Volume/tome 48, issue/numéro 3

March/mars 2022
Crux Mathematicorum is a problem-solving journal at the secondary and university undergraduate levels,
published online by the Canadian Mathematical Society. Its aim is primarily educational; it is not a research
journal. Online submission:
https://publications.cms.math.ca/cruxbox/

Crux Mathematicorum est une publication de résolution de problèmes de niveau secondaire et de premier
cycle universitaire publiée par la Société mathématique du Canada. Principalement de nature éducative,
le Crux n’est pas une revue scientifique. Soumission en ligne:
https://publications.cms.math.ca/cruxbox/

The Canadian Mathematical Society grants permission to individual readers of this publication to copy articles for
their own personal use.
c CANADIAN MATHEMATICAL SOCIETY 2022. ALL RIGHTS RESERVED.
ISSN 1496-4309 (Online)
La Société mathématique du Canada permet aux lecteurs de reproduire des articles de la présente publication à des
fins personnelles uniquement.

c SOCIÉTÉ MATHÉMATIQUE DU CANADA 2022. TOUS DROITS RÉSERVÉS.

ISSN 1496-4309 (électronique)

Supported by / Soutenu par :

• Intact Financial Corporation
• University of the Fraser Valley

Editorial Board

Editor-in-Chief Kseniya Garaschuk University of the Fraser Valley

MathemAttic Editors John McLoughlin University of New Brunswick

Shawn Godin Cairine Wilson Secondary School
Kelly Paton Quest University Canada
Olympiad Corner Editors Alessandro Ventullo University of Milan
Anamaria Savu University of Alberta
Articles Editor Robert Dawson Saint Mary’s University
Associate Editors Edward Barbeau University of Toronto
Chris Fisher University of Regina
Edward Wang Wilfrid Laurier University
Dennis D. A. Epple Berlin, Germany
Magdalena Georgescu BGU, Be’er Sheva, Israel
Chip Curtis Missouri Southern State University
Philip McCartney Northern Kentucky University

Guest Editors Yagub Aliyev ADA University, Baku, Azerbaijan

Andrew McEachern York University
Vasile Radu Birchmount Park Collegiate Institute
Vanessa Radzimski University of the Fraser Valley
Aaron Slobodin University of Victoria
Chi Hoi Yip University of British Columbia
Samer Seraj Existsforall Academy
Translators Rolland Gaudet Université de Saint-Boniface
Frédéric Morneau-Guérin Université TÉLUQ
Editor-at-Large Bill Sands University of Calgary
 IN THIS ISSUE / DANS CE NUMÉRO

118 In Memoriam
119 MathemAttic Article Contest winners announcement
120 MathemAttic: No. 33
121 Problems: MA161–MA165
123 Solutions: MA136–MA140
128 Problem Solving Vignettes: No. 21 Shawn Godin
133 From the bookshelf of ... Asia Matthews
135 From the lecture notes of ... Fok-Shuen Leung
138 Olympiad Corner: No. 401
138 Problems: OC571–OC575
141 Solutions: OC546–OC550
146 Focus On . . . : No. 50 Michel Bataille
152 Problems: 4721–4730
156 Bonus Problems: B101–B125
160 Solutions: 4671–4680

Crux Mathematicorum
Founding Editors / Rédacteurs-fondateurs: Léopold Sauvé & Frederick G.B. Maskell
Former Editors / Anciens Rédacteurs: G.W. Sands, R.E. Woodrow, Bruce L.R. Shawyer,
Shawn Godin

Crux Mathematicorum
with Mathematical Mayhem
Former Editors / Anciens Rédacteurs: Bruce L.R. Shawyer, James E. Totten, Václav Linek,
Shawn Godin
118/ In Memoriam

In Memoriam
Yulia Zdanovskaya, a 21-year old mathematician, was killed on March 8th, 2022
during a Russian forces attack on Kharkiv, Ukraine.
In 2017, Yulia represented Ukraine at the European Girls’ Mathematical Olympiad
and won a silver medal at the competition. Currently, Yulia was a student in
computer mathematics program at the Kiev National University. Passionate about
math, computer science and teaching, she recently joined Teach for Ukraine to
become a mentor and a role model for new generations of young people. Yulia was
from Kharkiv and refused to leave her hometown; the last message she sent was
“Thanks, but I will stay in Kharkiv until our victory”.

Crux Mathematicorum, Vol. 48(3), March 2022

 MathemAttic Article Contest winners /119

MathemAttic Article Contest

winners

We received several excellent submissions to our MathemAttic Article Contest and

are happy to announce three winners (listed alphabetically by last name):
• Amit Kumar Basistha with the article “Pell’s equation and problems solving”
• Alex Bloom with the article “Telescoping sums”
• Rishi Nair with the article “Introduction to Knot theory”
The articles will appear in the upcoming April, May and June issues of Crux in
the MathemAttic section.
Congratulations to the winners!

Copyright © Canadian Mathematical Society, 2022

120/ MathemAttic

MATHEMATTIC
No. 33
The problems featured in this section are intended for students at the secondary school
level.

Click here to submit solutions, comments and generalizations to any

problem in this section.

To facilitate their consideration, solutions should be received by June 1, 2022.

MA161. On an n × n board, there are 21 dominoes. Each domino covers

exactly two squares and no two dominoes touch one another, even at a corner.
Determine the minimal value of n.

MA162. From an 8 × 8 chessboard, the central 2 × 2 block rises up to form a

barrier. Queens cannot be placed on the barrier, and may not attack one another
across this barrier. Determine the maximal number of Queens which can be placed
on the chessboard so that no two attack each other.

MA163. In the diagram, the log A has radius R. A hole of radius r is drilled
through the centre of log A at right angles to the axis. Another log B of radius r
passes through the hole. Find the length S in terms of R and r.

MA164. ABCD is a parallelogram. M and N are points on the sides AB

and AD respectively, such that AB = 4AM and AD = 3AN . Let K be the point
AC
of intersection of M N and AC. Determine the ratio AK .

Crux Mathematicorum, Vol. 48(3), March 2022

 MathemAttic /121

MA165. Proposed by Dominique Mouchet.

In the diagram below, ADE is an equilateral triangle with O on DA and C on ED
such that OABC is a square. Let r represent the radius of the circles inscribed in
triangles CDO and AF B, r0 represent the radius of the circle inscribed in triangle
ECF and R represent the radius of the circle tangent to OA, AF and CO. Show
that Rr0 = r2 .

.................................................................

Les problèmes dans cette section sont appropriés aux étudiants de l’école secondaire.

Cliquez ici afin de soumettre vos solutions, commentaires ou

généralisations aux problèmes proposés dans cette section.

Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au plus tard le 1 juin 2022.

MA161. Sur un échiquier de taille n × n, on retrouve 21 dominos. Chaque

domino occupe deux carrés contigus; aussi, les dominos ne se touchent pas, deux
dominos n’ayant même pas un coin en commun. Déterminer la valeur minimale
de n.

MA162. Sur un échiquier de taille 8 × 8, on place au centre un obstacle

de taille 2 × 2. Des reines ne peuvent pas être placées sur les carrés occupés par
cet obstacle, et ne peuvent pas attaquer d’autres reines en passant par l’obstacle.
Déterminer le nombre maximal de reines pouvant être placées sur cet échiquier
modifié, de façon à ce qu’aucune de ces reines en attaque une autre.

Copyright © Canadian Mathematical Society, 2022

122/ MathemAttic

MA163. Un billot A est de rayon R. Or un trou de rayon r est percé

perpendiculairement à l’axe vertical de A, tel qu’illustré ci-bas, et un deuxième
billot B, de rayon r, est inséré dans ce trou. Déterminer la longueur S en termes
de R et r.

MA164. Soit ABCD un parallélogramme. M et N sont des points sur les

côtés AB et AD respectivement, de façon à ce que AB = 4AM et AD = 3AN .
AC
Soit alors K le point d’intersection de M N et AC. Déterminer la valeur du AK .

MA165. Proposed by Dominique Mouchet.

ADE est un triangle équilatéral, OABC est un carré, avec O ∈ AD et C ∈
DE. On note r le rayon des cercles inscrits des triangles ODC et ABF (les deux
triangles sont isométriques), r0 celui du triangle CEF , et R le rayon du cercle
tangent aux segments OA, OC et AF . Montrer que Rr0 = r2 .

Crux Mathematicorum, Vol. 48(3), March 2022

 MathemAttic /123

MATHEMATTIC
SOLUTIONS
Statements of the problems in this section originally appear in 2021: 47(8), p. 372–373.

MA136. Proposed by Ed Barbeau.

Determine all sets consisting of an odd number 2m + 1 of consecutive positive
integers, for some integer m ≥ 1 such that the sum of the squares of the smallest
m + 1 integers is equal to the sum of the squares of the largest m integers.
We received 7 submissions of which 5 were correct and complete. We present the
solution by Prithwijit De.
Let Ω2m+1 = {a, a + 1, . . . , a + 2m} for some m ≥ 1 be the set with the desired
property. Let for k ≥ 0,

Sk+1 = a2 + (a + 1)2 + · · · + (a + k)2 .

Then according to the problem

Sm+1 = S2m+1 − Sm+1

whence
2Sm+1 = S2m+1 .
This leads to
m
X 2m
X
2 (a + k)2 = (a + k)2
k=0 k=0

which on simplifying yields

a2 − (2m2 )a − m2 (2m + 1) = 0

whose roots are a = −m, 2m2 + m. Clearly for a = −m, Ω2m+1 contains negative
integers. So a 6= −m and we must have a = 2m2 + m. Thus

Ω2m+1 = {2m2 + m + k|0 ≤ k ≤ 2m}

for each m ≥ 1.

MA137. Triangle ABC has area 1. X, Y are points on the side AB and Z
a point on the side AC such that XY = 2AX, XZ is parallel to Y C and Y Z is
parallel to BC. Determine the area of triangle XY Z.
Originally from the 2010 Alberta High School Mathematics Competition, Part I,
problem 15.

Copyright © Canadian Mathematical Society, 2022

124/ MathemAttic

There were 7 solutions submitted, of which 6 were correct. We presents the solution
by Doddy Kastanya.

Note that the triangles AYZ and ABC are similar.

Z
Y

C
B

Since
AY AZ AX 1
= = = ,
AB AC AY 3

then
[XY Z] = 32 [AY Z] = ( 32 )( 19 )[ABC] = 2
27 .

MA138. Proposed by Aravind Mahadevan.

Prove that csc 6◦ + csc 78◦ − csc 42◦ − csc 66◦ = 8.

Originally from Hong Kong 2021-008.

We received 8 solutions, of which 1 was complete and correct. Numerous other

solutions concluded with the right answer, but were deemed incomplete due to the
usage of trigonometric identities that were not explicitly stated. We present the
solution by Konstantine Zelator, summarized by the editor for brevity.

We do some initial simplification as follows:

csc 6◦ + csc 78◦ − csc 42◦ − csc 66◦ = (csc 6◦ − csc 66◦ ) − (csc 78◦ − csc 42◦ )
sin 66◦ − sin 6◦ sin 78◦ − sin 42◦
= ◦ ◦
− .
sin 6 sin 66 sin 42◦ sin 78◦

Crux Mathematicorum, Vol. 48(3), March 2022

 MathemAttic /125

Now, we will be using the following well-known identities:

A−B A+B
Å ã Å ã
sin A − sin B = 2 sin cos ,
2 2
sin(A + B) = sin A cos B + sin B cos A,
cos(A + B) = cos A cos B − sin A sin B,
cos 2A = 1 − 2 sin2 A = 2 cos2 A − 1, sin 2A = 2 sin A cos A,
sin(90◦ − A) = cos A, cos(90◦ − A) = sin A.
Continuing the computation, we find that
sin 66◦ − sin 6◦ sin 78◦ − sin 42◦ 2 sin 30◦ cos 36◦ 2 sin 18◦ cos 60◦
◦ ◦
− ◦ ◦
= −
sin 6 sin 66 sin 42 sin 78 sin 6 sin(60 + 6 ) sin(30◦ + 12◦ ) cos 12◦
◦ ◦ ◦
◦
cos 36 sin 18◦
= −
sin 6◦ (sin 6◦ cos 60◦ + sin 60◦ cos 6◦ ) (sin 30◦ cos 12◦ + sin 12◦ cos 30◦ ) cos 12◦
cos 36◦ sin 18◦
= 1 2 ◦ 1 ◦ ◦ ◦
− 1 ◦ 1 ◦ ◦ ◦
2 cos 12 + 2 cos 30 · 2 sin 12 cos 12
2
2 sin 6 + 2 sin 60 · 2 sin 6 cos 6
◦ ◦
cos 36 sin 18
= 1 1−cos 12◦
1 ◦ ◦
− 1 1+cos 24◦
1
2 2 + 2 sin 60 · sin 12 2 2 + 2 cos 30◦ · sin 24◦
◦
cos 36 sin 18◦
= 1 ◦ ◦ 1 ◦ + 1

− 1 1 ◦ ◦ ◦ + 1


2 sin 60 sin 12 − 2 cos 12 4 2 2 cos 24 + sin 24 cos 30 4
cos 36◦ sin 18◦
= 1 − 1
− 2 (cos 60◦ cos 12◦ − sin 60◦ sin 12◦ ) + 14 ◦ ◦ ◦
2 (sin 30 cos 24 + sin 24 cos 30 ) +
◦ 1
4
◦ ◦ ◦ ◦
cos 36 sin 18 cos 36 sin 18
= 1 ◦ 1 − 1 ◦ 1 = 1 ◦ 1 − 1 ◦ 1
− 2 cos 72 + 4 2 sin 54 + 4 − 2 sin 18 + 4 2 cos 36 + 4
√
−1+ 5
We finally substitute in the well-known values sin 18◦ = 4 and cos 36◦ =
√
1+ 5
4 , and simplify to get that the answer is 8.

MA139. The shape below was created by pasting together 25 unit squares.
When a similar shape is created with n squares, its perimeter is 100 units. Deter-
mine n.

Originally from the 2014-2015 Nova Scotia Math League, Game 1, Team question
2.
We received 6 solutions, of which 4 were correct. We present the solution by
Corneliu Manescu-Avram.
Let p be the perimeter of the shape containing n squares. We shall prove that
p = 2n + 2. We call a square interior if it does not contain sides on the perimeter.

Copyright © Canadian Mathematical Society, 2022

126/ MathemAttic

When the number of interior squares increases by 1, then n becomes n + 4 and p

becomes p + 8. As a result,

(p + 8) − 2(n + 4) = p − 2n,

whence we deduce by induction that the value of the expression p − 2n remains

unchanged no matter how many new interior squares are introduced. In the base
case there is only one interior square, so the initial value is

p − 2n = 12 − 2 · 5 = 2.

