos High School, Drama, Greece; Corneliu Mănescu-Avram, Ploieşti, Romania;
Chirciu, Colegiul Nat, ional Zinca Golescu, Pites, ti, Romania; Nicuşor Zlota, Traian Vuia Technical College,
Focşani, Romania; Ioan Viorel Codreanu, Satulung, Maramures, Romania; Miguel Amengual Covas, Cala
Figuera, Mallorca, Spain; Mihaly Bencze, Brasov, Romania; Nguyen Viet Hung, Hanoi University of Science,
Vietnam.

Mathematical Reflections 6 (2020) 4

J539. Let α > 0 be a real number. Prove that if a, b, c are real numbers in the interval [α, 30α], then
7 7 7 6 6 6
+ + ≥ + + .
a + 6b b + 6c c + 6a a + 5b b + 5c c + 5a

Proposed by Mihaela Berindeanu, Bucharest, România

Solution by Polyahedra, Polk State College, USA

Since 30b ≥ a,
7 6 1 1 (a − b)2 (30b − a)
− + − = ≥ 0.
a + 6b a + 5b 42a 42b 42ab(a + 6b)(a + 5b)
Summing with the other two analogous inequalities completes the proof.

Also solved by Taes Padhihary, Disha Delphi Public School, India.

Mathematical Reflections 6 (2020) 5

J540. Find all functions f ∶ R → R such that

f (x + ⌊y⌋) = ⌊f (x)⌋ + f (y),

for all x, y ∈ R.

Proposed by Besfort Shala, University of Primorska, Koper, Slovenia

First solution by Polyahedra, Polk State College, USA

Setting x = y = 0 yields 0 = ⌊f (0)⌋, so f (0) = c for some c ∈ [0, 1). Setting x = 0 gives f (⌊y⌋) = f (y) for all real
y. Let k = ⌊f (1)⌋. As an induction hypothesis, suppose that m is a nonnegative integer and f (m) = km + c.
Setting x = 1 and y = m we get f (1 + m) = k + km + c = k(1 + m) + c.
Therefore, for all nonnegative integer m, f (m) = km + c, and then letting x = m and y = −m, we also
have c = f (m − m) = km + f (−m), so f (−m) = k(−m) + c as well.
Hence, for all real y, f (y) = f (⌊y⌋) = k⌊y⌋ + c. Finally, if f (x) = k⌊x⌋ + c for any integer k and c ∈ [0, 1),
then both sides of the functional equation equal k⌊x⌋ + k⌊y⌋ + c for all real x, y.

Second solution by Joel Schlosberg, Bayside, NY, USA

For a function f ∶ R → R such that f (x + ⌊y⌋) = ⌊f (x)⌋ + f (y) for all x, y ∈ R, let c = f (1) − f (0) and d = f (0).
Then
⌊d⌋ = ⌊f (0)⌋ = f (0 + ⌊0⌋) − f (0) = 0
so d ∈ [0, 1), and
f (1) = ⌊f (0)⌋ + f (1) = f (0 + ⌊1⌋) = f (1 + ⌊0⌋) = ⌊f (1)⌋ + f (0)
so c = ⌊f (1)⌋ is an integer.
Since f (1) = c⌊1⌋ + d and if f (n) = c⌊n⌋ + d for some n ∈ N,

f (n + 1) = f (1 + ⌊n⌋) = ⌊f (1)⌋ + f (n) = c + c⌊n⌋ + d = c⌊n + 1⌋ + d

so by induction, f (n) = c⌊n⌋ + d for all n ∈ N. Also, f (0) = c⌊0⌋ + d and for n ∈ N

f (−n) = f (n + ⌊−n⌋) − ⌊f (n)⌋ = d − ⌊c⌊n⌋ + d⌋ = c⌊−n⌋ + d

so f (z) = c⌊z⌋ + d for all z ∈ Z.

Then for y ∈ R,
f (y) = f (0 + ⌊y⌋) − ⌊f (0)⌋ = f (⌊y⌋) = c⌊y⌋ + d.
Conversely, if f (x) = c⌊x⌋ + d for constants c ∈ Z and d ∈ [0, 1),

f (x + ⌊y⌋) = c⌊x + ⌊y⌋⌋ + d = c⌊x⌋ + c⌊y⌋ + d

= ⌊c⌊x⌋ + d⌋ + c⌊y⌋ + d = ⌊f (x)⌋ + f (y).

Also solved by Arkady Alt, San Jose, CA, USA; Alina Craciun, Miron Costin Theoretical High School,
Pas, cani, România; Taes Padhihary, Disha Delphi Public School, India.

Mathematical Reflections 6 (2020) 6

 Senior problems

S535. Find all triples (p, q, r) of primes such that

1 1 1 1
+ + =
p−1 q r+1 2

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

Solution by Li Zhou, Polk State College, USA

Clearly, (p, q, r) = (5, 5, 19), (13, 3, 11), and (43, 7, 2) are solutions. We show that they are the only ones.
First, we must have p ≥ 5 and q ≥ 3.
If p = 5, then the equation becomes (q − 4)(r − 3) = 16, yielding only (q, r) = (5, 19). If p = 7, then the
equation becomes (q − 3)(r − 2) = 9, yielding no solution. Consider p ≥ 11.
If q = 3, then the equation becomes (p − 7)(r − 5) = 36, yielding only (p, r) = (13, 11). If q = 5, then the
equation becomes (3p − 13)(3r − 7) = 100, yielding no solution. Consider q ≥ 7.
If r ≥ 3, then
1 1 1 1 1 1 1
+ + ≤ + + < .
p − 1 q r + 1 10 7 4 2
Thus r = 2, and the equation becomes (p − 7)(q − 6) = 36, yielding only (p, q) = (43, 7).

Also solved by Joel Schlosberg, Bayside, NY, USA; Taes Padhihary, Disha Delphi Public School, India.

