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 2019. 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 2019 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
Shaun Fallat University of Regina
Chip Curtis Missouri Southern State University
Allen O’Hara University of Western Ontario

Guest Editors Vasile Radu Birchmount Park Collegiate Institute

Aaron Slobodin Quest University Canada
Ethan White University of British Columbia
Thi Nhi Dang University of the Fraser Valley

Editor-at-Large Bill Sands University of Calgary

 IN THIS ISSUE / DANS CE NUMÉRO

378 Editorial Kseniya Garaschuk

379 MathemAttic: No. 7
379 Problems: MA31–MA35
382 Solutions: MA6–MA10
386 Problem Solving Vignettes: No. 7 Shawn Godin
390 Teaching Problems: No. 4 John McLoughlin
395 Olympiad Corner: No. 375
395 Problems: OC441–OC445
397 Solutions: OC411–OC420
407 Focus On . . . : No. 37 Michel Bataille
413 Problems: 4461–4470
417 Solutions: 4401–4420
442 Snapshot

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
378/ Editorial

EDITORIAL
Have you heard of the pancake problem? You have a stack of all different size
pancakes that you want to order from largest on the bottom to smallest on top.
You are allowed to insert a spatula at any point in the stack and use it to flip all
pancakes above it.
Let’s try a small case of 3 pancakes. How many flips are required to order the
following stacks?

More generally, what is the maximum number of flips required for n pancakes?
This problem was first proposed by Jacob E. Goodman, under the pseudonym
Harry Dweighter, in 1975 when it appeared as Elementary Problem E2569 in
American Mathematical Monthly. Here is what we know so far. In 1979, Bill
Gates and Christos Papadimitriou gave an upper bound of 5n+5 3 (yes, that Bill
Gates!). In 2008, the bound was improved to 18 11 n. In 2011, this problem was
proved to be NP-hard. Not so elementary after all.
It gets even more interesting in biology context. This “flipping” operation can be
applied to create reversals in the gene sequence, which allows us to study genome
rearrangements in evolution. For example, cabbage is only 3 flips away from
turnip! Maybe one day we will find out what came first.
Kseniya Garaschuk

Crux Mathematicorum, Vol. 45(7), September 2019

 MathemAttic /379

MATHEMATTIC
No. 7
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 October 30, 2019.

MA31. Given that the areas of an equilateral triangle with side length t and
a square with side length s are equal, determine the value of st .

MA32. Jack and Madeline are playing a dice game. Jack rolls a 6-sided
die (numbered 1 to 6) and Madeline rolls an 8-sided die (numbered 1 to 8). The
person who rolls the higher number wins the game. If Jack and Madeline roll the
same number, the game is replayed. If a tie occurs a second time, then Jack is
declared the winner. Which person has the better chance of winning? What are
the odds in favour of this person winning the game?

MA33. Observe that » »

2 23 = 2 23 .
Determine conditions for which
» »
a cb = a cb .

where a, b, c are positive integers.

MA34. Try to replace each ∗ with a different digit from 1 to 9 so that the
multiplication is correct. (Each digit from 1 to 9 must be used once.)

∗ ∗ ∗ ∗
× ∗
∗ ∗ ∗ ∗
Determine whether a solution is possible. If so, determine whether the solution is
unique.

MA35. A polygon has angles that are all equal. If the sides of this polygon
are not all equal, show that the polygon must have an even number of sides.

Copyright © Canadian Mathematical Society, 2019

380/ MathemAttic

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

Les problémes proposés dans cette section sont appropriés aux étudiants de l’école sec-
ondaire.

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 30 octobre 2019.

La rédaction souhaite remercier Rolland Gaudet, professeur titulaire à la retraite à

l’Université de Saint-Boniface, d’avoir traduit les problèmes.

MA31. Étant donné que la surface d’un triangle équilatéral de côté t et celle
d’un carré de côté s sont égales, déterminer la valeur de st .