Therefore, it is always true that p = 2n + 2. Letting p = 100, we find that the

answer is
p−2 100 − 2
n= = = 49.
2 2

MA140.
If f (x) = 1 − x − x3 , what are all real values of x which satisfy
3
1 − f (x) − (f (x)) > f (1 − 5x)?

Originally from the Canadian National Mathematics League, Contest 6, April

1994, problem 6-6.
We received 8 solutions, of which 5 were correct. We present the solution by Digby
Smith.
Note that

f (b) − f (a) = (1 − b − b3 ) − (1 − a − a3 )
= (a − b) + (a3 − b3 )
= (a − b) + (a − b)(a2 + ab + b2 )
= (a − b)(a2 + ab + b2 + 1).

Since
b 2 3b2
Å ã
a2 + ab + b2 + 1 = a + + + 1 > 0,
2 4
we find that

a < b ⇐⇒ f (b) − f (a) = (a − b)(a2 + ab + b2 + 1) < 0

⇐⇒ f (a) > f (b).

In other words, f is a strictly decreasing function. Since we given that

f (f (x)) = 1 − f (x) − (f (x))3 > f (1 − 5x),

Crux Mathematicorum, Vol. 48(3), March 2022

 MathemAttic /127

we can use the strictly decreasing property to get

f (x) < 1 − 5x
1 − x − x3 < 1 − 5x
x(x − 2)(x + 2) = x3 − 4x > 0.

Performing interval analysis or sign analysis on this last inequality shows that the
solution is all real x such that −2 < x < 0 or x > 2.

42 and 1,000,000 – the problem

Show that for any sequence of 42 integers, one can insert plus signs, multipli-
cation signs, and parentheses between them so that the resulting expression is
divisible by one million.

For the solution, see page 131.

Sources: Moscow Mathematical Olympiad 2000, Problem B2, contributed

by S. A. Shestakov. See, e.g., R. Fedorov, A. Belov, A. Kovaldzhi,
and I. Yashchenko, eds., Moscow Mathematical Olympiads, 2000–2005,
MSRI/AMS 2011.

The interested reader should also see the book by Peter Winkler: Mathematical
Puzzles, CRC Press 2021 – which has hundreds more great puzzles!

Copyright © Canadian Mathematical Society, 2022

128/ Problem Solving Vignettes

PROBLEM SOLVING
VIGNETTES
No. 21
Shawn Godin
Pythagorean Triples
The Pythagorean theorem, a2 + b2 = c2 , where a and b are the lengths of the
legs of a right angled triangle and c is the length of its hypotenuse, has a history
spanning several millennia and a plethora of cultures. Over the years, hundreds
of proofs have been supplied. It is one of the first mathematical theorems that
children see in school and, in some cases, is one of the few mathematical results
people retain in adulthood. Its use as a metric in Euclidean space means that it
is present explicitly and implicitly in many topics of school mathematics.
A Pythagorean triple is a triple of positive integers, (a, b, c) with a, b < c that
satisfy the Pythagorean theorem (by convention, when a Pythagorean triple is
listed, its elements are generally listed in increasing order). That is, a, b, and c are
the side lengths of an integer sided right triangle. The smallest, and perhaps best
known, example is (3, 4, 5). A quick check with a spreadsheet for 1 ≤ a, b ≤ 25
reveals ten more examples:

(5, 12, 13), (6, 8, 10), (7, 24, 25), (8, 15, 17), (9, 12, 15),
(10, 24, 26), (12, 16, 20), (15, 20, 25), (18, 24, 30), (20, 21, 29).
From these examples we can see that (6, 8, 10), (9, 12, 15), (12, 16, 20), (15, 20, 25),
and (18, 24, 30) are all multiples of (3, 4, 5) and correspond to dilating the original
triangle by an integer factor. Similarly, (10, 24, 26) is a multiple of (5, 12, 13). In
general, if a, b, c are integers that satisfy a2 + b2 = c2 , then, for any integer k

(ka)2 + (kb)2 = k 2 a2 + k 2 b2 = k 2 (a2 + b2 ) = k 2 c2 = (kc)2 .

Hence, if (a, b, c) is a Pythagorean triple, then so is (ka, kb, kc) for any integer k.
It makes sense to define a primitive Pythagorean triple as one where the side
lengths are relatively prime, that is, gcd(a, b, c) = 1. Thus, from our examples so
far, (3, 4, 5), (5, 12, 13), (7, 24, 25), (8, 15, 17), and (20, 21, 29) are primitive.
A common starting place for searching for Pythagorean triples is to look at squares,
modulo 4, as in their remainders upon division by 4.
n n2 (mod 4)
0 0
1 1
2 0
3 1

Crux Mathematicorum, Vol. 48(3), March 2022

 Shawn Godin /129

Since the squares, modulo 4, are 0 and 1, then a2 + b2 can take on three possible
values, modulo 4. If a and b are both even, then a2 + b2 ≡ 0 (mod 4). In this case,
any Pythagorean triple would not be primitive since a, b, and, thus also, c would
be even. Similarly, if a and b are both odd, a2 + b2 ≡ 2 (mod 4). However, in
this case, the result is not a square modulo 4, which tells us that no Pythagorean
triples have the two smallest values being odd. That means, if we want a primitive
Pythagorean triple, one of a and b is even and the other is odd, which also makes
c odd. Hence no primitive Pythagorean triple has even c. A quick check of
the Pythagorean triples from our list reinforces the properties that we have just
discovered in action.
Looking more closely, a pattern appears in the first three primitive Pythagorean
triples: (3, 4, 5), (5, 12, 13), and (7, 24, 25). In all three examples, c = b + 1 and
b + c = a2 . If we let c = b + 1 and insert it into the Pythagorean theorem, we get
a2 + b2 = c2
a2 + b2 = (b + 1)2
a2 + b2 = b2 + 2b + 1
a2 = 2b + 1 = b + (b + 1) = b + c
showing that the second observation also holds. Further, since a2 = 2b+1 it means
that a must be odd and since b > 0, this forces a > 1 (otherwise we would have the
degenerate Pythagorean triple (a, 0, a), which isn’t very exciting). Thus, if we let
a = 2k + 1, for some non-negative integer k, then a2 = 2b + 1 yields b = 2k 2 + 2k
and c = 2k 2 + 2k + 1. Therefore
(2k + 1, 2k 2 + 2k, 2k 2 + 2k + 1), k ∈ Z+
is an infinite family of Pythagorean triples. However, since 2k 2 + 2k and 2k 2 +
2k + 1 are consecutive integers, they are necessarily relatively prime. Therefore
the Pythagorean triples shown above are primitive.
This might lead us to wonder, what if c = b + 2? Repeating the analysis yields
another infinite family
(2k, k 2 − 1, k 2 + 1), k ∈ Z+ , k > 1.
I will leave the details as an exercise for interested readers. In this case, if k is
odd, all the entries in the triple are even, so the triple is not primitive. If k is even,
then k 2 − 1 and k 2 + 1 are consecutive odd integers, and are therefore relatively
prime. Hence, if we only consider even k, the family described above would be
primitive.
We could continue in this matter and look at what happens if c = b + 3, c = b + 4,
. . . . Or, one might wonder, is there a way to describe all primitive Pythagorean
triples?
When I was in school and first encountered the algebraic identities
(x + y)2 = x2 + 2xy + y 2 and (x − y)2 = x2 − 2xy + y 2 ,

Copyright © Canadian Mathematical Society, 2022

130/ Problem Solving Vignettes

I noticed that (x + y)2 − (x − y)2 = 4xy. Then, if xy is a perfect square, it means

√
that (x − y, 2 xy, x + y) would be a Pythagorean triple. At the time, algebra
was new to me. Although I understood that this meant my method would always
create a Pythagorean triple, I didn’t quite believe it. I remember filling up several
pages of examples before I could accept the result.
Clearly, if x and y are both perfect squares then so is xy. For example, if x =
100 = 102 and y = 49 = 72 , then xy = 4900 = 702 . Is this the only way that xy
can be a perfect square? Consider x = 50 and y = 72. Neither of these numbers
are perfect squares, however, xy = 3600 = 602 . What is going on? If we look at
the largest factor of x and y that is a perfect square, we can write x = 2 × 52 and
y = 2 × 62 and
2
xy = 2 × 52 × 2 × 62 = 22 × 52 × 62 = (2 × 5 × 6) = 602 .

Hence, for xy to be a perfect square, we need x = km2 and y = kn2 for some
positive integers k, m, n. However, if k > 1, then the resulting Pythagorean triple
is not primitive. Hence
(m2 − n2 , 2mn, m2 + n2 ), m, n ∈ Z+ , m > n, (1)
is a Pythagorean triple. However, if d = gcd(m, n), then d2 is a factor of each term
of the triple. Therefore if we want the triple to be primitive we need gcd(m, n) = 1.
While this condition is necessary for the triple to be primitive, it is not sufficient.
For example, picking (m, n) = (5, 3) yields the Pythagorean triple (16, 30, 34)
which is not primitive. A quick check modulo 2 shows that if m and n are both
odd, the resulting Pythagorean triple will be made up of three even terms which
will therefore not be primitive. Thus, m and n will generate a primitive triple
only if m and n are relatively prime and have opposite parity (one is even and the
other is odd). I will leave verifying this fact to the interested reader.
Formula (1) is called Euclid’s formula as it appeared as proposition 29 in book X
of Euclid’s The Elements. All Pythagorean triples are of the form given in (1),
or a multiple of such a triple. For example (9, 12, 15) cannot be generated by (1),
however m = 2, n = 1 yields our famous (3, 4, 5) and (9, 12, 15) = (3×3, 3×4, 3×5).
A little algebra shows that (1) does indeed yield a Pythagorean triple. Hence, the
formula is sufficient to produce primitive Pythagorean triples, but is it necessary?
Assume that (a, b, c) is a primitive Pythagorean triple. Thus a, b, and c are
pairwise coprime (why?). We have shown already that a and b have opposite
parity, so assume, without loss of generality, that a is odd, making b even and c
odd.
Rearranging the Pythagorean theorem we get
a2 + b2 = c2
c2 − a2 = b2
c+a b
(c − a)(c + a) = b2 ⇒ =
b c−a

Crux Mathematicorum, Vol. 48(3), March 2022

 Shawn Godin /131

Since we are assuming, without loss of generality, that a < b < c, then a, b, and c
are distinct positive integers, c+a c+a m
b is rational and c − a > 0. Let b = n , where
m
gcd(m, n) = 1 (that is, n is in lowest terms). Then

b m c−a n
= ⇒ =
c−a n b m
which means that
c a m
+ = (2)
b b n
c a n
− = (3)
b b m
adding (2) and (3) yields
c m n
2 = +
b n m
c m2 + n2
= . (4)
b 2mn
Similarly, subtracting (3) from (2) yields
a m n
2 = −
b n m
a m2 − n2
= . (5)
b 2mn

If m and n are both odd, then m2 − n2 ≡ 0 (mod 4), from our discussion at the
start of the column, and thus

a m2 − n2 4k 2k
= = = , for some k ∈ Z+ .
b 2mn 2mn mn
However, since a and b are relatively prime ab is reduced to lowest terms which,
means that a must be even, a contradiction. Hence, m and n must have opposite
parity which means the expressions on the right sides of (4) and (5) are in lowest
terms. Since both sides of the equations in both (4) and (5) are already in lowest
terms, we get Euclid’s formula and we are done.
Here are some problems related to Pythagorean triples for your enjoyment.
1. If (a, b, c) is a primitive Pythagorean triple, show that exactly one of a and
b is a multiple of 3 and that c is never a multiple of 3.
2. If (a, b, c) is a primitive Pythagorean triple, show that 60 | abc.
3. If (a, b, c) is a primitive Pythagorean triple, show that the altitude from the
hypotenuse of the triangle with sides a, b, c does not have integer length.
4. Determine which positive integers are part of some primitive Pythagorean
triple.

Copyright © Canadian Mathematical Society, 2022

132/ Problem Solving Vignettes

5. Show that no primitive Pythagorean triple contains 2022. Determine all

non-primitive Pythagorean triples that contain 2022.
6. Find all primitive Pythagorean triples (a, b, c), such that the triangle with
sides a, b, c has inradius r = 2013. (Proposed by Neculai Stanciu, “George
Emil Palade” General School, Buzău, Romania and Titu Zvonaru, Comăneşti,
Romania.)

42 and 1,000,000 – the solution

Consider a sequence of 5 numbers. Some consecutive subsequence adds up to a

multiple of 5. (Proof: Look at a, a + b, a + b + c, a + b + c + d, a + b + c + d + e.
If they are all different modulo 5, then one is 0 (mod 5). If not, then two are
the same. Subtract them to get the desired subsequence.)
So break the 42 integers into groups of 7. For the first group, get a subsequence
of the first 5 with a sum divisible by 5, and multiply it by any unused numbers
in the first 5. It is still divisible by 5. If the 6th and 7th numbers have the same
parity, add them to get an even number and multiply by that to get a multiple
of 10 out of the first group. If they have opposite parity, multiply them to get
an even number, and do the same. So any group of 7 yields a multiple of 10,
and so the 42 numbers can be operated on to get a multiple of one million.
This solution was found by: Ben Katz, Richard Johnson, Maurizio Codogno,
Piotr Zielinski, Ross Millikan, Barry Cox, Jim Tilley, Nick Wedd, Peter
Saltzman, and Dan Dima.
Now, one can wonder if this can be done with 41 numbers. One might think
that if we had 41 ones, it would not be possible. But it is: Piotr Zielinski found
(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+1)∗(1+1)∗(1+1)∗(1+1)∗(1+1)∗(1+1)
= (1 + 2 ∗ 2 ∗ 2 ∗ 3 ∗ 3 ∗ (1 + 2 ∗ 2 ∗ 2 ∗ 3 ∗ 3 ∗ 3)) ∗ 2 ∗ 2 ∗ 2 ∗ 2 ∗ 2 ∗ 2
= (1 + 72 ∗ (1 + 216)) ∗ 64 = (1 + 72 ∗ 217) ∗ 64 = (1 + 15624) ∗ 64 = 106 .
Then one might think that with 41 elevens it would not be possible. But Piotr
found that then one needs only 17 numbers:
11+(11∗11∗11)∗(11+(11∗(11+(11∗11∗11))∗(11+(11∗11∗11∗(11+(11+11))))))
= 11+1331∗(11+(11∗(11+1331)∗(11+(1331∗(11+22)))))
= 11 + 1331 ∗ (11 + (11 ∗ 1342 ∗ (11 + (1331 ∗ 33))))
= 11+1331∗(11+(14762∗(11+43923))) = 11+1331∗(11+(14762∗43934))
= 11+1331∗(11+648553708) = 11+1331∗(11+648553708)
= 11+1331∗648553719 = 863225000000.
So now it is hard to say. Perhaps one can achieve a factor of one million with
any list of 41 numbers.