Mathematical Reflections 6 (2020) 7

S536. Let S = {2, 4, 6, . . . , 2020} be the set of positive even integers not greater than 2020 and T = {3, 6, 9, . . . , 2019}
be the set of positive multiples of 3 less than 2020. Evaluate

∑ ∑ ∣A ∪ B∣.
A⊆S B⊆T

Proposed by Li Zhou, Polk State College, USA

Solution by the author

We first notice that ∣S∣ = 1010, ∣T ∣ = 673, and ∣S ∩ T ∣ = 336. For each i ∈ S, half of all 2∣S∣ subsets of S contain
i, so
∑ ∑ ∣A∣ = ∣S∣ ⋅ 2 ⋅ 2∣T ∣ = 1010 ⋅ 21682 .
∣S∣−1
A⊆S B⊆T

Likewise,
∑ ∑ ∣B∣ = 2 ⋅ ∣T ∣ ⋅ 2 = 673 ⋅ 21682 .
∣S∣ ∣T ∣−1
A⊆S B⊆T

Similarly, for each i ∈ S ∩ T , half of all 2∣S∣ subsets of S contain i and half of all 2∣T ∣ subsets of T contain i,
thus
∑ ∑ ∣A ∩ B∣ = ∣S ∩ T ∣ ⋅ 2 ⋅ 2∣T ∣−1 = 336 ⋅ 21681 .
∣S∣−1
A⊆S B⊆T

Finally, from the beloved PIE we get

∑ ∑ ∣A ∪ B∣ = 2 (∣S∣ + ∣T ∣) − 2∣S∣+∣T ∣−2 ∣S ∩ T ∣ = 1515 ⋅ 21682 .

∣S∣+∣T ∣−1
A⊆S B⊆T

Mathematical Reflections 6 (2020) 8

S537. Let ABC be a triangle with ∠B = 50○ . Let D be a point on the segment BC such that ∠BAD = 30○
and AD = BC. Find ∠CAD.

Proposed by Marius Stănean, Zalău, Romania

First solution by Li Zhou, Polk State College, USA

Let E be the reflection of D across AB, then ∠AEB = ∠ADB = 100○ = ∠CBE and AE = AD = CB, so
AEBC is an isosceles trapezoid. Therefore, ∠CAB = ∠EBA = 50○ , thus ∠CAD = 20○ .

Second solution by Brian Bradie, Christopher Newport University, Newport News, VA, USA
Applying the Law of Sines to ∆BAD yields
BD AD BC
2BD = = = ;
sin 30 ○ sin 50 ○ sin 50○
thus,
BC = 2 sin 50○ BD and DC = (2 sin 50○ − 1)BD.
Now, let x denote the measure of ∠CAD. Applying the Law of Sines to ∆CAD yields

2 sin 50○ BD (2 sin 50○ − 1)BD

=
sin(100○ − x) sin x
or
2 sin 50○ sin(100○ − x)
= = sin 100○ cot x − cos 100○ = cos 10○ cot x + sin 10○ .
2 sin 50○ − 1 sin x
Next, multiply the numerator and denominator of the fraction on the left side by cos 50○ to obtain
2 sin 50○ sin 100○ cos 10○
= =
2 sin 50○ − 1 sin 100○ − cos 50○ cos 10○ − cos 50○
cos 10○
=
cos(30○ − 20○ ) − cos(30○ + 20○ )
cos 10○
= ;
sin 20○
thus,
cos 10○ − sin 10○ sin 20○
cot x = .
cos 10○ sin 20○
But,
cos 10○ = cos(20○ − 10○ ) = cos 20○ cos 10○ + sin 20○ sin 10○ ,
so
cos 10○ − sin 10○ sin 20○ = cos 20○ cos 10○
and
cot x = cot 20○ .
Therefore, ∠CAD = 20○ .

Also solved by Joel Schlosberg, Bayside, NY, USA; Ivko Dimitric, Pennsylvania State University Fayette,
Lemont Furnace, PA, USA; Alina Craciun, Miron Costin Theoretical High School, Pas, cani, România; Taes
Padhihary, Disha Delphi Public School, India; Arighna Pan, Nabadwip Vidyasagar College, India; Corne-
liu Mănescu-Avram, Ploieşti, Romania; Nicuşor Zlota, Traian Vuia Technical College, Focşani, Romania;
Miguel Amengual Covas, Cala Figuera, Mallorca, Spain.

Mathematical Reflections 6 (2020) 9

S538. Let P (x) = xn + a1 xn−1 + a2 xn−2 + . . . + an−1 x + an be a polynomial with real coefficients and n an even
positive integer. If P (x) has n non-negative real roots, prove that
√ √ a1
1+ n
an ≤ n P (−1) ≤ 1 − .
n

Proposed by Nguyen Viet Hung, Hanoi University of Science, Vietnam

Solution by Joel Schlosberg, Bayside, NY, USA

By comparing coefficients of (x − r1 )(x − r2 ) ⋅ . . . ⋅ (x − rn ) = xn + a1 xn−1 + a2 xn−2 + ⋯ + an−1 x + an (or via
Viete’s formulas),
√ n
1/n
1 + n an = 1 + ∏ ri
i=1

and
a1 1 n
1− = 1 + ∑ ri
n n i=1
√
while n P (−1) = ∏ni=1 (1 + ri )1/n .
1/n 1/n
According to Mahler’s inequality, for any q1 , . . . , qn , r1 , . . . , rn ≥ 0, ∏ni=1 qi + ∏ni=1 ri ≤ ∏ni=1 (qi + ri )1/n
so for q1 = ⋯ = qn = 1,
√ n
1/n
n √
1 + n an = 1 + ∏ ri ≤ ∏(1 + ri )1/n = n P (−1).
i=1 i=1

By the arithmetic mean-geometric mean inequality,

√ n
1 n 1 n a1
n
P (−1) = ∏(1 + ri )1/n ≤ ∑(1 + ri ) = 1 + ∑ ri = 1 − .
i=1 n i=1 n i=1 n

Also solved by Li Zhou, Polk State College, USA; Arkady Alt, San Jose, CA, USA; Nicuşor Zlota, Traian
Vuia Technical College, Focşani, Romania; Taes Padhihary, Disha Delphi Public School, India; Arighna Pan,
Nabadwip Vidyasagar College, India.

Mathematical Reflections 6 (2020) 10

S539. There is a street with n houses on the left side and n houses on the right side. Moreover, the houses
which are in front of one to another are identical. We have to paint the houses with m different colors
in such a way that no two neighboring houses nor two face to face houses are of the same color. In
how many ways can this coloring be done?

Proposed by Mircea Becheanu, Canada

Solution by Nicole Lipschitz and Arturo Acunã, San José, Costa Rica
Let n = 1. Then, the number of ways to color the houses would be m(m − 1), as we will have m options for
the first house to be colored, lets call it a1 , and (m − 1) options for the one infront, b1 , since it must be a
different color.

a1 b1

Now, we will prove that, for any 2n houses (n on each side of the street) and m colors, the number of
ways to color them (lets call it k(n)) is:

m(m − 1)((m − 1) + (m − 2)2 )n−1 .

Assume this is true for all positive integers smaller or equal to n and lets name the houses on the right
hand side a1 , a2 , ⋯, an+1 , and the ones on the left hand side b1 , b2 , ⋯, bn+1 . Therefore, we know that there are
k(n) ways to color the first n houses on both sides of the street, so we only have the houses an+1 and bn+1
left to color.

Mathematical Reflections 6 (2020) 11

 a1 b1

an−1 bn−1

an bn

an+1 bn+1

Lets color first an+1 . Since the adjecent house, an , was already colored, there are (m − 1) posible ways
to color an+1 , yet we will separate it in two cases. On one hand, if the color of an+1 is the same as bn , then
there are (m − 1) color options for bn+1 (since it must be different from the house infront and adjecent to
itself, which are both colored in the same way). On the other hand, if the color of an+1 is any of the other
(m − 2) colors, there are (m − 2) color options for bn+1 . So, altogether, there are (m − 1) + (m − 2)2 ways to
color an+1 and bn+1 .
Therefore, the number of ways to color the 2(n+1) houses would be equal to:

k(n) ⋅ ((m − 1) + (m − 2)2 ) = m(m − 1)((m − 1) + (m − 2)2 )n−1 ⋅ ((m − 1) + (m − 2)2 )

= m(m − 1)((m − 1) + (m − 2)2 )n

Hence, we have concluded our proof for this problem.

Also solved by Li Zhou, Polk State College, USA.

Mathematical Reflections 6 (2020) 12

S540. Let s(x) denote the sum of digits of the positive integer x. Find all positive integers n such that
s(n) = 3s(n + 1).

Proposed by Titu Andreescu, USA and Marian Tetiva, România

Solution by Joel Schlosberg, Bayside, NY, USA

For a positive integer n, let z be the number of rightmost 9s in the decimal representation of n, and let m
be the integer formed when the z 9s are removed (if z = 0, m = n; if all of the digits of n are 9, m = 0). The
decimal representation of n + 1 replaces the z rightmost 9s by zeroes, adds 1 to the rightmost digit of m or
changes m from zero to one, and leaves any other digits of m unchanged. Therefore

s(n) = s(m) + 9z

and
s(n + 1) = s(m) + 1.
If s(n) = 3s(n + 1), s(m) = 9z−3
2 . Since s(m) is an integer, z is odd (and thus nonzero). Conversely, if z is
an odd positive integer and s(m) = 9z−3
2 , s(n) = 3s(n + 1).

Also solved by Li Zhou, Polk State College, USA; Corneliu Mănescu-Avram, Ploieşti, Romania; Nicuşor
Zlota, Traian Vuia Technical College, Focşani, Romania.

Mathematical Reflections 6 (2020) 13

 Undergraduate problems

U535. Let (an )n≥1 be a sequence of real numbers such that

1 n ak
lim ∑ = 1.
n→∞ n
k=1 k

Evaluate

2n
a1 + ... + ak
lim ∑ .
n→∞
k=n+1 k3

Proposed by Florin Stanescu, România

Solution by Joel Schlosberg, Bayside, NY, USA

By the Stolz-Cesaro theorem,

∑nk=1 ak /k an ∑n ak
lim n = 1 Ô⇒ lim = 1 Ô⇒ lim k=1 = 1.
n→∞ ∑ n→∞ ∑n k
k=1 1
n→∞ n
k=1

Let
a1 + ⋯ + ak
mn = min
k∈{n+1,...,2n} k2
and
a1 + ⋯ + ak
Mn = max .
k∈{n+1,...,2n} k2
Since a1 + ⋯ + ak is asymptotic to 1 + ⋯ + k = 21 k(k + 1), both mn and Mn tend to 1/2 as n → ∞. Since

2n
1 2n
a1 + ⋯ + ak 2n
1
mn ∑ ≤ ∑ 3
≤ M n ∑ ,
k=n+1 k k=n+1 k k=n+1 k

and the well-known approximation ∑nk=1 k1 = ln n + γ + O(1/n) with γ the Euler-Mascheroni constant implies
k=n+1 k = ln 2, by the squeeze theorem
that limn→∞ ∑2n 1

2n
a1 + ⋯ + ak ln 2
lim ∑ = .
n→∞
k=n+1 k3 2

Also solved by Arkady Alt, San Jose, CA, USA; Moubinool Omarjee, Lycée Henri IV, Paris, France.

Mathematical Reflections 6 (2020) 14

U536. Evaluate √ √
∫ x( 1 + x + 1 − x) dx.

Proposed by Nguyen Viet Hung, Hanoi University of Science, Vietnam

First solution by Henry Ricardo, Westchester Area Math Circle

We use integration by parts:
dv
u ³¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ·¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ µ
© √ √ 2x 2
∫ x ( 1 + x + 1 − x) dx = ((1 + x)3/2 − (1 − x)3/2 ) − ∫ (1 + x)3/2 − (1 − x)3/2 dx
3 3
2x √ 2x √
= (1 + x) 1 + x − (1 − x) 1 − x
3 3
4
+ ((1 + x) + (1 − x)5/2 ) + C
5/2
15
2(1 − x)5/2 2(1 − x)3/2 2(1 + x)5/2 2(1 + x)3/2
= − + − +C
5 √ 3 5 √ 3
(6x2 + 2x − 4) 1 + x (6x2 − 2x − 4) 1 − x
= + + C.
15 15

Second solution by Henry Ricardo, Westchester Area Math Circle

We write √ √ √ √
I = ∫ x( 1 + x + 1 − x) dx = ∫ x 1 + x dx + ∫ x 1 − x dx = I1 + I2 .