Crux Mathematicorum, Vol. 48(3), March 2022

 Asia Matthews /133

From the bookshelf of . . .

Asia Matthews
This new feature of MathemAttic brings attention to books of potential interest to
the readers. Some of these will be reviews whereas others will be hearty recom-
mendations from the contributors. If you have a book related to mathematics that
would be of interest to secondary school students and/or teachers, feel welcome to
send along a submission to [email protected]. Publishers are also wel-
come to send along books for possible review.

Things to Make and Do in the Fourth Dimension

by Matt Parker
ISBN 978-0374535636, 480 pages
Published by Doubleday Canada, 2014.

I picked this book up completely by chance at a bookstore in Boise, Idaho on one

leg of a trip with my family through the southwest. I laughed out loud in the
bookstore. I laughed out loud in the campground. I stole paper from the recycling
bins at the National Park office so that I could write and think about the new
ideas that I was finding in this little gem.
The writing is hilarious – just what you want from a
stand-up comedian and mathematician. “You might
want to laminate this and take it to parties” he says
about the full page of 78 graphs showing the differ-
ent ways that six people at a party can either know
each other or not know each other. This book is a
diamond in the rough world of mathematical commu-
nication. As a guild, we aren’t that great at selling
our craft, and so we are fortunate that there are people
like Matt Parker and Steven Strogatz who can explain
what we do in a way that is not frightening. As it
turns out, Parker was already on my radar since I’d
seen some of his Numberphile videos. In fact, I recom-
mend that you stop reading right now and go watch every video that he has made
https://standupmaths.com. What’s that? You were hoping for a comedy sketch
about spreadsheets? Yes.
I’m going to be honest: I haven’t read the whole book yet. But understand that
this is intentional: I’m saving some for later. But one thing I’m really happy to
have found early along is an optimized way to tie my shoes! You can find it at
the end of Chapter 8: Knot a Problem. My kids and I spent a good few hours
untangling factor graphs together at the pub, and I used the same idea a few years
ago when I ran a Professional Development day with local teachers. I also use
excerpts from this book in my mathematics for liberal arts course. Even students

Copyright © Canadian Mathematical Society, 2022

134/ From the bookshelf of . . .

with bad school math experiences in their past can get drawn in Parker is able
to push you that little bit, just enough to get momentum to hold your balance as
the ideas begin to take form. And the book is a great example of how one can use
different styles and rhetorical voice to express sophisticated ideas. In every section
Parker nudges the reader a little more forward toward that vital mathematical act
of generalization. For example, after talking about flat-packing and space filling
with polyhedral forms, he starts a new section of the chapter with “Moving on
. . . Get an orange”. What a lovely way to introduce the reader to the notion of
sphere-packing.
This book also contains a weird and delightful set of random mathematical history.
For example, Parker tells how Kepler uses an algorithm to find himself a wife; and
Chapter 3 begins, “In 1994 NASA calculated a strange numerical code and con-
cealed it in the backwaters of their website, where it remains to this day”, and then
describes how to find this secret. The book is a treasure trove of truly notewor-
thy old and contemporary mathematics. Parker cleverly weaves the new and old
together in a manner that reflects the interconnectivity of so many mathematical
topics. “Despite the fame of those limited-edition five Platonic solids, there are in
fact four more, which were completely unknown to the ancients. (Honestly, first
parallel lines and now this.)”
There are many open problems just tantalizingly hanging around in the book.
Parker explains each of them in a way that makes you feel that you should have
thought of this question, and how possibly can it not be known already. Including
open problems in a book is risky, because at any moment someone may find a
solution. No problem: Parker has made a website where he keeps things up to
date www.makeanddo4D.com – the author truly is devoted to getting the word out
that mathematics is fun, interesting, relevant, and alive.
If you do go out and buy this book, and I hope that you will, you can find
Parker’s favourite definition of maths (p. 40), his answer to why mathematicians
spend their time doing things for no practical reason (p. 51), and, on the last
two pages of Chapter 3, a really nice description of what it looks like when actual
mathematicians do math (together, and at the pub).

.................................................................

This book is a recommendation from the bookshelf of

Asia Matthews. Asia is a professor at Quest University
Canada and an associate editor of the College Math Jour-
nal. Asia is passionate about getting people to have fun
with math and she has been involved in enrichment pro-
grams for over ten years, including teaching at MathPath
for four years. She lives in Qathet, BC with her partner
and kids, and two catty cats.

Crux Mathematicorum, Vol. 48(3), March 2022

 Fok-Shuen Leung /135

From the lecture notes of . . .

Fok-Shuen Leung

Fok-Shuen Leung is Professor of Teaching and Undergraduate

Chair of the Mathematics Department at the University of
British Columbia. He is also an Academic Director of UBC
First Year Experience, and winner of the Killam Teaching
Prize and the PIMS Education Prize. He is particularly in-
terested in the transition from high school to university math,
and novice instructor training.

.................................................................

The first problem is from a first-year differential calculus course for students with-
out high school calculus. It requires some persistence but no “tricks”. A nice
feature of the problem is that students can confirm the correct answer using a
physical argument. In fact, a clever student might be able to proceed directly to
the physical argument and bypass the differentiation!
The second problem is from a first-year integral calculus course. I like this prob-
lem because it requires students to understanding something about the essence of
integration — “chop up and add”, as my colleague Lior Silberman puts it — and
not just follow a formula.

Problems
1. Imagine a disc of radius 1 m spinning anticlockwise at π rad/sec. A tight
elastic connector joins the point A on the rim of the disc to the stationary
point B located 1 m away from the disc, as shown below. Find the rate of
change of the angle θ at the moment α = π3 .

2. Let R denote the finite region enclosed by the curves y = x and y = x2 , and
let S denote the solid obtained by rotating R about the line y = x. Find the
volume of S.

Copyright © Canadian Mathematical Society, 2022

136/ From the lecture notes of . . .

Solutions
1. We can relate α and θ by dropping the perpendicular from point A.

sin(α)
It follows that tan(θ) = 2−cos(α) . Differentiating, we get

dθ cos(α)(2 − cos(α)) − sin2 (α) dα

sec2 (θ) = .
dt (2 − cos(α))2 dt

Plugging in α = π3 yields sec2 (θ) dθ

dt = 0; and since sec(θ) is plainly nonzero, we
conclude that dθ
dt = 0.
In fact, we can observe that when α = π3 , θ = π6 : the triangle with vertices at
A, B and the centre of the disc is a right triangle, and the elastic is connector is
tangent to the disc. In other words, when α = π3 , θ is at a local maximum, and
dθ
dt = 0.

2. Think of the line y = x as a diagonal axis. We’ll calculate the volume of S by

taking it to be the sum of the volume of thin disks perpendicular to this diagonal
axis.

In the picture above, at position z on the diagonal axis, we have a disk of thickness
dz and radius r. To write r√in terms of z, we note that the dotted radius√is a
segment of the line y = −x+ 2z. This line intersects y = x2 when x2 = −x+ √ √2z,
2
√ −1+ 1+4 2z
or 0 = x +x− 2z. By the quadratic formula, this has solution x = 2 .
(The other solution describes the point of intersection in the second quadrant.)

Crux Mathematicorum, Vol. 48(3), March 2022

 Fok-Shuen Leung /137

By Pythagoras,
p √ å2
√ √ √ » √
Ç
−1 + 1 + 4 2z z
»
2
r =2 −√ = 1+3 2z+z 2 − 1 + 4 2z− 2z 1 + 4 2z.
2 2

The first three terms are more straightforward to antidifferentiate, and we handle
those first. The volume of S is equal to
√ √ √
Z 2
2
Ä ä √
2 √ » Z √ ã2 Å»
πr dz = π z+ √3 z 2 + 1 3
−π
1 + 4 2z + 2z 1 + 4 2z dz
2 3z
0
0 0
√
Z √2 » √
√
√ Z 2 » √
14 2π
= 3 −π 1 + 4 2z dz − 2π z 1 + 4 2z dz
0 0
√
For
√ both integrals in the last line, we substitute u = 1 + 4 2z, getting du =
4 2z dz,
√ √
Z 2 » √ 1
Z 9
√ 1 9
13 2
1 + 4 2z dz = √ u du = √ u3/2 = .
0 4 2 1 6 2 1 6

and
√
Z 2 √ »
1
Z 9
√ 1 2 5/2 2 3/2
Å ã 9
149
1 + 4 2z dz =
z u(u − 1) du = u − u = .
0 32 1 32 5 3 1 60
Ä √ √ √ ä √
Thus the volume of S is equal to π 143 2 − 136 2 − 14960 2 = 602π
.

Copyright © Canadian Mathematical Society, 2022

138/ OLYMPIAD CORNER

OLYMPIAD CORNER
No. 401
The problems featured in this section have appeared in a regional or national mathematical
Olympiad.

Click here to submit solutions, comments and generalizations to any

problem in this section

To facilitate their consideration, solutions should be received by June 1, 2022.

OC571. The Tournament of Towns is held once per year. The 43rd such
tournament took place in 2021 and, amazingly, 43 is a divisor of 2021: 2021 ÷ 43 =
47. How many times more will humanity witness such a wonderful event?

OC572. The convex n-gon A1 A2 . . . An , with n > 4, is such that An−1 An A1

and An A1 A2 are isosceles triangles, as are Ai−1 Ai Ai+1 for 2 ≤ i ≤ n − 1. Prove
that there are at least two equal sides among any four sides of the n-gon.

OC573. There were 20 participants in a chess tournament. Each of them

played every other participant twice, once as white and once as black. We say
that participant X is no weaker than participant Y if X has won at least the same
number of games playing white as Y and also has won at least the same number
of games playing black as Y . Do there always exist two participants such that one
is no weaker than the other?

OC574. For any n ∈ N, consider the system (Sn ):

®
x2 + ny 2 = z2,
nx2 + y 2 = t2 ,

with unknowns x, y, z, t ∈ N. Let

M1 = {n ∈ N | System (Sn ) has infinitely many solutions}

and
M2 = {n ∈ N | System (Sn ) has no solutions}.

Prove that:
a) 7 ∈ M1 and 10 ∈ M2 ;
b) sets M1 and M2 are infinite.

Crux Mathematicorum, Vol. 48(3), March 2022

 OLYMPIAD CORNER /139

OC575. Let a1 , a2 , a3 , . . . , a9 be a random arrangement of the the numbers

1, 2, 3, . . . , 9. What is the greatest possible value of
√ √ √ √ √
|a1 − 3a2 | + |a2 − 3a3 | + |a3 − 3a4 | + · · · + |a8 − 3a9 | + |a9 − 3a1 |?

.................................................................

Les problèmes présentés dans cette section ont déjà été présentés dans le cadre d’une
olympiade mathématique régionale ou nationale.

Cliquez ici afin de soumettre vos solutions, commentaires ou

généralisations aux problèmes proposés dans cette section.

Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au plus tard le 1 juin 2022.

OC571. Le Tournoi des Villages se tient une fois par année. Or, le 43-ième
tel tournoi a eu lieu en 2021 et, remarquablement, 43 est un diviseur de 2021, car
2021 ÷ 43 = 47. Combien de fois verrons-nous une autre telle situation ?

OC572. Un polygone à n côtés, n > 4, est dénoté A1 A2 . . . An . Or,

An−1 An A1 et An A1 A2 sont des triangles isocèles, tout comme Ai−1 Ai Ai+1 pour
2 ≤ i ≤ n − 1. Démontrer que parmi n’importe quels quatre côtés du polygone, il
y en aura au moins deux de mêmes longueurs.

OC573. Un tournoi d’échecs a accueilli 20 participants. Chaque participant

a joué contre chaque autre participant deux fois, une fois avec les pièces blanches
et une fois avec les pièces noires. On dira que le participant X n’est pas plus faible
que le participant Y si, avec les pièces blanches, il a gagné au moins autant de
fois que Y , et si, avec les pièces noires, il a gagné au moins autant de fois que Y .
Est-il nécessairement le cas qu’il existe deux participants avec l’un pas plus faible
que l’autre ?

OC574. Pour n ∈ N, soit les système (Sn ):

®
x2 + ny 2 = z 2 ,
nx2 + y 2 = t2 ,

aux variables x, y, z, t ∈ N. Dénotons

M1 = {n ∈ N | le système (Sn ) possède un nombre infini de solutions}

Copyright © Canadian Mathematical Society, 2022

140/ OLYMPIAD CORNER

et
M2 = {n ∈ N | le système (Sn ) ne possède aucune solution}.
Démontrer que:
a) 7 ∈ M1 et 10 ∈ M2 ;
b) les ensembles M1 et M2 sont infinis.

OC575. Soit a1 , a2 , a3 , . . . , a9 un arrangement aléatoire des nombres

1, 2, 3, . . . , 9. Déterminer la plus grande valeur possible pour
√ √ √ √ √
|a1 − 3a2 | + |a2 − 3a3 | + |a3 − 3a4 | + · · · + |a8 − 3a9 | + |a9 − 3a1 |.

Crux Mathematicorum, Vol. 48(3), March 2022

 OLYMPIAD CORNER /141

OLYMPIAD CORNER
SOLUTIONS
Statements of the problems in this section originally appear in 2021: 47(8), p. 377–378.

OC546. Let a, b, c be three real numbers. For each positive integer n,

an + bn + cn is an integer. Prove that there exist three integers p, q, r such that
a, b, c are the roots of the equation x3 + px2 + qx + r = 0.
Originally from the 2009 Vietnam Mathematical Olympiad, Problem 4.
We received 9 submissions of which 8 were correct and complete. We present a
typical solution.
From the identity

6abc = (a + b + c)3 − 3(a + b + c)(a2 + b2 + c2 )

we see that 6abc is an integer, say 6abc = m. From the identity

6(abc)2 = (a2 + b2 + c2 )3 − 3(a2 + b2 + c2 )(a4 + b4 + c4 )

we get that 6(abc)2 is also an integer. But 6(abc)2 = m2 /6 from which it follows
that m is a multiple of 6. Thus abc is an integer.
From the identity

2(ab + bc + ca) = (a + b + c)2 − (a2 + b2 + c2 )

we get that 2(ab + bc + ca) is an integer, say 2(ab + bc + ca) = k. From the identity

2(ab + bc + ca)2 = (a2 + b2 + c2 )2 − (a4 + b4 + c4 ) + 4abc(a + b + c)

we get that 2(ab + bc + ca)2 is also an integer. But 2(ab + bc + ca)2 = k 2 /2 from
which it follows that k is a multiple of 2. Thus ab + bc + ca is an integer.
Let p = a + b + c, q = ab + bc + ca, and r = abc. By Vieta’s formula a, b, c
are the roots of the monic cubic polynomial x3 − px2 + qx − r. Since p, q, and r
are integers, it follows that a, b, c are the roots of a monic cubic polynomial with
integer coefficients, as required.