In I1 , make the substitution t2 = 1 + x, and in I2 let u2 = 1 − x. This gives us

I = I1 + I2 = ∫ (2t4 − 2t2 ) dt + ∫ (2u4 − 2u2 ) du

2 2 2 2
= t5 − t3 + u5 − u3 + C
5 3√ 5 3 √ √ √
2(1 + x)2 1 + x 2(1 + x) 1 + x 2(1 − x)2 1 − x 2(1 − x) 1 − x
= − + − + C
5 √ 3 √ 5 3
(6x2 + 2x − 4) 1 + x (6x2 − 2x − 4) 1 − x
= + + C.
15 15

Also solved by Olimjon Jalilov, National University of Uzbekistan, Tashkent, Uzbekistan; Brian Bradie,
Christopher Newport University, Newport News, VA, USA; Taes Padhihary, Disha Delphi Public School,
India; Alina Craciun, Miron Costin Theoretical High School, Pas, cani, România; Corneliu Mănescu-Avram,
Ploieşti, Romania; Marin Chirciu, Colegiul Nat, ional Zinca Golescu, Pites, ti, Romania; Marin Chirciu, Co-
legiul Nat, ional Zinca Golescu, Pites, ti, Romania; G. C. Greubel, Newport News, VA, USA; Mihaly Bencze,
Brasov, Romania; Moubinool Omarjee, Lycée Henri IV, Paris, France; Nicuşor Zlota, Traian Vuia Technical
College, Focşani, Romania.

Mathematical Reflections 6 (2020) 15

U537. Let k be a positive integer. Evaluate

arctan nk arctan(nk + 1)
lim n2k ( − ).
n→∞ nk nk + 1

Proposed by Dinu Ovidiu Gabriel, Bălceşti, Vâlcea, România

Solution by Brian Bradie, Christopher Newport University, Newport News, VA, USA
With
π 1 π 1 1
arctan nk = − arctan k = − k + O ( 3k )
2 n 2 n n
and
π 1 π 1 1
arctan(nk + 1) = − arctan k = − k +O( k ),
2 n +1 2 n +1 (n + 1)3
it follows that
arctan nk arctan(nk + 1) π 1 2nk + 1 1
− = ⋅ − + O ( 4k ) ,
n k n +1
k 2 n (n + 1) n (n + 1)
k k 2k k 2 n
and
arctan nk arctan(nk + 1) π n2k 2nk + 1 1
n2k ( − ) = ⋅ − + O ( 2k ) .
nk nk + 1 2 nk (nk + 1) (nk + 1)2 n
Thus,
arctan nk arctan(nk + 1) π
lim n2k ( − )= .
n→∞ n k n +1
k 2

Also solved by Alina Craciun, Miron Costin Theoretical High School, Pas, cani, România; Moubinool
Omarjee, Lycée Henri IV, Paris, France; Olimjon Jalilov, National University of Uzbekistan, Tashkent,
Uzbekistan; Arkady Alt, San Jose, CA, USA.

Mathematical Reflections 6 (2020) 16

U538. Let p > 2 be a prime and r > 1 be an integer. Define the set

S = {a ∈ Z∣aa ≡ a (mod pr ), 2 ≤ a ≤ pr − 2}.

Prove that
∣S∣ ≤ pr−2 (p − 1)(p − ϕ(p − 1)).

Proposed by Navid Safaei, Sharif University of Technology, Tehran, Iran

Solution by the author

A the outset, we prove that if xx ≡ x (mod pr ) then gcd(x, p) = 1, otherwise, since x(xx−1 − 1) ≡ 0 (mod pr ).
Since x ≡/ 0 (mod pr ), then xx ≡ 1 (mod pr ). Thus, gcd(x, p) = 1. Now, let us prove the following lemma.

Lemma: Let x ∈ S, then x is not a primitive root mod pr

Proof: Assume as a contraction that pr−1 (p − 1) divides x − 1. Write x = k(pr−1 (p − 1)), for some positive
integer k, then k(pr−1 (p − 1)) ≤ pr − 3, thus k = 1. Therefore, x = pr − pr−1 + 1. Since vp ((pr − pr−1 + 1)2p − 1) = r,
we find that x2p ≡ 1 (mod ()pr ). Hence, 2p is divisible by pr−1 (p − 1) and pr−1 (p − 1) ≤ 2p. It would be absurd
unless p = 3, r = 2. In that case x = 7 is the only solution and 7 is not a primitive root mod 9. That is, since
ord79 = 3 and we are done.

Back to our problem, since the total number of primitive roots modulo pr is ϕ(pr−1 (p − 1)) = pr−2 (p −
1)ϕ(p − 1), we find that

∣S∣ ≤ pr−1 (p − 1) − pr−2 (p − 1)ϕ(p − 1) = pr−2 (p − 1)(p − ϕ(p − 1)).

Mathematical Reflections 6 (2020) 17

U539. Calculate
1 −2x2 ln x
∫ √ dx
0 1 − x2 (x4 − x2 + 1)

Proposed by Paolo Perfetti, Roma, Italy

Solution by the author

1 −2x2 ln x 1 −2x2 ln x 1
∫ √ dx = ∫ √ √ √ dx =
0 1 − x2 (x4 − x2 + 1) 0 1 − x2 (1 − x 1 − x2 )(1 + x 1 − x2 )
1 −x2 ln x 1 1
=∫ √ ( √ + √ ) dx
0 1−x 1−x 1−x
2 2 1 + x 1 − x2

x = cos t with −π/2 ≤ t ≤ 0 yields

1 −x2 ln x dx 0 −(cos t)2 ln(cos t)
∫ √ √ =∫ (− sin t)dt
0 1 − x2 1 − x 1 − x2 −π/2 − sin t(1 + sin t cos t)

x = cos t with 0 ≤ t ≤ π/2 yields

1 −x2 ln x dx π/2 −(cos t)2 ln(cos t)
∫ √ √ = −∫ (− sin t)dt
0 1 − x2 1 + x 1 − x2 0 sin t(1 + sin t cos t)
By summing we get
π/2 −(cos t)2 ln(cos t)
∫ dt
−π/2 1 + sin t cos t
By x = arctan t we get
+∞ ln(1 + t2 )
∫ dt
−∞ 2(1 + t2 )(1 + t + t2 )
Ln(z − tk )
By introducing the complex functions Fk (z) = , k = 0, 1, t0 = i, t1 = −i z ∈ C, Ln(z) = Ln(ρeit ) ≐
(1 + z 2 )2
ln ρ + it, we can write
+∞ Ln(t − t0 ) +∞ Ln(t − t1 )
∫ dt + ∫ dt
−∞ 2(1 + t )(1 + t + t )
2 2 −∞ 2(1 + t2 )(1 + t + t2 )

To evaluate the first integral, we cut the complex plane along the set Imz = [i, +∞) and we take −3π/2 ≤
Argz ≤ π/2. We perform the clockwise integral over the two curves

γ1 (t) = {z ∈ C ∶ z = t, −r ≤ t ≤ r},
γ2 (t) = {z ∈ C ∶ z = re−it , −π ≤ t ≤ 0},

lim ∫ F0 (z)dz = −2πiRes(F0 (−i))

r→∞ γ1 ∪γ2

The integral over γ2 tends to zero by observing that ∣z − t0 ∣ ≤ ∣z∣ + 1

Ln(z − t0 ) ln(∣z∣ + 1) + π ln(∣z∣ + 1) + π

∣F0 (z)∣ = ∣ ∣≤ ≤C
(1 + z )(1 + z + z )
2 2 ∣∣z∣ − 1∣ (∣z∣ − ∣z∣ − 1)
2 2 ∣z∣4

if ∣z∣ ≥ rC > 0 and rC is large enough. Thus we have

0 ln(r + 1) + π
∣∫ F0 (z)dz∣ = ∣∫ F0 (re−it )(−i)re−it dt∣ = rπC →0
γ2 −π r4

Mathematical Reflections 6 (2020) 18

Clearly
+∞ ln(t − i)
lim ∫ F0 (z)dz = ∫ dt
r→∞ γ1 −∞ (1 + t2 )(1 + t + t2 )
and analogously occurs for F1 (z).

Ln(−2i) Ln(−2i) ln 2 iπ
Res(F0 (−i)) = = = +
−2i ⋅ −i −2 −2 4

√
2 −i − i)
3
− i)
−2iπ
Ln(e 3 Ln( −1
Res(F0 (e )) = = =
−2πi
3
2
(1 + e
)(2e + 1) (1 + e )(2e + 1)
−4iπ −2iπ −4iπ −2iπ
3 3 3 3
√ √
+ − + 3) − 1 √ √
1 7iπ 7iπ
ln(2 3) ln(2 7iπ
2
√ √ =
12
√ 6
= (i 3 + 3)(ln(2 + 3) − )
( 2 + i 2 )(−i 3)
1 3 −i 3 + 3 12 6

Doing analogous calculation for the integral with k = 1 we get (in this case we cut in the complex plane in
Imz = [−i, +∞), −π/2 ≤ Argz ≤ 3π/2 and the path

γ3 (t) = {z ∈ C ∶ z = t, −r ≤ t ≤ r},
γ4 (t) = {z ∈ C ∶ z = reit , 0 ≤ t ≤ π},

is run counterclockwise
Ln(2i) ln 2 +
iπ
ResF1 (i) = = 2
2i ⋅ i −2
√
ln(2 + 3) + i 7π
2iπ 1
2iπ Ln(e 3 + i)
ResF1 (e 3 )= 4iπ 2iπ = 2
√ √
12
)(1 + 2e 3 )
(1 + e 3 ( 12 − i 23 )i 3
1 √ √ 7iπ
= (3 − i 3)(ln(2 + 3) + )
12 6

Finally the integral is

1 − ln 2 iπ 1 √ √ 7iπ
− ⋅ 2πi ( + + (i 3 + 3)(ln(2 + 3) − )) +
2 2 4 12 6
1 − ln 2 iπ 1 √ √ 7iπ −π 2 π √
⋅ 2πi ( − + (3 − i 3)(ln(2 + 3) + )) = + √ ln(2 + 3)
2 2 4 12 6 12 2 3

Mathematical Reflections 6 (2020) 19

 √ √
U540. Let x be a rational number. Prove that Q( 3 x) = Q( 3 2) if and only is x = 2q 3 or x = 4q 3 , for some
rational number q.

Proposed by Mircea Becheanu, Canada

First solution by Gabriel Dospinescu, France

It is clear that the condition x = 2q 3 or x = 4q 3 , with q ∈ Q is sufficient.√ We will
√ prove that the condition is
√
necessary as well. Let a, b, c be rational numbers such that 3 x = a + b 3 2 + c 3 4. We introduce the√rational
polynomial P (X) = a + bX + cX 2 . Then P (X) √
3
− x = F (X) is also a rational polynomial and F ( 3 2) = 0.
Since X 3 −√ 2 is the minimal polynomial of 3 2 over √ Q, it follows that X 3 −√2∣F (X). Therefore, we also
√ √ √
have F (ω 2) = 0, where ω + ω + 1 = 0. From P (ω 2)3 = x we obtain P (ω 3 2) ∈ { 3 x, ω 3 x, ω 2 3 x}. We
3 2 3

distinguish three
√ cases:
√
Case 1: P (ω 3 2) = 3 x. It means that
√ √ √ √
a + b 2ω + c 4ω 2 = a + b 2 + c 4.
3 3 3 3

Then we have √ √
b + c 2 = bω + c 2ω 2 .
3 3

taking the complex conjugation one obtains also

√ √
b + c 2 = bω 2 + c 2ω.
3 3

By adding these equalities and using the realtion ω + ω 2 = −1 we obtain:

√ √
2b + 2c 2 = −b − c 2.
3 3

√ √
This gives b +√c 3 2 = 0 which means that b = c = 0 or 3 2 is rational. This case is impossible.
√
Case 2: P (ω 3 2) = ω 3 x. It means that
√ √ √ √ √ √
a + b 2ω + c 4ω 2 = aω + b 2ω + c 4ω ⇔ a + c 4ω 2 = aω + c 4ω ⇔
3 3 3 3 3 3

√ √
⇔ a(1 − ω) = c 4ω(1 − ω) ⇔ a = c 4ω.
3 3

√ √
From this we √ have a = c = 0 which gives 3 x = b 3 2 and x = 2b3 .
√
Case 3: P (ω 3 2) = ω 2 3 x. It means that
√ √ √ √ √ √
a + b 2ω + c 4ω 2 = aω 2 + b 2ω 2 + c 4ω 2 ⇔ a + b 2ω = aω 2 + b 2ω 2 .
3 3 3 3 3 3

WE multiply the last equality by ω to obtain

√ √ √ √
a + b 2 = aω + b 2ω 2 ⇔ a(1 − ω) = b 2(ω 2 − 1) ⇔ a = −b 2(ω + 1) ⇔
3 3 3 3

√
a = b 2ω 2 .
3

This shows that a = b = 0 and then x = 4c3 .