OC547. In a non-equilateral triangle ABC, I is the incentre and O is the

circumcentre. Prove that ∠AIO ≤ 90◦ if and only if 2BC ≤ AB + AC.
Originally from the 2002 French IMO selection test, Problem 5.
We received 8 correct solutions. We present the solution by Theo Koupelis.

Copyright © Canadian Mathematical Society, 2022

142/ OLYMPIAD CORNER

Let a, b, c be the side lengths of triangle ABC. Let R,r,s,E be the circumradius,
inradius, semiperimeter, and area of the triangle, respectively. From the law of
cosines in triangle AIO we have
OI 2 + AI 2 − OA2
cos ∠AIO = .
2 × OI × AI
If D is the point of tangency of the incircle and AB, we have

AI 2 = ID2 + AD2 = r2 + (s − a)2 .

Also, Euler’s theorem in geometry implies OI 2 = R(R − 2r). Therefore the follow-
ing are equivalent

∠AIO ≤ 90◦ iff cos ∠AIO ≥ 0 iff OI 2 + AI 2 ≥ OA2

iff R(R − 2r) + r2 + (s − a)2 ≥ R2
iff r2 + (s − a)2 ≥ 2Rr
E2 abc E
iff + (s − a)2 ≥ 2 × ×
s2 4E s
1 abc
iff (s − a)(s − b)(s − c) + (s − a)2 ≥
s 2s
iff 8(s − a)(s − b)(s − c) + 8s(s − a)2 ≥ 4abc
iff 8(s − a)[2s2 − s(a + b + c) + bc] ≥ 4abc
iff 2(s − a) ≥ a
iff b + c ≥ 2a.

OC548. Each of the digits 1, 3, 7, 9 occurs at least once in the decimal

representation of some positive integer. Prove that one can permute the digits of
this integer such that the resulting integer is divisible by 7.
Originally from the 2004 Pan-African Mathematical Olympiad, Problem 5.
We received 6 correct solutions. We present a typical solution.
Let N be the integer with 4 or more digits containing each of the digits 1, 3, 7, 9
at least once. Rearrange the digits to produce M = 10000 × a + b, where b has
4 digits, 1, 3, 7, 9 in some order, to be decided next. Compute the remainder
r ≡ 10000 × a (mod 7).
For r = 0 select b = 1379, r = 1 select b = 1973, r = 2 select b = 1937, r = 3
select b = 1397, r = 4 select b = 1739, r = 5 select b = 3719, and for r = 6 select
b = 1793.
Then M is divisible by 7, since

1379 + 0 = 7 × 197, 1973 + 1 = 7 × 282, 1937 + 2 = 7 × 277, 1397 + 3 = 7 × 200,

1739 + 4 = 7 × 249, 3719 + 5 = 7 × 532, 1793 + 6 = 7 × 257.

Crux Mathematicorum, Vol. 48(3), March 2022

 OLYMPIAD CORNER /143

OC549. In the triangle ABC we have AB = AC and ∠BAC = 90◦ . Consider

points M and P on AB such that AM = BP . Let D be the midpoint of BC, R a
point on CM , and Q a point on BC such that A, R, Q are collinear and the line
AQ is perpendicular to CM . Show that:
(a) ∠AQC ∼
= ∠P QB;
(b) ∠DRQ = 45◦ .
Originally from the 2004 Romanian Mathematical Olympiad (District Round),
Grade 7, Problem 4.
We received 9 correct solutions. We present two solutions.
Solution 1, by UCLan Cyprus Problem Solving Group.

(a) We will show that the triangles AQC and P QB are similar. Since ∠P BC =
∠ACQ = 45◦ it is enough to show that BQ PB AC
= CQ or equivalently that
BQ
CQ = PAC
B
= AM
AC = tan ϑ where ϑ = ∠M CA = ∠QAB. By Sine Law on
triangles ABQ and ACQ we have

BQ AQ CQ CQ CQ
= = = = .
sin ϑ sin 45◦ sin (∠CAQ) sin (90◦ − ϑ) cos ϑ
BQ
Thus CQ = tan ϑ as required.
(b) Since D is the midpoint of BC, then AD = BD = CD and since ∠DCA =
45◦ then ∠CAD = 45◦ and ∠ADC = 90◦ . So D lies on the circle with
diameter AC. So does R and therefore, ∠DRQ = ∠DCA = 45◦ .

Copyright © Canadian Mathematical Society, 2022

144/ OLYMPIAD CORNER

Solution 2, by Jaimin Patel.

Start with four identical right triangles.

Arrange them on the sides of a square. As midpoint of the diagonal, D is

The empty space in the center is a centre for the big and small squares.
square, as well. So, ∠DRQ = 45◦ . This solves part (b).

Remove two triangles, the one at the Fold the triangle from the left, as
top and the one from the right. shown. Part (a) follows.
Editor’s Comment. Solution 1 is rigorous, whereas Solution 2 is a visual solu-
tion. The graphic nature of the solution by Jaimin Patel appeals to the reader’s
geometrical intuition. The text under each picture supports this intuition.

Crux Mathematicorum, Vol. 48(3), March 2022

 OLYMPIAD CORNER /145

OC550. The real numbers x, y, and z are not all equal and satisfy:
1 1 1
x+ = y + = z + = k.
y z x
Determine all possible values of k.
Originally from the 2006 Nordic Mathematical Contest, Problem 2.
We received 15 submissions, of which 13 were correct and complete. We present
a typical solution.
If two of the numbers x, y, z are equal, then it is obvious that they are all equal,
1
which is not allowed. Let x, y, z be mutually distinct with xyz 6= 0. From z+ = k
x
1 kx − 1 1
and x + = k we get z = , and y = . Substituting into the final
y x k−x
equation we get
1 1 x
k=y+ = + .
z k − x kx − 1
Clearing denominators we get

(k 2 − 1)(x2 − kx + 1) = 0.
1
If x2 − kx + 1 = 0 then y = = x, which is not allowed.
k−x
Therefore, k 2 = 1, equivalently, k = 1 or k = −1. These two values are obtained
with the following solutions

1 x+1
Å ã
(x, y, z, k) = x, − ,− , −1 , with x(x + 1) 6= 0,
x+1 x

and
1 1−x
Å ã
(x, y, z, k) = x, ,− ,1 , with x(1 − x) 6= 0.
1−x x

Copyright © Canadian Mathematical Society, 2022

146/ Focus On... Linear algebra (I)

FOCUS ON...
No. 50
Michel Bataille
Linear algebra (I)
Introduction
Linear algebra occupies quite a large space in undergraduate syllabuses and regu-
larly appears in problem corners. So it seems of interest to present a selection of
problems centered on various objects pertaining to linear algebra (determinants,
matrices, ...). In this number and the next one, we have tried to choose examples
that lead to some valuable tricks or methods, often worthy of being kept in mind.
In this first part, we consider results about determinants, rank and trace. In part
II, we will focus on eigenvalues and polynomials of matrices.

Determinants
We begin with the following easy problem proposed in the October 2009 issue of
Mathematics Magazine:
Let A be an n×n matrix with integer entries and such that each column
of A is a permutation of the first column. Prove that if the entries in
the first column do not sum to 0, then this sum divides det(A).
In each column, the entries add up to the same sum s (because they are the
same entries as in the first column up to order). By adding rows 2, 3, . . . , n to
the first row,
the determinant remains unchanged while its first row becomes
s s · · · s . Thus, s can be factored and det(A) = s det(A0 ) where A0 is a
matrix with integer entries. The result follows since det(A0 ) is an integer.
Note that the calculation of det(A0 ) could continue
by subtracting column 1 from
columns 2, 3, . . . , n. This leads to 1 0 · · · 0 as the first row and to an easy
expansion. The method is exemplified by problem 1789 of Mathematics Magazine,
proposed in February 2008:
For nonzero real numbers a1 , a2 , . . . , an , define s = nk=1 a1k and
P

á ë
t + a1 t ... t
t t + a2 ... t
A= .. .. .. .. ,
. . . .
t t ... t + an

where t is a real number with st 6= −1. Find det(A).

n
t
For k = 1, 2, . . . , n, let bk = ak , so that st = bk . Then, factoring row k by ak ,
P
k=1

Crux Mathematicorum, Vol. 48(3), March 2022

 Michel Bataille /147

Å n
ã
we obtain det(A) = ak · det(B), where
Q
k=1
á ë
1 + b1 b1 b1 ... b1
b2 1 + b2 b2 ... b2
B= .. .. .. .. .
. . . ··· .
bn bn bn ... 1 + bn
Again, the useful observation is that in each column the entries add up to 1 + st.
Following the remark above, we obtain that
1
0 0 ... 0
1 0 ... 0
b2
det(B) = (1 + st) .
.. .. . .
..
. . · · · ..
bn 0 0 . . . 1
Å n ã
Thus, det(B) = 1 + st and det(A) = ak · (1 + st).
Q
k=1

Induction is often used in the calculation of a determinant. Here is an example:

If a1 , . . . , an , b1 , . . . , bn are complex numbers, prove that
á ë
0 a1 · · · an
b1
det .. = −(a1 b1 + · · · + an bn ) (1)
. In
bn

0 a1
Clearly, = −a1 b1 , hence the result holds for n = 1. Assuming that the
b1 1
result holds for some integer n ≥ 1 and expanding the determinant
0 a1 · · · an+1
b1
∆n+2 = ..
. In+1
bn+1
along the last row, we find
∆n+2 = −(a1 b1 +· · ·+an bn )+(−1)n+3 bn+1 (−1)n+2 an+1 = −(a1 b1 +· · ·+an bn +an+1 bn+1 ),
which completes the induction step.
We can resort to (1) for a variant of solution to 3928 [2014 : 124 ; 2015 : 133]:
Let A ∈ Mn (C) with rank(A) ≤ 1 and complex numbers x1 , x2 , . . . , xn
0 x1 · · · xn
Ü ê
n x1
with x2k = 1. If B = and In+1 is the unit
P
..
k=1 . A
xn

Copyright © Canadian Mathematical Society, 2022

148/ Focus On... Linear algebra (I)

matrix of size n + 1, prove that

Ö è
x1
det(In+1 + B) = x1 ···


xn A .. .
.
xn

Because rank(A) ≤ 1, we can find row vectors U = u1 · · · un and V =
v1 · · · vn such that A = U T V [this is obvious if A = 0; if rank(A) = 1, this


follows from the fact that the columns of A are of the form v1 U T , v2 U T , . . . , vn U T ].
n
Then, achieving a block-multiplication and using x2k = 1, it is readily checked
P
k=1
that á ëá ë
0 x1 · · · xn 0 v1 · · · vn
u1 x1
In+1 + B = .. .. .
. In . In
un xn
From (1) we deduce

det(In+1 + B) = (u1 x1 + · · · + un xn )(v1 x1 + · · · + vn xn ).

The result follows since we have

Ö è Ö è
x1 x1
x1 · · · xn A

.. = x1 ··· xn U

T


·V ..
. .
xn xn
= (u1 x1 + · · · + un xn )(v1 x1 + · · · + vn xn ).

The relation (1) also provides a quick solution to 3188 [2007 : 112 ; 2007 : 447]:
Let z1 , z2 , . . . , zn be the zeroes of the complex polynomial A(z) = z n +
an−1 z n−1 + · · · + a1 z + a0 , where a0 6= 0. Prove that
 
n z1 z2 ... zn
 z1 1 + z12 1 ... 1 
2
 
det 
 z2 1 1 + z2 . . . 1  = a21 .
. . . . .
. . . . .

. . . . . 
zn 1 1 ... 1 + zn2

As in the featured solution, we remark that the required determinant is det(A2 ) =

(det A)2 if we set
 
0 1 ... 1
1 
A = .
 
 ..

D  n
1

Crux Mathematicorum, Vol. 48(3), March 2022

 Michel Bataille /149

where Dn denotes the diagonal matrix diag(z1 , z2 , . . . , zn ). Now, from (1) we

obtain

0 1 ... 1
1/z1
det(A) = (z1 z2 . . . zn ) .
.. In
1/zn
1 1 1
Å ã
= −(z1 z2 . . . zn ) + + ··· +
z1 z2 zn
n
= (−1) a1 ,
n n
the latter because zk = (−1)n−1 a1 . The result follows.
P Q
j=1 k=1,k6=j

About idempotent matrices

We review some properties of idempotent matrices through a problem set in The
Mathematical Gazette in 2004:
Let A be an n × n matrix with complex entries. Show that A satisfies
A2 = A (i.e. is idempotent) if, and only if,

rk (A) ≤ tr (A) and rk (In − A) ≤ tr (In − A),

where rk () and tr () denote, respectively, the rank and the trace of a

matrix.
Suppose that A is idempotent. We first show that rk (A) = tr (A). If A is invert-
ible, then A = In and rk (A) = tr (A) = n. If A = On , rk (A) = tr (A) = 0. Now,
suppose that 0 < r = rk (A) < n. If X ∈ ker(A) ∩ im (A), then X = AY for some
Y , hence 0 = AX = A2 Y = AY = X. This shows that ker(A) ∩ im (A) = {0}.
Since in addition dim(ker(A)) + dim(im (A)) = n (the familiar rank formula), Cn
is the direct sum of ker(A) and im (A) and there exists a basis (X1 , . . . , Xn ) of Cn
such that (X1 , . . . , Xr ) is a basis of im (A) and (Xr+1 , . . . , Xn ) is a basis of ker(A).
Since AXi = Xi for i = 1, . . . , r and AXi = 0 for i = r + 1, . . . , n, A is similar to
the matrix with the unit r × r matrix in the north-west corner and 0s elsewhere.
Thus, we have tr (A) = r = rk (A).
Since In − A is idempotent as well, we similarly have rk (In − A) = tr (In − A).
Conversely, suppose that rk (A) ≤ tr (A) and rk (In − A) ≤ tr (In − A). Then

n = rk (In ) = rk (A+In −A) ≤ rk (A)+rk (In −A) ≤ tr (A)+tr (In −A) = tr (In ) = n

and so rk (A) + rk (In − A) = n.

The rank formula then yields dim ker(A) + dim ker(In − A) = n; since in addition
we have ker(A) ∩ ker(In − A) = {0}, we see that Cn = ker(A) ⊕ ker(In − A).

Copyright © Canadian Mathematical Society, 2022

150/ Focus On... Linear algebra (I)

Now, any column vector X can be written as X 0 + X 00 where X 0 ∈ ker(A) and

X 00 ∈ ker(In − A), hence, noticing that A(In − A) = (In − A)A, we obtain

A(In − A)X = (In − A)AX 0 + A(In − A)X 00 = (In − A)0 + A0 = 0.