Mathematical Reflections 6 (2020) 20

 Second solution by Corneliu Mănescu-Avram, Ploieşti, Romania
√
Note that x ≠ 0, otherwise Q( 3 x) = Q(0) =√Q. √
√
If x = 2q 3 , with q ∈ Q∗ , then Q( 3 x) = Q(q 3 2) = Q( 3 2).
√ √
If x = 4q 3 , with q ∈ Q∗ , then Q( 3 x) = Q ( √
3 ) = Q(
2q 3
2).
2
√ √ √ √
Conversely, suppose √ that Q( x
3) = Q( 3
2) for some
√ q ∈ Q ∗
. The
√ set√ {1, Q( 3
2), Q( 3
4)} is the base of
the vector
√ space
√
3
2) over This means that 3
2) = {a + b 3
2√+ c 3
4∣a, b, c ∈ and that the equality
√
Q( Q. Q( Q}
a + b 3 2 + c 3 4 = 0, a, b, c ∈ Q implies
√ that a
√ = b = c = 0. From 3
x ∈ 3
2) we deduce that there are a, b, c ∈ Q
√
Q(
not all zero, such that x = a + b 2 + c 4. Then we have to prove that a = b = 0 or a = c = 0. Calculating
3 3 3

third powers, we have

√ √
x = a3 + 2b3 + 4c3 + 12abc + 3(a2 b + 2b2 c + 2c2 a) 2 + 3(ab2 + 2bc2 + ca2 ) 4,
3 3

whence, we deduce that

a2 b + 2b2 c + 2c2 a = 0 and ab2 + 2bc2 + ca2 = 0 (1)
√
If b = c = 0, then x = a ∈ Q, contradiction.
3

If c = 0, then the equalities (1) become a2 b = ab2 = 0, therefore a = 0, b ≠ 0.

If b = 0, then the equalities (1) become c2 a = ca2 = 0, therefore a = 0, c ≠ 0.

Indeed, if we put b/a = u, c/b = v and c/a = uv and dividing equalities (1) by a3 one obtains u + 2u3 v +
2u v = 0 and uv + u2 + 2u3 v 2 = 0. Now, we multiply the first equality by v and comparing the two equalities
2 2

one have u2 = 2u2 v 3 . This is a contradiction, as u and v are rational numbers.

Mathematical Reflections 6 (2020) 21

 Olympiad problems

O535. Let ABCD be a convex quadrilateral with an incircle. Let ω1 , ω2 , ω3 , ω4 be the excircles of ABCD
tangent to segments AB, BC, CD, DA, respectively. Prove that the lengths of the internal common
tangent segments to the circles (ω1 , ω3 ) and (ω2 , ω4 ) are equal.

Proposed by Waldemar Pompe, Warsaw, Poland

Solution by Li Zhou, Polk State College, USA

Let I, r, and s be the incenter, inradius, and semi-perimeter of ABCD, and Ii and ri be the center and radius
of ωi . Label other points as in the figure. First, it is easy to see that AJ = AS = BE = BF , AH = AE = BS,
DH = DG = CT , and DK = DT = CG = CF . Next, since △JI1 A ∼ △HAI, JI1 /AJ = HA/IH, that is,
r1 = AH ⋅ BE/r. Likewise, r3 = CF ⋅ DG/r. Therefore,

AH ⋅ BE ⋅ CF ⋅ DG AE ⋅ BF ⋅ CG ⋅ DH
r1 r3 = = = r2 r4 .
r2 r2
Now, △P XI1 ∼ △I3 Y P , so P X/XI1 = I3 Y /Y P , that is, P X ⋅ P Y = r1 r3 . Similarly, QX ⋅ QY = r1 r3 .
Therefore, P X(P Q − QY ) = (P Q − P X)QY , thus P X = QY . Hence, P X + P Y = QX + QY = QJ + QK =
JK = s, that is, P X and QY are the smaller zero of the quadratic polynomial x2 − sx + r1 r3 = x2 − sx + r2 r4 .
Finally, XY = XQ − QY = JK − 2QK = s − 2QY , completing the proof.

Mathematical Reflections 6 (2020) 22

O536. Let a, b, c be the side-lengths of a triangle and let S be its area. Let R and r be the circumradius and
inradius of the triangle, respectively. Prove that
√
6r
a2 + b2 + c2 ≤ 4S + 3 [(a − b)2 + (b − c)2 + (c − a)2 ] .
R

Proposed by Marius Stănean, Zalău, Romania

First solution by the author

The inequality can be rewritten as
√
6r
16(ab + bc + ca) − 5(a + b + c) ≤ 4S
2
.
R

But ab + bc + ca = s2 + r2 + 4Rr, so this becomes

√
6r
s2 + sr ≥ 4r2 + 16Rr,
R
or √
s2 s r 6r 4r2 16r
2
+ ( )( ) ≥ + .
R R R R R2 R
If we denote x2 = 1 − 2r
R ∈ [0, 1), then by Blundon’s Inequality

s2 (1 − x2 )2 (1 − x)(x + 3)3
≥ 2 + 5(1 − x 2
) − − 2x 3
= .
R2 4 4
Hence, it suffices to prove that
√
(1 − x)(x + 3)3 1 − x2 3(1 − x2 )(1 − x)(x + 3)3
+ ≥ (1 − x2 )2 + 8(1 − x2 ),
4 2 4
or √
1 − x2 3(1 − x2 )(1 − x)(x + 3)3 (x − 1)2 (x + 3)(5x + 3)
≥ ,
2 4 4
or
(x − 1)2 (x + 3)2 [3(x + 1)3 (x + 3) − (5x + 3)2 ] ≥ 0,
that is
x2 (x − 1)2 (x + 3)2 (3x2 + 18x + 11) ≥ 0,
clearly true. The equality holds when x = 0 so when the triangle is equilateral.
Note: The inequality is a slightly stronger version of Hadwiger-Finsler Reverse Inequality.