Since X is arbitrary, it follows that A(In − A) = On and A is idempotent.

About rank 1 matrices

As we remarked in a previous section, a square matrix M with rank 1 can be
written as U T V for some nonzero row vectors U, V . Noticing that V U T is a
scalar, we deduce that

M 2 = (U T V )(U T V ) = U T (V U T )V = (V U T )U T V = (tr(M ))M.

This nice property allows an elementary solution to problem 4081 [2015 : 397 ; 405]:
Determine all A, B ∈ M2 (R) such that:
 Å ã
2 2 22 44
 A +B = ,


Å14 28ã
10 20
 AB + BA = .


2 4
Å ã Å ã
3 6 1 2
Let H = and K = . We show that the solutions for the pairs
2 4 0 0
(A, B) are (H, K), (K, H), (−H, −K), (−K, −H).
It is readily checked that these pairs are indeed solutions.
Conversely, let (A, B) be an arbitrary solution and let S = A + B and D = A − B.
Then Å ã
2 2 2 2 4
S = A + B + AB + BA = 16
1 2
and Å ã
1 2
D2 = A2 + B 2 − (AB + BA) = 12 (2).
1 2
We observe that S 2 and D2 have rank 1 so that S, D cannot be O2 nor be invertible,
hence S and D have rank 1 as well. It follows that S 2 = tr(S)S and D2 = tr(D)D
and therefore tr(S 2 ) = (tr(S))2 and tr(D2 ) = (tr(D))2 . From (2), we then deduce
that tr(S) = 8 or −8 and tr(D) = 6 or −6 and so

1 1
Å ã Å ã
2 4 1 2
S= S 2 = ±2 , D= D2 = ±2 .
tr(S) 1 2 tr(D) 1 2

Since A = 12 (S + D) and B = 12 (S − D), a short calculation shows that (A, B) is

one of the pairs (H, K), (K, H), (−H, −K), (−K, −H).
We conclude this section about rank 1 matrices with the following interesting
exercise:

Crux Mathematicorum, Vol. 48(3), March 2022

 Michel Bataille /151

Let M ∈ Mn (F) be a matrix with rank 1. Prove that det(In + M ) =

1 + tr(M ). [Here and in what follows, F denotes an arbitrary field.]
We obtain a basis of Fn by completing a basis (M X1 ) of im(M ) with appro-
priate column vectors X2 , . . . , Xn . Since M (M X1 ), M X2 , . . . , M Xn are collinear
to M X1 , the matrix of the endomorphism X 7→ M X in this basis is a matrix
M 0 whose entries are 0 except perhaps on the first row, which is of the form
(α1 , . . . , αn ). Thus In + M is similar to a triangular matrix with 1 + α1 , 1, 1, . . . , 1
as the diagonal entries. As a result, tr(M ) = tr(M 0 ) = α1 and det(In + M ) =
1 + α1 = 1 + tr(M ).

A simple, useful property of the trace

In addition to the usual properties of the the trace (linearity, tr(AB) = tr(BA)),
it is worth mentioning the following obvious one: for any matrix A in Mn (F),
tr(AAT ) is the sum of the squares of all entries of A. This result can sometimes
give rise to simple solutions. For example, it can be used to solve a problem posed
in the June 2007 issue of Mathematics Magazine (slightly modified):
Let P, Q ∈ Mn (R) be such that P = P 2 = P T , Q = Q2 = QT and
P QP = P . If P, Q have the same rank, prove that P = Q.
Equivalently, we show that tr((P − Q)(P − Q)T ) = 0. An easy calculation gives

tr((P − Q)(P − Q)T ) = tr(P ) + tr(Q) − tr(P Q) − tr(QP ).

We know that tr(P Q) = tr(QP ); moreover, since P, Q are idempotent with the
same rank, we have tr(P ) = rk (P ) = rk (Q) = tr(Q). As a result, tr((P − Q)(P −
Q)T ) = 2(tr(P ) − tr(P Q)).
To conclude, we observe that P Q is idempotent (since (P Q)2 = (P QP )Q = P Q)
and that
rk (P ) = rk (P QP ) ≤ rk (P Q) ≤ rk (P ).
Thus, rk (P ) = rk (P Q), that is, tr(P ) = tr(P Q) and tr((P − Q)(P − Q)T ) = 0
follows.

Exercises
1. Let A = (ai,j ) ∈ M4 (F) be such that ai,i = a2 , ai,5−i = b2 , ai,j = ab for
j 6= i, 5 − i (i, j = 1, 2, 3, 4). Calculate det(A).
2. Let A ∈ Mn (F) satisfy A2 = kA (k ∈ F) and let r = rk (A). Prove that
det(In + A) = (1 + k)r .
3. Let A ∈ Mn (R) satisfy AAT = A2 . Prove that AT = A.

Copyright © Canadian Mathematical Society, 2022

152/ Problems

PROBLEMS
Click here to submit problems proposals as well as solutions, comments
and generalizations to any problem in this section.

To facilitate their consideration, solutions should be received by June 1, 2022.

4721. Proposed by Greg Dresden and Myanna Nash.

It’s well-known that if we take the sum along a diagonal in Pascal’s triangle, like
this, Ç å Ç å Ç å Ç å
n n−1 n−2 n−3
+ + + + ...
0 1 2 3
then we get the Fibonacci numbers fn which satisfy fn = fn−1 + fn−2 .
But what if we take sums along diagonals at a more gentle slope? Let’s define the
numbers an as
Ç å Ç å Ç å Ç å
n n−1 n−2 n−3
an = + + + + ....
0 2 4 6

Prove that an = an−1 + an−2 + an−4 .

4722. Proposed by George Apostolopoulos.

Let I be the incenter of triangle ABC and let A0 , B 0 and C 0 be the intersections
of the rays AI, BI and CI with the circumcircle of the triangle. Prove that
Area(A0 B 0 C 0 )≥ Area(ABC).

4723. Proposed by Michel Bataille.

Let (un )n≥1 be the sequence defined by u1 = 1 and the recursion

un+1 = un + 22n−1 · u2n .

Express un as a function of n.

4724. Proposed by D.M. Bătineţu - Giurgiu and Daniel Sitaru.

Find all x, y > 0 such that:
1 1 1
8
+ 8
=
(x + 1) (y + 1) 8(xy + 1)4

Crux Mathematicorum, Vol. 48(3), March 2022

 Problems /153

4725. Proposed by Hidefumi Katsuura.

Z π√
3π π2
Prove that < sin xdx < .
4 0 4
(Hint: Consider the graph of a cycloid.)

4726. Proposed by Daniel Sitaru.

Find: Ã
n
(−1)k
Å ã
X n
lim n
· 2n−k ·
n→∞ k+1 k
k=0

4727. Proposed by Robert Frontczak.

For all n ≥ 0, prove that
n Ç å Å ãk
X 2n + 1 5 F6n+3
= 2n+1 ,
2k + 1 4 2
k=0

where Fn is the nth Fibonacci number

4728. Proposed by Antonio Garcia.

Let x, y, z ≥ 1. Show that
4xy 4yz 4xz
+ + ≥ 3.
(1 + y)2 (1 + z)2 (1 + x)2

4729. Proposed by George Stoica.

Let P ∈ R[x] of degree at least two, with simple zeroes and with at least one real
zero. Prove that P 0 /P has at least one pair of complex conjugate critical numbers.

4730. Proposed by Ion Patrascu and Ion Cotoi.

Let Γ be the center of the circle O of radius r, and let P be a point outside of Γ
with OP < 2r. The perpendicular bisector of OP intersects Γ at points Q and R,
and the segment P Q cuts Γ at C. The perpendicular bisector of P C intersects Γ
at points B and B 0 . The rays P B 0 , P B and P R intersect Γ at the points A, A0
and C 0 , respectively. Show that:
a) The triangles P AB and P A0 B 0 are equilateral.
b) The lines AA0 , BB 0 , CR, QC 0 and P O are concurrent.

.................................................................

Copyright © Canadian Mathematical Society, 2022

154/ Problems

Cliquez ici afin de proposer de nouveaux problèmes, de même que pour

offrir des solutions, commentaires ou généralisations aux problèmes
proposés dans cette section.

Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au plus tard le 1 juin 2022.

4721. Soumis par Greg Dresden et Myanna Nash.

Il est bien connu que si l’on considère la somme le long d’une diagonale du triangle
de Pascal comme suit :
Ç å Ç å Ç å Ç å
n n−1 n−2 n−3
+ + + + ...
0 1 2 3

alors on obtient les nombres de Fibonacci fn qui satisfont fn = fn−1 + fn−2 .

Qu’arrive-t-il si l’on considère plutôt la somme le long de diagonales suivant une
pente plus douce ? Définissons les nombres an comme suit :

Ç å Ç å Ç å Ç å
n n−1 n−2 n−3
an = + + + + ....
0 2 4 6

Montrez que an = an−1 + an−2 + an−4 .

4722. Soumis par George Apostolopoulos.

Soit I le centre du cercle inscrit au triangle ABC. Soit encore A0 , B 0 et C 0 les
intersections des demi-droites AI, BI et CI avec le cercle circonscrit à ce triangle.
Montrez que Aire(A0 B 0 C 0 )≥ Aire(ABC).

4723. Soumis par Michel Bataille.

Soit (un )n≥1 une suite définie par u1 = 1 et par la relation de récurrence un+1 =
un + 22n−1 · u2n . Exprimez un en fonction de n.

4724. Soumis par D.M. Bătineţu - Giurgiu et Daniel Sitaru.

Trouvez tous x, y > 0 pour lesquels :

1 1 1
8
+ 8
=
(x + 1) (y + 1) 8(xy + 1)4

Crux Mathematicorum, Vol. 48(3), March 2022

 Problems /155

4725. Soumis par Hidefumi Katsuura.

Z π√
3π π2
Montrez que < sin x dx < .
4 0 4
(Indice : Considérez le graphe d’une cyclode.)

4726. Soumis par Daniel Sitaru.

Trouvez Ã
n
(−1)k
Å ã
X n
lim n
· 2n−k ·
n→∞ k+1 k
k=0

4727. Soumis par Robert Frontczak.

Pour tout n ≥ 0, montrez que
n Ç å Å ãk
X 2n + 1 5 F6n+3
= 2n+1 ,
2k + 1 4 2
k=0

où Fn désigne le nième nombre de Fibonacci.

4728. Soumis par Antonio Garcia.

Étant donné x, y, z ≥ 1, montrez que
4xy 4yz 4xz
+ + ≥ 3.
(1 + y)2 (1 + z)2 (1 + x)2

4729. Soumis par George Stoica.

Soit P ∈ R[x] un polynôme de degré au moins deux ayant des zéros simples et
ayant au moins un zéro réel. Montrez que P 0 /P a au moins une paire de conjugués
complexes comme points critiques.

4730. Soumis par Ion Patrascu et Ion Cotoi.

Soit Γ un cercle de centre O et de rayon r. Soit P un point situé à l’extérieur du
cercle Γ et vérifiant OP < 2r. La médiatrice du segment OP coupe Γ aux points
Q et R, alors que le segment P Q rencontre Γ en C. La médiatrice du segment P C
coupe Γ aux points B et B 0 . Les demi-droites P B 0 , P B et P R rencontrent quant
à elles Γ respectivement en A, A0 et C 0 . Démontrez que :
a) Les triangles P AB et P A0 B 0 sont équilatéraux.
b) Les droites AA0 , BB 0 , CR, QC 0 et P O sont concurrentes.

Copyright © Canadian Mathematical Society, 2022

156/ Bonus Problems

BONUS PROBLEMS
These problems appear as a bonus. Their solutions will not be considered for publication.

B101. Proposed by Mihaela Berindeanu.

In M ABC with ]BAC = 90◦ , let S be the intersection point of symmedians. If
BC
AS = , then find ]BCA.
8

B102. Proposed by Nguyen Viet Hung.

Prove the following identities
4n
+ (−1)n 2n



2n 2 2n 2 2n 2 2n 2 2n n





(a) 0 + +
2 + ··· +
4 =
2n ,
2
4n n 2n



2n − (−1)

2 2n 2 2n 2 2n 2
(b) 2n n




1 + 3 + 5 + · · · + 2n−1 = .
2

B103. Proposed by Thanos Kalogerakis.

Let K and L be the midpoints of the respective sides AB and CD of a convex
quadrilateral ABCD, and let M and N be midpoints of the respective diagonals
AC and BD, while letting E and F be the points where the line M N meets the
lines AD and BC. For any point P on the line through L that is parallel to M N ,
complete the parallelogram P QRS so that

Q ∈ P L, E ∈ SP, R ∈ QF, K ∈ RS.

Prove that P QRS has the same area as the given quadrilateral ABCD.

B104. Proposed by Daniel Sitaru.

Find all real roots of the equation:

5x3 − 9x2 − 15x + 3 = 0.

Crux Mathematicorum, Vol. 48(3), March 2022

 Bonus Problems /157

B105. Proposed by Michel Bataille.

Let (Lm )m≥0 be the Lucas sequence defined by L0 = 2, L1 = 1 and the recursion
Lm+1 = Lm + Lm−1 and let n be a positive integer. Evaluate in closed form
n Ç å
X 2n
L4k .
n−k
k=0

B106. Proposed by Navid Safaei.

Let P (x) = a0 + a1 x + · · · + ad xd be a polynomial such that |aj | ≥ i6=j |ai |.
P
Prove that if a0 6= 0 and P (x) has a root r on the unit circle, then r is a root of
unity and either P (x) = Q(xk ) (for some k ≥ 2 and polynomial Q(x) with real
coefficients) or r = ±1.

B107. Proposed by Daniel Sitaru.

In ∆ABC, let a, b, c be side lengths, s denote the semiperimeter and r the inradius.
Show that the following relationship holds:

a b c 108r2
√ +√ +√ ≥ √ √ √ .
s+b s+c s+a a s+b+b s+c+c s+a

B108. Proposed by Goran Conar.

For n ∈ N and any a ∈ R, calculate
1 1 1 1 1
√
n
+ √
n
+ √
n
+ ··· + √
n
+ √
n
.
a+ a2 a+ a4 a+ a6 a + a2(n−2) a + a2(n−1)

B109. Proposed by George Stoica.

Prove that for any a ∈ (1, 2) and any integer n ≥ 1, there exist ε0 , ε1 , . . . , εn
∈ {−1, 1} such that |ε0 + ε1 a + . . . + εn an | < 1/(a − 1).

110. Proposed by Michel Bataille.

Let k, n be integers such that 0 ≤ k ≤ n. Evaluate
Z π
sin[(2n + 1 − k)x]
In,k = (cos x)k · dx.
0 sin x

B111. Proposed by Goran Conar.

Prove that x4 +y 3 +z 2 −3xyz = 0 has infinitely many solutions in positive integers.

Copyright © Canadian Mathematical Society, 2022

158/ Bonus Problems

B112. Proposed by Florin Stanescu.

Let A, B ∈ Mn (C) with n ≥ 2 be two matrices such that AB = BA and det(B) =
1. Show that | det(A + zB)| ≤ 1 for all z on the unit circle if and only if A is the
zero matrix.

B113. Proposed by George Stoica.

A ring R is a field if and only if, given a, b, c ∈ R with a 6= b, the equation
ax − xb = c has a unique solution x in R.

B114. Proposed by George Stoica.

If zi ∈ C satisfy | arg(zi )| ≤ t ≤ π/2, i = 1, 2, prove that |z1 + z2 |2 ≥ 4|z1 z2 | cos2 t.
When do we have equality?

B115. Proposed by Michel Bataille.

Evaluate !
n
1X
lim Un and lim Un − Uj
n→∞ n→∞ n j=1

where
n Ç å
1X n (−1)k−1
Un = .
n k+1 k
k=1

B116. Proposed by George Stoica.

Let ε ∈ (0, 1). For each n ∈ N, prove that the equation xn+1 = xn + ε · 2n has a
unique positive solution xn , which satisfies limn→∞ xn = 2.

B117. Proposed by Goran Conar.

Solve xy + y x = 2xy, where x and y are non-zero integers.

B118. Proposed by Florentin Visescu.

Prove that a given triangle ABC is equilateral if and only if
√ √ √ √
(3 3 cos A − sin A)(3 3 cos B − sin B)(3 3 cos C − sin C) = 3 3.

B119. Proposed by George Stoica.

Let f (x) ∈ Z[x] be a monic polynomial with f (0) 6= 0. If f (x) has only one root
(counting multiplicity) x0 ∈ C such that |x0 | ≥ 1, then f (x) is irreducible in Z.

Crux Mathematicorum, Vol. 48(3), March 2022

 Bonus Problems /159

B120. Proposed by Daniel Sitaru.

If x, y, z ∈ (0, 1) and xy + yz + zx = 1, show that
(1 + x2 )(1 + y 2 ) (1 + y 2 )(1 + z 2 ) (1 + z 2 )(1 + x2 ) 48
+ + ≥ 2
x2 y 2 y2 z2 z 2 x2 x + y2 + z2

B121. Proposed by Mihaela Berindeanu.

Let z ∈ C and |z| = 2 be. Show that 4 ≤ |z + 3| + |z 2 − z + 4| ≤ 11.

B122. Proposed by George Apostolopoulos.

Let ABC be an acute triangle with centroid G, inradius r and circumradius R. Let
AG, BG and CG intersect the sides BC, CA and AB at A1 , B1 and C1 respectively.
Prove that √
3 2r p p p 3R
√ ≤ A1 A2 + B 1 B 2 + C 1 C 2 ≤ √ .
R 2 r

B123. Proposed by Nguyen Viet Hung.

Find the greatest positive integer k such that the following inequality

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

B124. Proposed by Daniel Sitaru.

Let 4ABC be a triangle with internal bisectors wa , wb , wc and inradius r. Prove
that:
a4 b4 c4
+ + ≥ 144r3 .
wa wb wc

B125. Proposed by George Stoica.

Let f, g : R → (0, ∞) be continuously differentiable functions satisfying:
ã0
f (x) f (x) f (x)
Å
= −f (x) for all x ∈ R; lim = 1 and lim = 0.
g(x) x→−∞ g(x) x→∞ g(x)
Prove that
Z x Å Z x ã
g(t)dt < ∞ and f (x) = g(x) exp − g(t)dt for all x ∈ R.
−∞ −∞

Copyright © Canadian Mathematical Society, 2022

160/ Solutions

SOLUTIONS
No problem is ever permanently closed. The editor is always pleased to consider for
publication new solutions or new insights on past problems.

Statements of the problems in this section originally appear in 2021: 47(8), p. 392–397.

4671. Proposed by Minh Ha Nguyen.

Given a triangle ABC and a point M on the side BC, let r1 and r2 be inradii
of triangles ABM and ACM , respectively. Determine the position of M which
yields the maximum value of r1 + r2 .
We received 6 correct and 2 incomplete solutions for this problem. The correct
solutions can be put on a spectrum from purely trigonometric to non-trigonometric.
Solution 1, by Michel Bataille.
Let a = BC, b = CA, c = AB as usual and let x = BM, y = M C and m = AM .
Without loss of generality, we suppose that ∠ABC = B ≥ C = ∠ACB and under
this assumption we show
q that r1 + r2 is maximum when the angle θ = ∠AM B
tan(C/2)
satisfies θ = 2 arctan tan(B/2) . [Note that π − B > θ > C.]

We have a cot θ = x cot C − y cot B (see a proof at the end) and x + y = a. We

readily deduce that

a sin C sin(θ + B) sin(θ + B) sin(θ − C)

x= · =c· and y = b · .
sin A sin θ sin θ sin θ
c sin B b sin C
From the Law of Sines, we also have m = sin θ = sin θ . We can now calculate

sin(θ + B) sin B (1 + cos θ) sin B

Å ã Å ã
x+m−c=c + −1 =c + cos B − 1
sin θ sin θ sin θ
Å ã
θ
= c cos B − 1 + (sin B) cot
2

and similarly, y + m − b = b cos C − 1 + (sin C) tan θ2 .

tan θ2 and r2 = y+m−b

y+m−b
Since r1 = x+m−c
2 2 tan π−θ
2 = 2 cot θ2 , we obtain

θ θ
2(r1 + r2 ) = c(cos B − 1) tan + b(cos C − 1) cot + k.
2 2
where k is independent of θ.
b(cos C−1)
It follows that r1 + r2 is maximum when f (t) = c(cos B − 1)t + t is
maximum (tan C2 < t < tan π−B
2 ).

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /161

The derivative of f satisfies

c(cos B − 1) 2 b(cos C − 1) c(cos B − 1) 2 tan(C/2)
Å ã Å ã
f 0 (t) = t − = t − ,
t2 c(cos B − 1) t2 tan(B/2)
q q
tan(C/2) tan(C/2)
hence f 0 (t) = 0 iff t = tan(B/2) and f 0 (t) > 0 iff t < tan(B/2) . We deduce that
q
tan(C/2)
r1 + r2 is maximum when tan θ2 = tan(B/2) and the announced result follows.

Note It is readily seen that r1 = r2 when r1 + r2 is maximum.

−−→
Proof of a cot θ = x cot C − y cot B. Let → −u = a1 BC. Since aM = yB + xC, we
−−→ − −−→ − −→ −
have aAM · → u = y AB · → u + xAC · → u , hence aAM cos θ = bx cos C − cy cos B. But
we have c = AM sin θ
sin B and b = AM sin θ
sin C , hence

aAM cos θ = xAM sin θ cot C − yAM sin θ cot B.

Solution 2, by Prithwijit De.

I assume that in triangle ABC, ∠B > ∠C. Let M ∈ BC be such that ∠AM B = θ
and ∠BAM = α. Let AM = d. Then
[ABC] = (1/2)bc sin A = [ABM ] + [ACM ]
= (d/2)(BM sin θ + CM sin (π − θ))
= (ad/2) sin θ
bc sin A bc
whence d = = , where R is the circumradius of triangle ABC.
a sin θ 2R sin θ
Observe that
d bc/2R
r1 = =
α θ
Å ã
α θ
cot + cot sin θ cot + cot
2 2 2 2
and
d bc/2R
r2 = = ã.
A−α A−α
Å ã Å Å ã
θ θ
cot + tan sin θ cot + tan
2 2 2 2
Observe that α = π − B − θ. Therefore
B+θ A−α θ−C
Å ã Å ã Å ã
α
cot = tan and cot = cot .
2 2 2 2
Now
B+θ cos(B/2)
Å ã
θ
tan + cot =
B+θ
Å ã
2 2 θ
cos sin
2 2
and
θ−C cos(C/2)
Å ã
θ
cot + tan = .
θ−C
Å ã
2 2 θ
sin cos
2 2

Copyright © Canadian Mathematical Society, 2022

162/ Solutions

Therefore

B+θ
ã Å
cos
1 2 B θ
ã= = 1 − tan tan
B+θ
Å Å ã
θ B θ 2 2
sin θ tan + cot cos cos
2 2 2 2

and
θ−C
ã Å
sin
1 2 C θ
ã= = 1 − tan cot .
θ−C
Å Å ã
θ θ C 2 2
sin θ cot + tan sin cos
2 2 2 2

Therefore
Å ã
bc B θ C θ
r1 + r2 = 2 − tan tan − tan cot
2R 2 2 2 2

and it is maximised when

B θ C θ
f (θ) = tan tan + tan cot
2 2 2 2

is minimised. Observe that since all quantities involved in the definition of f are
positive, the A.M-G.M inequality gives us
…
B C
f (θ) ≥ 2 tan tan
2 2

θ tan(C/2)
with equality if, and only if tan= . Thus the maximum value of
2 tan(B/2)
θ tan(C/2)
r1 + r2 is attained when tan = which corresponds to that position
2 tan(B/2)
of M on the side BC which makes
Ç … å
bc B C
r1 = r2 = 1− tan tan .
2R 2 2

Solution 3, by Theo Koupelis.

Let a, b, c be the side lengths of triangle ABC, s1 , s2 be the semiperimeters of

triangles ABM, ACM, respectively, and h the height from A to BC. We also let
BM = x and AM = y. Then r1 s1 = 12 hx, and r2 s2 = 21 h(a − x). Also, using the

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /163

x
law of cosines in triangle ABM we get y 2 = c2 + x2 − · (a2 + c2 − b2 ). Therefore,
a
a−x 2x 2(a − x)
Å ã Å ã
h x h
r1 + r2 = + = +
2 s1 s2 2 y+c+x y+b+a−x
ah −2x2 + (2a + b − c)x + a(c + y)
= ·
a+b+c (b − c)x + ac + ay
2x(x − a)
ï ò
ah
= 1−
a+b+c (b − c)x + ac + ay
2
ß ™
ah
= 1+ 2 · [(b − c)x + ac − ay] .
a+b+c a − (b − c)2

Thus, the maximum value of r1 + r2 is achieved when the function

»
f (x) = (b − c)x + ac − ay = (b − c)x + ac − a2 x2 − a(a2 + c2 − b2 )x + a2 c2

achieves its maximum. We have

2a2 x − a(a2 + c2 − b2 )
f 0 (x) = 0 ⇐⇒ b − c = ay 0 = ,
2ay

from which, after squaring, we get

a2 a2 − (b − c)2 4y 2 − (b + c)2 + a2 = 0.
  

1 
Also, f 00 (x) = − (a + c)2 − b2 b2 − (a − c)2 < 0. Thus, the maximum
 
4ay 3
value of r1 + r2 is achieved when M ≡ Mo , where

1» √ A
AMo = yo = (b + c)2 − a2 = bc · cos ,
2 2
or when

1 h 2 » i a 2c + p(b + c)2 − a2
2 2
BMo = xo = a + c − b + (b − c) (b + c) − a = ·
2 2 p .
2a 2 b + c + (b + c)2 − a2

Obviously 0 < xo < a, and therefore there is a single solution for the position of
M for which r1 + r2 = max. It is easy to show that at maximum we have r1 = r2 .
Indeed,
a 2c + 2yo xo c + xo + yo xo s1
xo = · =⇒ = =⇒ = =⇒ r1 = r2 .
2 b + c + 2yo a − xo a + b + yo − xo a − xo s2

Finally, if E, r, s are the area, inradius, and semiperimeter of triangle ABC, respec-
tively, we have E = rs = r1 s1 +r2 s2 . Therefore, at maximum we have r1 = r2 = ro
and thus
rs (a + b + c)r
ro = = p .
s + yo a + b + c + (b + c)2 − a2

Copyright © Canadian Mathematical Society, 2022

164/ Solutions

Solution 4 by Marie-Nicole Gras.

A A

acute obtuse
B H M C H B M C

Let a = BC, b = CA, c = AB, be the side lenghts of 4ABC, and A = ∠BAC,
B = ∠CBA and C = ∠ACB be the measures of its angles; let H be the foot of
altitude from the vertex A.

Without loss of generality, we can suppose B ≥ C.

Point M is between B and C, and it is well known that

−AM + BM + c B
r1 = tan ,
2 2
−AM + CM + b C
r2 = tan ,
2 2
AM = BM + c − 2BM c cos B = (BM − c cos B)2 + c2 sin2 B.
2 2 2

p
We put x = BM − c cos B = HM . Then x ≥ 0, so that AM = x2 + c2 sin2 B,
BM = x + c cos B, CM = a − BM = c cos B + b cos C − BM , and then
p
− x2 + c2 sin2 B + x + c(1 + cos B) B
r1 = tan ,
p 2 2
2
− x + c sin B − x + b(1 + cos C)
2 2 C
r2 = tan ·
2 2

Let f (x) = 2(r1 + r2 ); then we have

Å ã Å ã
B C p 2 2 B C
f (x) = − tan + tan x + c sin B + tan − tan
2 x
2 2 2 2

B
C
+ c 1 + cos B tan + b 1 + cos C tan ·
2 2

We compute

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /165

Å ã Å ã
B C x B C
f 0 (x) = − tan + tan p + tan − tan ;
2 2 x2 + c2 sin2 B 2 2

and f 0 (x) = 0 if and only if

Å ã Å ã
B C B C p 2
tan + tan x = tan − tan x + c2 sin2 B.
2 2 2 2

Squaring the two members, we deduce

C 2 2 C 2 2
Å ã Å ã
B B
x + c2 sin2 B


tan + tan x = tan − tan
2 2 2 2
Å ã2
B C B C
⇐⇒ 4 tan tan x2 = tan − tan c2 sin2 B
2 2 2 2
sin B2 sin C2 2 sin2 B−C
2
 
⇐⇒ 4 B C
x = B C
bc sin B sin C since c sin B = b sin B
cos 2 cos 2 cos2 2 cos2 2
sin B2 sin C2 2 sin2 B−C
2 B B C C
⇐⇒ 4 x = 4 bc sin cos sin cos
cos B2 cos C2 cos2 B2 cos2 C
2
2 2 2 2
B−C
⇐⇒ x2 = bc sin2 ·
2

Since B ≥ C and x ≥ 0, it follows that f 0 (x) = 0 if and only if x = x0 =

√ B−C
bc sin ·
2
We deduce that f 0 (x) has same sign as −(x − x0 ) and f (x) is increasing if x < x0
and decreasing if x > x0 ; then we have: for all x ∈ R, f (x) ≤ f (x0 ).