Mathematical Reflections 6 (2020) 23

 Second solution by Li Zhou, Polk State College, USA
Let s = (a + b + c)/2. From Gerretsen’s inequality s2 ≥ 16Rr − 5r2 and Euler’s inequality R ≥ 2r we have
s2 ≥ 16Rr − 5Rr/2 = 27Rr/2. Using the well-known relations S = rs and ab + bc + ca = 4Rr + r2 + s2 we get
√
6r
4S + 3 [(a − b)2 + (b − c)2 + (c − a)2 ] − a2 − b2 − c2
R
√
6r
= 4rs + 20s2 − 16 (4Rr + r2 + s2 )
R
√ √ √
6r 6r ⎛ 27Rr ⎞
≥ 4rs − 36r2 = 4r s− ≥ 0.
R R⎝ 2 ⎠

Also solved by Corneliu Mănescu-Avram, Ploieşti, Romania; Marin Chirciu, Colegiul Nat, ional Zinca
Golescu, Pites, ti, Romania; Ioan Viorel Codreanu, Satulung, Maramures, Romania; Nicuşor Zlota, Traian
Vuia Technical College, Focşani, Romania; Arkady Alt, San Jose, CA, USA.

Mathematical Reflections 6 (2020) 24

O537. Let i1 < ⋯ < il and j1 ≤ ⋯ ≤ jm be nonnegative integers such that

2i1 + ⋯ + 2il = 2j1 + ⋯ + 2jm .

Prove that l ≤ m.

Proposed by Titu Andreescu, USA and Marian Tetiva, România

First solution by Li Zhou, Polk State College, USA

If l > m, then there is a minimal k ∈ {1, . . . , m} such that jk > ik , thus

2j1 + ⋯ + 2jk−1 ≤ 2i1 + ⋯ + 2ik−1 < 2ik−1 +1 ≤ 2ik < 2jk .

Therefore, in modulo 2jk , 2j1 + ⋯ + 2jm ≡ 2j1 + ⋯ + 2jk−1 , while 2i1 + ⋯ + 2il has a residue greater than or equal
to 2i1 + ⋯ + 2ik , which is a desired contradiction.

Second solution by Joel Schlosberg, Bayside, NY, USA

Given any finite sum 2j1 + ⋯ + 2jm where j1 , . . . , jm are nonnegative integers, replace any identical summands
2j , 2j with 2j+1 . If the replacement process is iterated, eventually no duplicates will remain, since the number
of terms in the sum is a positive integer which decreases with each substitution. Rearrange the terms of the
resulting sum in increasing order as 2i1 + ⋯ + 2il , so that i1 < . . . < il . Each substitution leaves the total sum
unchanged, so 2i1 + ⋯ + 2il = 2j1 + ⋯ + 2jm .
The digits in the binary representation of 2i1 + ⋯ + 2il uniquely determine i1 , . . . , il , so the replacement
process applied to any sum 2j1 + ⋯ + 2jm equaling 2i1 + ⋯ + 2il will yield the same sequence of increasing
values i1 , . . . , il . Since the number of terms in the sum is nonincreasing during the process, l ≤ m.

Mathematical Reflections 6 (2020) 25

O538. Find the greatest constant λ such that the inequality

a2 + b2 + c2 2 a b c
+ λ ≥ (1 + λ) ( + + )
ab + bc + ca 3 b+c c+a a+b
holds for all positive real numbers a, b, c.

Proposed by Nguyen Viet Hung, Hanoi University of Science, Vietnam

Solution by the author

Multiplying the both sides by ab + bc + ca, the inequality becomes
2 a b c
a2 + b2 + c2 + λ(ab + bc + ca) ≥ (1 + λ) ( + + ) (ab + bc + ca),
3 b+c c+a a+b
or
2 abc abc abc
a2 + b2 + c2 + λ(ab + bc + ca) ≥ (1 + λ) (a2 + b2 + c2 + + + ),
3 b+c c+a a+b
or
1 1 1
(1 − 2λ)(a2 + b2 + c2 ) + 3λ(ab + bc + ca) ≥ 2(1 + λ)abc ( + + ),
b+c c+a a+b
or
2bc b + c
(1 − 2λ)(a2 + b2 + c2 − ab − bc − ca) ≥ (1 + λ) ∑ a ( − ),
cyc b+c 2
or
(1 + λ)a
∑ (1 − 2λ + ) (b − c)2 ≥ 0,
cyc b+c
or
Sa (b − c)2 + Sb (c − a)2 + Sc (a − b)2 ≥ 0
where
(1 + λ)a
Sa = 1 − 2λ + ,
b+c
(1 + λ)b
Sb = 1 − 2λ + ,
c+a
(1 + λ)c
Sc = 1 − 2λ + .
a+b
Now we take c = b then obtain
(Sb + Sc )(a − b)2 ≥ 0.

Mathematical Reflections 6 (2020) 26

 This is equivalent to
Sb + Sc ≥ 0,
or
2(1 + λ)b
2 − 4λ + ≥ 0,
a+b
or
a + 2b
≥ λ,
2a + b
or
t+2 a
≥ λ (t = ).
2t + 1 b
This is true for all t > 0 if and only if
t+2 1
λ ≤ inf = .
t>0 2t + 1 2
Next, we will show that the inequality holds for λ = 12 . Indeed, for λ = 12 , the inequality becomes

a(b − c)2
∑ ≥0
cyc b+c

which is clearly true and we are done.

Also solved by Arkady Alt, San Jose, CA, USA; Nicuşor Zlota, Traian Vuia Technical College, Focşani,
Romania.

Mathematical Reflections 6 (2020) 27

O539. Let A1 A2 B1 B2 C1 C2 be a convex hexagon, in which all angles are obtuse. Let A1 A2 ∩B1 B2 = C, B1 B2 ∩
C1 C2 = A and C1 C2 ∩ A1 A2 = B. Let O be the circumcenter of ABC. Suppose that ∠B2 OC1 =
∠BAC, ∠C2 OA1 = ∠CBA and ∠A2 OB1 = ∠ACB. Prove that

A1 A2 + B1 B2 + C1 C2 ≤ A2 B1 + B2 C1 + C2 A1 .