It follows that f (x) = 2(r1 + r2 ) has maximum value when

√ B−C √ B−C
x = HM = bc sin , then BM = c cos B + bc sin ·
2 2

It remains to prove that M is between B and C. For that, we compute:

B−C
AM 2 = b2 sin2 C + bc sin2
2
B−C
= bc sin B sin C + bc sin2
2
B 2
ñ Å ã ô
B B C C B C C
= bc 4 sin cos sin cos + sin cos − sin cos
2 2 2 2 2 2 2 2
B+C A
= bc sin2 = bc cos2 ,
2 2

Copyright © Canadian Mathematical Society, 2022

166/ Solutions

√ A
and then AM = bc cos ·
2
√ A √
Since x ≥ 0, M belongs to the half-line HC. We notice AM = bc cos < bc,
2
and AM < b, since c ≤ b; then M is between H and C.

− If ∠ABC is acute, this prove that M is between B and C.

− If ∠ABC is obtuse, it remains to prove that AM > AB:

A
We deduce from Al-Kashi formula a2 = (b + c)2 − 4bc cos2 2 that 4AM 2 = (b +
c)2 − a2 .

Since ∠ABC is obtuse, we have b2 ≥ a2 + c2 and it follows, since also b ≥ c,

4AM 2 = b2 + 2bc + c2 − a2 ≥ 2bc + 2c2 ≥ 4c2 , QED.

Then, we have shown that the position of M , on the side BC, which yields the
maximum value of r1 + r2 , is given by
√ B−C
BM = c cos B + bc sin ·
2

Solution 5 by C. R. Pranesachar.
Let a = BC, b = CA, c = AB be the side lengths of triangle ABC; let s be
its semiperimeter, F its area, R its circumradius, r its inradius, ha its altitude
through A (see figure on page 4). We have
F 2R2 sin A sin B sin C a
r= = = sin B sin C
s s s
and
ha = c sin B = 2R sin B sin C.
Let ∠AM B = t. Then by the sine rule applied to triangles AM B and AM C, we
have
c sin(B + t) c sin B b sin C b sin(t − C)
BM = , AM = = , CM = .
sin t sin t sin t sin t
Then
1 B
r1 = (AB + BM − AM ) tan
2 2
1 c sin(B + t) c sin B
Å ã
B
= c+ − tan
2 sin t sin t 2
1 sin(B + t) sin B
Å ã
B
= c 1+ − tan
2 sin t sin t 2
1 B 1 B
= c(1 + cos B) tan + c(sin B cot t − sin B csc t) tan .
2 2 2 2

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /167

Also
1 C
r2 = (AC + CM − AM ) tan
2 2
1 b sin(t − C) b sin C
Å ã
C
= b+ − tan
2 sin t sin t 2
1 sin(t − C) sin C
Å ã
C
= b 1+ − tan
2 sin t sin t 2
1 C 1 C
= b(1 + cos C) tan + b(− sin C cot t − sin C csc t) tan .
2 2 2 2
Note that
1 B 1 C
c(1 + cos B) tan = b(1 + cos C) tan = R sin B sin C.
2 2 2 2
So Å ã
B C
r1 + r2 =2R sin B sin C + R sin B sin C tan − tan cot t
2 2
Å ã
B C
−R sin B sin C tan + tan csc t.
2 2
If we set
B C
tan = u, tan = v,
2 2
we see that
d
(r1 + r2 ) =R sin B sin C((u − v)(− csc2 t) + (u + v) csc t cot t)
dt
−(u − v) + (u + v) cos t
Å ã
=R sin B sin C .
sin2 t
d u−v
Thus dt (r1 + r2 ) = 0 implies cos t = u+v . Again

d2 −2 cos t (u + v)
Å Å ã ã
(r1 + r2 ) =R sin B sin C (−(u − v) + (u + v) cos t) − .
dt2 sin2 t sin t

Therefore when
u−v
cos t = ,
u+v
we have
d2 (u + v)
(r1 + r2 ) = −(R sin B sin C) < 0,
dt2 sin t
because sin t > 0 for 0 < t < π. So r1 + r2 is maximum when
u−v
cos t = .
u+v

Copyright © Canadian Mathematical Society, 2022

168/ Solutions

√
2 uv
Since sin t = , we have
u+v
(u − v) (u + v)
Å ã
max(r1 + r2 ) = 2R sin B sin C + R sin B sin C (u − v) · √ − (u + v) · √
2 uv 2 uv
√
= 2R sin B sin C + R sin B sin C · (−2 uv)
√
= 2R sin B sin C · (1 − uv)
Ç … å
s−a
= 2R sin B sin C · 1 −
s
Ç … å
s−a
= ha 1 − ,
s

because
…
B C (s − c)(s − a) (s − a)(s − b) s−a
tan tan = = .
2 2 ca ab s

Also note that

(1 − uv) 2r 2r
2R sin B sin C √ = √ = » .
(1 + uv) 1 + uv 1 + s−a
s

Thus Ç … å
s−a 2r
max(r1 + r2 ) = ha 1− = » .
s 1+ s−a
s

This completes the proof.

Remark. Walther Janous noted that this problem has been settled in a much
more general setting in [1], p. 351 - 352, entry 2.14.
[1] D. S. Mitrinović et al., Recent Advances in Geometric Inequalities. Kluwer
Acad. Publ. Dordrecht 1989.

4672. Proposed by Nguyen Viet Hung.

Let a, b, c be rational numbers such that
p p p
a2 − ab + b2 + b2 − bc + c2 = a2 + ac + c2 .
√
Prove that a2 + b2 + c2 is a rational number.
We received solutions from 15 authors, some proving in several different ways.
Two of them were chosen below because of being short and elegant.

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /169

Solution 1, by Michel Bataille.

Let ω = exp(2πi/3) and x, y ∈ R. Then

|x + ωy|2 = |x + ω 2 y|2 = (x + ωy)(x + ω 2 y) = x2 − xy + y 2 .

It follows that the hypothesis can be written |z1 | + |z2 | = |z1 + z2 | where

z1 = b + aω, z2 = −b − cω 2 .

Since z1 , z2 6= 0, we deduce that z1 = λz2 for some positive real number λ, that
is, b + aω = λ(c − b + cω). As a result, we have b = λ(c − b) and a = λc,
hence
√ bc = a(c − b), from which we obtain a2 + b2 + c2 = (c + a − b)2 . Thus,
a + b2 + c2 = |c + a − b|, a rational number.
2

Solution 2, by UCLan Cyprus Problem Solving Group.

Consider points O, A, B, C in the plane such that OA = a, OB = b, OC = c and
∠AOB = ∠BOC = 60◦ and ∠AOC = 120◦ .
By cosine rule we get
p p p
AB = a2 − ab + b2 , BC = b2 − bc + c2 , AC = a2 + ac + c2 .

Thus AB + BC = AC and the points A, B, C are collinear.

In the triangle OAC we have that OB is an angle bisector. So

(OA)(OC)
b2 = OB 2 = ((OA + OC)2 − BC 2 )
(OA + OC)2
ac
= ((a + c)2 − (a2 + ac + c2 ))
(a + c)2
ac 2
Å ã
= .
a+c

So,
ã2
(a2 + c2 )(a + c)2 + (ac)2 a4 + 2a3 c + 3a2 c2 + 2ac3 + c4 a2 + ac + c2
Å
a2 +b2 +c2 = 2
= = .
(a + c) (a + c)2 a+c
√
Thus a2 + b2 + c2 is a rational number.

4673. Proposed by Todor Zaharinov.

Let P be a point in the plane of triangle ABC not on any line joining two of its
vertices. Let P A1 A2 be the triangle with centroid A and vertices A1 ∈ BA and
A2 ∈ AC. By cyclically permuting A, B, C we similarly define triangle ∆P B1 B2
with centroid B, and P C1 C2 with centroid C. Prove that the triangles A1 B1 C1
and A2 B2 C2 have equal areas.

Copyright © Canadian Mathematical Society, 2022

170/ Solutions

We received ten solutions, all correct, and will sample two of them.
Solution 1, a composite of the almost identical solutions from Michel Bataille and
from the UCLan Cyprus Problem Solving Group.
We use barycentric coordinates. Let A = (1, 0, 0), B = (0, 1, 0) and C = (0, 0, 1).
Assume P = (x, y, z) with x + y + z = 1.
Since A1 belongs to the line AB, we have A1 = (r, 1 − r, 0) for some r. Similarly
A2 = (s, 0, 1 − s) for some s.
The centroid of P A1 A2 has coordinates 31 (x + r + s, y + 1 − r, z + 1 − s), from
which we deduce that r = 1 + y and s = 1 + z.
Thus A1 = (1 + y, −y, 0), A2 = (1 + z, 0, −z). Cyclically we get

B1 = (0, 1 + z, −z), C1 = (−x, 0, 1 + x), B2 = (−x, 1 + x, 0), C2 = (0, −y, 1 + y).

For the signed areas of the triangles A1 A2 A3 and B1 B2 B3 we have

1+y −y 0
[A1 B1 C1 ]
= 0 1+z −z
[ABC]
−x 0 1+x
=(1 + x)(1 + y)(1 + z) − xyz
1+z 0 −z
[A2 B2 C2 ]
= −x 1+x 0 = .
[ABC]
0 −y 1+y

So the triangles A1 B1 C1 and A2 B2 C2 have equal area.

Solution 2, by J. Chris Fisher.

The problem is a special case of the following theorem:
Given a triangle ABC, let B2 , C2 , and A2 be arbitrary points on the lines AB, BC,
and CA, respectively, and define
C1 to be the point where the line parallel to BC through B2 intersects CA,
A1 to be the point where the line parallel to CA through C2 intersects AB,
B1 to be the point where the line parallel to AB through A2 intersects BC.

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /171

Then the triangles A1 B1 C1 and A2 B2 C2 have equal areas.

Proof of the theorem. Because the result belongs to affine geometry, we may take
(without loss of generality) the origin at B, and let C = (1, 0) and A = (0, 1)
be the unit points of the x- and y-axes. As in the accompanying figure we take
B2 = (0, b) to be the point chosen on AB, so that C1 = (1 − b, b) is the point
where y = b intersects AC (which is the line y = 1 − x). Similarly, the choice of
C2 = (c, 0) gives us A1 = (0, c), and A2 = (a, 1 − a) gives B1 = (a, 0). The areas
of triangles A1 B1 C1 and A2 B2 C2 are therefore, respectively
0 c 1 a 1−a 1
a 0 1 and 0 b 1 ,
1−b b 1 c 0 1
which both equal ab − bc − ca + c.
It remains to see that the theorem applies to the configuration of problem 4673.
In ∆P B1 B2 note that because B1 is on the line BC, while B is assumed to be
the centroid, the median B1 B must lie along BC and, therefore, BC contains the
midpoint of side P B2 . Similarly, we see that BC passes through the midpoint of
the side P C1 of ∆P C1 C2 . Consequently, the base B2 C1 of ∆P B2 C2 is parallel to
the line BC, which passes through the midpoints of the other two sides. In the
same way, CA||C2 A1 and AB||A2 B1 , as desired.
Editor’s comments. There is an interesting alternative statement for the theorem
of solution 2: Given a set of six points in the plane, no three collinear, form the
hexagon A0 A1 A2 A3 A4 A5 . If the vertices are arranged so that Ai Ai+1 ||Ai−1 Ai+2
for i = 1, 3, 5 (where the indices are taken modulo 6), then the triangles A0 A2 A4
and A1 A3 A5 have the same area.

Copyright © Canadian Mathematical Society, 2022

172/ Solutions

4674. Proposed by Michel Bataille.

Find all monic polynomials P (x) ∈ Z[x], of positive degree 2n, that have a complex
root of multiplicity n and satisfy P (0) = −1 and P (1) · P (−1) 6= 0.
We received 4 submissions, all of which were correct. We present the solution by
Oliver Geupel.
It is straightforward to verify that, if n is an odd number, then, for every integer
a 6= 0, the polynomial
P (x) = (x2 + ax − 1)n
satisfies the given conditions. We are going to prove that there are no further
solutions. In particular, there is no solution when n is even.
Suppose that P satisfies the requirements and that ξ ∈ C is a root of P of mul-
tiplicity n. Since P is monic, ξ is an algebraic integer. Let d be the degree of
the minimal polynomial of ξ. Because ξ is distinct from its conjugates, we have
nd ≤ 2n, that is, 1 ≤ d ≤ 2. If d = 1, then the minimal polynomial x − ξ of
ξ is a divisor of P (x) in the ring Z[x]. Then, ξ is a rational integer which is a
divisor of P (0) = −1. It follows that ξ ∈ {−1, 1}, a contradiction to the hy-
pothesis that P (1)P (−1) 6= 0. As a consequence, d = 1 is impossible and we
obtain that d = 2 and that P has the form P (x) = (x2 + ax + b)n with integers
a and b. Hence, bn = P (0) = −1, that is, n is odd and b = −1. Moreover,
−a2n = an (−an ) = P (1)P (−1) 6= 0 implies that a 6= 0. Hence the result.

4675. Proposed by George Apostolopoulos.

Let a, b and c be positive real numbers such that a2 + b2 + c2 = 3. Prove that

2(a4 + b4 + c4 ) − (a3 + b3 + c3 ) ≥ 3abc.

We received 26 submissions of which 23 were correct and complete.

We present a solution submitted independently by S̆efket Arslanagić, Theo Koupelis,
and Cao Minh Quang.
Using AM-GM we have a3 + b3 + c3 ≥ 3abc, and therefore it is sufficient to show
2
the stronger inequality a4 + b4 + c4 ≥ a3 + b3 + c3 . However, x4 − x3 ≥ x 2−1 holds,
since it is equivalent to 12 (x − 1)2 (2x2 + 2x + 1) ≥ 0. Thus, by adding the three
inequalities obtained when x is replaced by a, b, and c we have

a2 + b2 + c2 3
a4 + b4 + c4 − (a3 + b3 + c3 ) ≥ − = 0.
2 2
Equality occurs when a = b = c = 1.

4676. Proposed by Lorian Saceanu.

Let ABCD be a convex quadrilateral with E at the intersection of diagonals.
From E build the bisectors of the angles ∠AEB, ∠BEC, ∠CED, ∠DEA. We

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /173

get an orthodiagonal quadrilateral F GHI with F ∈ [AB], G ∈ [BC], H ∈ [CD],

I ∈ [DA]. Prove that:

[AF ] [BG] [CH] [DI]

· · · = 1.
[BF ] [CG] [DH] [AI]

We received 12 submissions, all quite similar, and present the common solution.
We extend the result by allowing E to be any point in the interior of the quadri-
lateral. Recall that in a triangle U V W , if the bisector of ∠V U W intersects V W
VX VU
at X, then W X = W U . Applying this theorem to the four triangles with vertex
E (and one side of the quadrilateral as its opposite side), we deduce that
AF AE BG BE CH CE DI DE
= , = , = , and = .
BF BE CG CE DH DE AI AE
Consequently,
AF BG CH DI AE BE CE DE
· · · = · · · = 1.
BF CG DH AI BE CE DE AE

Editor’s comments. Observe that we never made use of the convexity assumption:
the same conclusion follows should E be any point in the plane of an arbitrary
quadrilateral ABCD as long as the points where the four angle bisectors meet
the sides of the quadrilateral are well defined. Indeed, the analogous result would
apply to any polygon in the plane. The case of a triangle is especially amusing: For
a point P in the plane of triangle ABC, define the points D, E, and F to be the
intersections of the sides BC, CA, and AB with the bisectors of ∠BP C, ∠CP A,
and ∠AP B, respectively (where P is restricted to those points for which all three
points of intersection are well defined). Then the cevians AD, BE, and CF are
concurrent.

4677. Proposed by Seán M. Stewart.

Evaluate ∞
2x
Z Å ã
x
arctan dx.
0 1 + x2 x2 +4

There were 10 correct solutions and one incomplete solution.

Let I be the value of the integral. The answer is

Ç√
√ √
å
π 2+3 π π
I = ln √ = ln((2 2) − 1) = ln(9 − 4 2).
2 2+1 2 4

We will make use of the fact that

2x
arctan = arctan ux − arctan xv = arctan ux − arctan vx,
x2 + 1

Copyright © Canadian Mathematical Society, 2022

174/ Solutions

√ √
where u = 2 + 1 and v = 2 − 1 = u−1 .

Solution 1, by Niyazi Suleymanov.

Z ∞ Z ∞ ÅZ u
x(arctan u/x − arctan v/x) x2
ã
I= dx = dy dx
0 x2 + 4 0
2 2 2
v (y + x )(x + 4)
Z u ÅZ ∞
x2
ã
= dx dy
v 0 (y 2 + x2 )(x2 + 4)
Z uÅ ∞ Z ∞
y2 4
Z ã
dx dx
= − 2
v y 2 − 4 0 x2 + y 2 y − 4 0 x2 + 4
Z uï ò∞
y arctan(x/y) − 2 arctan(x/2)
= dy
v y2 − 4 0
Z u Å √
Z 2+1 Ç √ å
π y−2 2+3
ã
π dy π
= dy = = ln √ .
v 2 y2 − 4 2 √2−1 y + 2 2 2+1

Solution 2, by Brian Bradie.

Let ∞
x(arctan tx − arctan t−1 x)
Z
g(t) = dx.
0 x2 + 4
for t > 0. Observe that g(1) = 0 and that I = g(u). Then
Z ∞ Z ∞
0 x2 dx x2 dx
g (t) = + .
0 (1 + t2 x2 )(x2 + 4) 0 (t2 + x2 )(x2 + 4)
1
When t = 2 and t = 2 respectively, the first and second integrals are 4 and 1 times
∞ π/2 π/2
x2 dx (4 tan2 θ) d(2 tan θ) 1
Z Z Z
π
= = sin2 θ dθ = .
0 (x2 + 4)2 0 16 sec4 θ 2 0 8

For t 6= 21 ,
Z ∞ Z ∞Å
x2 dx 1 1 4
ã
= − dx
0 (1 + t2 x2 )(x2 + 4) 1 − 4t2 0 1 + t2 x2 x2 + 4
1 1 x ∞
ï ò
π
= arctan tx − 2 arctan = .
1 − 4t2 t 2 0 2t(1 + 2t)
This agrees with the answer π/2 when t = 1/2.
For t 6= 2,
Z ∞ Z ∞Å
x2 dx 1 4 t2
ã
= − dx
0 (x2 + t2 )(x2 + 4) 4 − t2 0 x2 + 4 x2 + t2
1 h x x i∞ π
= 2 arctan − t arctan = .
4 − t2 2 t 0 2(2 + t)

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /175

This agrees with the answer π/8 when t = 2.

Hence
1 1 2
Å ã
0 π
g (t) = + − .
2 2+t t 1 + 2t
Therefore ∞
t(2 + t)
Z Å ã
0 π
g(t) = g (t) dt = ln .
1 2 1 + 2t
Letting t = u, we find that
√ √ Ç√ å
π ( 2 + 1)( 2 + 3) π 2+3
I = · ln √ = ln √ .
2 2 2+3 2 2+1

Solution 3, by UC Lan Cyprus Problem Solving Group.

Integration by parts yields

ò∞ Z ∞ 2
2x (x − 1) ln(x2 + 4)
ï Å ã
1 2
I = 2 arctan ln(x + 4) + dx.
1 + x2 0 0 x4 + 6x2 + 1
Since, for x ≥ 0,
2x 2x ln(x + 2)2 4 ln(x + 2)
Å ã
arctan ln(x2 + 4) ≤ < ,
1 + x2 1 + x2 x
the integrated term vanishes and we find that
Z ∞ 2
(x − 1) ln(x2 + 4)
I= dx
0 (x2 + u2 )(x2 + v 2 )
u ∞ ln(x2 + 4) v ∞ ln(x2 + 4)
Z Z
= dx − dx,
2 0 x2 + u2 2 0 x2 + v 2
√ √ √ √
where u = 2 + 1, v = 2 − 1, u2 = 3 + 2 2 and v 2 = 3 − 2 2.
For w > 0 and t ≥ 0, define
∞
ln(x2 + t2 )
Z
f (t) = dx.
0 x2 + w 2
Then
∞ Z ∞
ln x ln wy
Z
f (0) = 2 2 + w2
dx = 2 2 + w2
wdy
0 x 0 (wy)
ïZ ∞ Z ∞
2 ln w ln y
ò
= dy + dy .
w 0 y2 + 1 0 y2 + 1
Observe that
Z 1 ∞ ∞
ln y ln(1/z) ln z
Z Å ã Z
dz
2
dy = − =− dz,
0 y +1 1 (1/z)2 + 1 −z 2 1 z2 +1

Copyright © Canadian Mathematical Society, 2022

176/ Solutions

so that
π ln w π ln w
f (0) = +0= .
w w
Now
∞ Z ∞
2t 2t 1 1
Z ï ò
f 0 (t) = dx = − dx
0 (x2 + t2 )(x2 + w2 ) 0 w2 − t2 x2 + t2 x2 + w 2
2t 1 1 x ∞ 2
ï ò ï ò
x π πt π
= 2 arctan − arctan = − = .
w − t2 t t w w 0 w 2 − t2 2 2w w(w + t)

Hence
π ln(w + t)
f (t) = .
w
Applying this result when (t, w) = (2, u), (2, v) yields
√ √ Ç√ å
π ln( 2 + 3) π ln( 2 + 1) π 2+3
I= − = ln √ .
2 2 2 2+1

Comments by the editor. Theo Koupelis followed the approach of Solution 3,

but used the calculus of residues to compute f (t) using the contour consisting of
[−R, R] and the upper semicircle of radius R and centre 0 as R → ∞. Three
solvers generalized the result, replacing 2x by another multiple of x and 4 by
another constant. The most interesting of these was due to C.R. Pranesachar who
showed that
Z ∞
2x tan θ 3 + tan θ/2
Å ã Å Å ãã
x π
arctan · dx = ln(sec θ + tan θ) − ln .
0 1 + x2 x2 + 4 2 3 − tan θ/2

The derivative of this integral with respect to θ is

Z ∞
2x2 (1 + x2 ) dx
sec2 θ · .
0 (x2 + 4)(x2 + (sec t + tan t)2 )(x2 + (sec t − tan t)2 )

This can be evaluated decomposing the integrand into partial fractions and then
integrating to evaluate the integral. Our result corresponds to θ = π/4.
Didier Pinchon followed the strategy of Solution 1 and evaluated
Z ∞
arctan(u/x)[x(x2 + 4)−1 ] dx
0

by parts. He pointed out that the generalization

Z ∞ Ç √ å
2rx a + r2 + b2 + r
Å ã Å ã
x dx π
arctan 2 · dx = ln √
0 b + r2 x2 + a2 2 a + r2 + b2 − r

was published by Seán Stewart in Scientia, Series A: Mathematical Sciences 31

(2021), 25-56.

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /177

4678. Proposed by Michel Bataille.

Let n be a nonnegative integer. Evaluate in closed form
bn/2c Ç å Å ãj
X n+1 1
j=0
n − 2j 5
bn/2c Ç ã .
1 j
åÅ
X n−j
−
j=0
n − 2j 5

We received 10 solutions, all of which were correct. We present the solution by

UCLan Cyprus Problem Solving Group.
In what follows all sums are over all values of the index for which the binomial
coefficient makes sense and it is non-zero. We have
X n + 1 Å 1 ãj X n + 1 Å 1 ãj √ X n + 1 Å 1 ã2j+1
Ç å Ç å Ç å
= = 5 √ .
j
n − 2j 5 j
2j + 1 5 j
2j + 1 5

So by the Binomial Theorem

√ ÇÅ
X n + 1 Å 1 ãj 1 n+1 1 n+1
Ç å å
5
ã Å ã
= 1+ √ − 1− √
j
n − 2j 5 2 5 5
√ √
( 5 + 1)n+1 − ( 5 − 1)n+1
= .
2 · 5n/2
Letting Ç åÅ ãj X n − j Å 1 ãj
Ç å
X n−j 1
A(n) = − = −
j
n − 2j 5 j
j 5
we have
X n + 2 − j Å 1 ãj
Ç å
A(n + 2) = −
j
j 5

1 j X n+1−j 1 j
Ç åÅ Ç åÅ
X n+1−j ã ã
= − + −
j
j 5 j
j−1 5

1 j−1
Ç åÅ
1X n+1−j
ã
= A(n + 1) − −
5 j j−1 5
1
= A(n + 1) − A(n) .
5
So the sequence A(n) satisfies a homogeneous secondÄ orderä recurrence relation
with characteristic equation x − x + 5 and roots 2 1 ± √15 . So
2 1 1

Ç√ ån Ç√ ån
5+1 5−1
A(n) = A √ +B √
2 5 2 5

Copyright © Canadian Mathematical Society, 2022

178/ Solutions

√ √
1+ 5 1− 5
for some constants A, B. Since A(0) = A(1) = 1 we get A = 2 and B = 2 .
Thus
√ √
( 5 + 1)n+1 − ( 5 − 1)n+1
A(n) = .
2n+1 · 5n/2
Thus the required expression is equal to 2n .

4679. Proposed by Daniel Sitaru.

Let (xn )n≥1 be a sequence of real numbers such that x1 = 71 , x2 = 1

5 and for n ≥ 2,

2nxn+1 · xn−1 = (n + 1)xn · xn−1 + (n − 1)xn · xn+1 .

Find
2 nxn
lim + xn .
n→∞ 3

16 correct solutions were received, plus one incomplete and one incorrect solution.
We give the solution presented essentially by most of the solvers. The main variant
was to work out the limit of the logarithm.

The answer is e−4/27 , which is about 0.86230.

Let yn = n/xn . Then y1 = 7, y2 = 10 and 2yn = yn+1 + yn−1 , so that yn = 3n + 4,

an arithmetic progression with common difference 3. Hence

n 1
xn = = − un
3n + 4 3

and
n2 1 4n2
Å ãÅ ã
nxn = = ,
3n + 4 un 3(3n + 4)2
4
where un = 3(3n+4) . Therefore

î ó(4n2 )/(3(3n+4)2 )
lim ( 23 + xn )nxn = lim (1 − un )1/un = (e−1 )4/27 = e−4/27 .
n→∞ n→∞

4680. Proposed by Pericles Papadopoulos.

In triangle ABC with |BC| > |AC| > |AB|, the circle C1 , centered at C and
with radius CA, meets the sides AB, BC at points P, Q, respectively and the
circle C2 , centered at B and with radius BA, meets the sides AC, BC at points
S, T , respectively. Let A0 be the second intersection point of C1 and C2 and let
O1 , O2 , O3 , O4 , O5 be the circumcenters of triangles CST, SAP, P BQ, BA0 Q, A0 CT ,
respectively. Prove that the points C, O1 , S, O2 , P, O3 , B, O4 , A0 , O5 are concyclic.

Crux Mathematicorum, Vol. 48(3), March 2022

 Solutions /179

We received 10 submissions. Our featured solution combines ideas from the similar
solutions by Michel Bataille, Prithwijit De, Marie-Nicole Gras, and the UCLan
Cyprus Problem Solving Group.

By assumption, the angle at A is the largest angle of ∆ABC. Comparing the two
sides of Figure 1, we see that the position of P on the line AB depends on whether
of not ∠BAC is acute or obtuse, and the same with S on AC. Consequently
care must be taken — angles that are equal in the diagram on the left might be
supplementary in the diagram on the right. (When the angle at A is a right angle,
the points P and S coincide with A.) Here we will use the symbol ∠XY Z to
represent a directed angle, namely the angle through which the line XY must be
rotated in the positive direction about the vertex Y to coincide with Y Z. With
this notation the supplement of ∠XY Z is ∠ZY X = −∠XY Z.

As a first step we show that the points A0 , B, C, P , and S all lie on the same circle,
call it ω.

Figure 1: The points A0 , B, C, P, S are concyclic.

Copyright © Canadian Mathematical Society, 2022

180/ Solutions

Since both BA = BS and CA = CP , we have

∠BP C = ∠AP C = ∠CAP = ∠ASB = ∠BSC;
therefore, B, P, S, C all lie on the circle ω. Moreover, since the triangles ABC
and A0 BC are oppositely congruent (they have opposite orientations but their
corresponding sides are equal), ∠BA0 C = ∠CAB = ∠CAP . But we have just
seen that ∠CAP = ∠BP C = ∠BSC, so that A0 joins B, P, S, and C on ω.

Figure 2: The points P, O2 , S, C (on left) and S, O1 , C, B (on right) are concyclic.
Next we see that O2 lies on ω (considered as the circle defined by the points
P, S, C); see the left side of Figure 2. Because ∆CAP is isosceles,
∠P CS = ∠P CA = 2∠P AC = 2∠P AS = ∠P O2 S,
with the final equality holding because the angle at the center of a circle is twice
any inscribed angle that is subtended by the same arc. It follows that the points
P, O2 , S, C are concyclic, as claimed.
Finally, the remaining centers Oi follow the same pattern as O2 :
• From BS = BT and O1 the center of the circle through S, T, C, we have
∠SBC = ∠SO1 C (right side Figure 2).
• From CP = CQ and O3 the center of the circle through P, Q, B, we have
∠P CB = ∠P O3 B.
• From CA0 = CQ and O4 the center of the circle through A0 , Q, B, we have
∠A0 CB = ∠A0 O4 B.
• From BA0 = BT and O5 the center of the circle through A0 , T, C, we have
∠A0 BC = ∠A0 O5 C.

Crux Mathematicorum, Vol. 48(3), March 2022