Proposed by Dominik Burek, Krakow, Poland

Solution by the author

We will use the following fact: let ABC be an acute triangle with circumcenter O. Let X and Y be points on
sides AB and AC, respectively, with ∠Y OX = ∠BAC. Prove that the perimeter of triangle AXY is greater
than or equal to BC.
Indeed, let X ′ be a point such that △OAX ≡ △OBX ′ . Let Y ′ be a point such that △OAY ≡ △OCY ′ .
Then OX ′ = OX, OY = OY ′ , and

∠X ′ OY ′ = ∠BOC − ∠BOX ′ − ∠Y ′ OC = 2∠BAC − ∠AOX − ∠Y OA

= 2∠BAC − ∠Y OX = ∠BAC.

It follows that △OXY ≡ △OX ′ Y ′ . In particular, XY = X ′ Y ′ . Hence

Perimeter(AXY ) = AX + XY + Y A = BX ′ + X ′ Y ′ + Y ′ C ≥ BC.

Mathematical Reflections 6 (2020) 28

O540. Let a, b, c be positive real numbers such that a + b + c = 3. Prove that
√ √ √
a+ b+ 3 c
3 3
a b c 7
√ +√ +√ + ≥ .
3
4(b6 + c6 ) + 7bc 3
4(c6 + a6 ) + 7ca 3
4(a6 + b6 ) + 7ab 12 12

Proposed by Hoang Le Nhat Tung, Hanoi, Vietnam

Solution by the author √ √ √

We have b6 +c6 = (b2 +c2 )(b4 −b2 c2 +c4 ) = (b2 +c2 ) [(b2 + c2 )2 − (bc 3)2 ] = (b2 +c2 )(b2 −bc 3+c2 )(b2 +bc 3+c2 ).
By AM-GM for three positive real numbers we have:
√ √ √ √ √ √ 9b2 − 12bc + 9c2
4(b6 + c6 ) = (b2 + c2 ) ⋅ 2(2 + 3)(b2 − bc 3 + c2 ) ⋅ 2(2 − 3)(b2 + bc 3 + c2 ) ≤ ⇔
3 3

3
√
3
4(b6 + c6 ) ≤ 3b2 − 4bc + 3c2 ⇔
√
4(b6 + c6 ) + 7bc ≤ 3b2 + 3bc + 3c2 ⇔
3

1 1 a a
√ ≥ 2 + bc + c2 )
⇔ √ ≥ 2 + bc + c2 )
3
4(b + c ) + 7bc
6 6 3(c 3
4(b + c ) + 7bc
6 6 3(b

Similarly for b and c. Now,

a b c a b c
√ +√ +√ ≥ + +
3
4(b6 + c6 ) + 7bc
3
4(c + a ) + 7ca
6 6 3
4(a + b ) + 7ab 3(b + bc + c ) 3(c + ca + a ) 3(a + ab + b2 )
6 6 2 2 2 2 2

(1)
Next, by Cauchy-Schwarz inequality, we have:

a b c a2 b2 c2
+ + = + + ≥
b2 + bc + c2 c2 + ca + a2 a2 + ab + b2 ab2 + abc + ac2 bc2 + bca + ba2 ca2 + cab + cb2
(a + b + c)2
(ab2 + abc + ac2 ) + (bc2 + bca + ba2 ) + (ca2 + cab + cb2 )
which equals to
a+b+c
ab + bc + ca
It follows that
a b c a+b+c 3
+ 2 + 2 ≥ = (2)
b2 + bc + c2 c + ca + a 2 a + ab + b2 ab + bc + ca ab + bc + ca
From (1) and (2) we conclude that

a b c 1
√ +√ +√ ≥ (3)
3
4(b6 + c6 ) + 7bc 3
4(c6 + a6 ) + 7ca 3
4(a6 + b6 ) + 7ab ab + bc + ca
√ √ √ √√ √ √
√ By AM-GM inequality for five positive real numbers: 3 a + 3 a + 3 a + a2 + a2 ≥ 5 5 3 a ⋅ 3 a ⋅ 3 a ⋅ a2 ⋅ a2 =
√ √
5 a5 = 5a ⇔ 3 3 a + 2a2 ≥ 5a ⇔ 3 3 a ≥ 5a − 2a2 , similarly for b and c.
5

Mathematical Reflections 6 (2020) 29

 Hence,
√ √ √
3( 3 a + b + 3 c) ≥ 5(a + b + c) − 2(a2 + b2 + c2 ) ⇔
3

√ √ √
3( 3 a + b + 3 c) ≥ 15 − 2(a2 + b2 + c2 )) ⇔
3

√ √ √
3( 3 a + b + 3 c + 1) ≥ 18 − 2(a2 + b2 + c2 ) = 2(a + b + c)2 − 2(a2 + b2 + c2 ) ⇔
3

√ √ √
3( 3 a + b + 3 c + 1) ≥ 2(a2 + b2 + c2 + 2ab + 2bc + 2ca) − 2(a2 + b2 + c2 ) ⇔
3

√ √ √
3( 3 a + b + 3 c + 1) ≥ 4(ab + bc + ca) ⇔
3

√ √ √
3( 3 a + b + 3 c) ≥ 4(ab + bc + ca) − 3 (4)
3

From (3) and (4) we conclude that

√ √ √
a+ b+
3 3 3
a b c c
√ +√ +√ + ≥
3
4(b6 + c6 ) + 7bc 3 4(c6 + a6 ) + 7ca 3 4(a6 + b6 ) + 7ab 12

1 ab + bc + ca 1
+ − (5)
ab + bc + ca 9 12
By AM-GM we get
√ √
1 ab + bc + ca 1 ab + bc + ca 1 2
+ ≥2 ⋅ =2 =
ab + bc + ca 9 ab + bc + ca 9 9 3
(5) ⇔
√ √ √
a+ b+ 3 c 2 1
3 3
a b c 7
√ +√ +√ ≥ ≥ − =
3
4(b6 + c6 ) + 7bc 3 4(c6 + a6 ) + 7ca 3 4(a6 + b6 ) + 7ab 12 3 12 12

and we are done. Equality occurs when a = b = c = 1.

Mathematical Reflections 6 (2020) 30