MA32. Jacques et Madeleine jouent aux dés un peu spéciaux. Jacques utilise
un dé ordinaire à 6 côtés (numérotés 1 à 6) tandis que Madeleine se sert d’un
dé spécial à 8 côtés (numérotés de 1 à 8). La personne obtenant le plus gros
chiffre gagne. Si Jacques et Madeleine roulent le même chiffre, on répète le jeu
une seconde fois, sans déclarer de gagneur au premier tour ; si un ex aequo a lieu
de nouveau, quel que soit le chiffre, Jacques est déclaré gagneur; autrement, le
plus gros chiffre gagne. Laquelle personne a la meilleure chance de gagner à long
terme ? Cette personne gagnante a quelle probabilité de gagner à long terme ?

MA33. Sachant que

» »
2 23 = 2 23 ,
déterminer des conditions selon lesquelles
» »
a cb = a cb ,

quelque soient a, b et c entiers positifs.

MA34. Dans le schéma qui suit, chaque ∗ représente un chiffre de 1 à 9, de

façon à ce que la multiplication soit correcte, chaque chiffre de 1 à 9 étant utilisé
une seule fois.
∗ ∗ ∗ ∗
× ∗
∗ ∗ ∗ ∗
Déterminer si une solution est possible et si elle est unique.

Crux Mathematicorum, Vol. 45(7), September 2019

 MathemAttic /381

MA35. Les angles d’un polygone sont égaux. Si les côtés du polygone ne sont
pas tous de même longueur, démontrer que le polygone doit avoir un nombre pair
de côtés.

Copyright © Canadian Mathematical Society, 2019

382/ MathemAttic

MATHEMATTIC
SOLUTIONS
Statements of the problems in this section originally appear in 2019: 45(2), p. 56–57.

MA6. A rectangular sheet of paper is labelled ABCD, with AB being one of

the longer sides. The sheet is folded so that vertex A is placed exactly on top of
the opposite vertex C. The fold line is XY , where X lies on AB and Y lies on
CD. Prove that the triangle CXY is isosceles.
Originally problem B4 from 2018 UK Junior Mathematical Olympiad.
We received 2 submissions both of which were correct and complete. We present
the solution by Richard Hess, modified by the editor.
Consider the below figure where XY is the perpendicular bisector of AC and O is
the center of the rectangle ABCD.

By symmetry, OX = OY . As XY is perpendicular to AC, it follows that CXY

is an isosceles triangle where CX = CY .

MA7. Sixteen counters, which are black on one side and white on the other,
are arranged in a 4 by 4 square. Initially all the counters are facing black side up.
In one ‘move’, you must choose a 2 by 2 square within the square and turn all
four counters over once. Describe a sequence of ‘moves’ of minimum length that
finishes with the visible colours of the counters of the 4 by 4 square alternating
(as shown in the diagram).

Originally problem B6 from 2018 UK Junior Mathematical Olympiad.

We received 2 submissions of which 1 was correct and complete. We present the
solution by Richard Hess, modified by the editor.

Crux Mathematicorum, Vol. 45(7), September 2019

 MathemAttic /383

Label the 1 × 1 squares left to right, bottom to top so that 1 refers to the bottom
left square and 13 refers to the upper left square. First notice that the order in
which the 2 × 2 squares are turned is inconsequential. The 1 × 1 squares in the
bottom left and upper right begin as black, and must become white, hence the
bottom left and upper right 2 × 2 squares must both by turned. This will be the
first and second move. The squares 2,5,12, and 15 belong to exactly one 2 × 2
square, excluding the two squares turned in our first two moves. Since squares
2,5,12, and 15 are white after the second move, and must become black, we are
forced to flip the middle bottom, middle right, middle left, and middle top 2 × 2
squares. After these next four moves, the checkerboard pattern is formed. It
follows that 6 is the minimum number of moves.

MA8. I have two types of square tile. One type has a side length of 1 cm and
the other has a side length of 2 cm. What is the smallest square that can be made
with equal numbers of each type of tile?
Originally problem B5 from 2018 UK Junior Mathematical Olympiad.
We received 2 solutions. We present a solution based on the submission by Doddy
Kastanya.
Suppose we have N squares of each type tiling a square of side length S (in cm2 ).
Then
S 2 = N · 1 + N · 4 = 5N.
The smallest S that satisfies this equation is 5, which implies N = 5. However
there is no possible arrangement of the tiles satisfying this, as can be seen from
the figure below. Any 2 × 2 tile placed in the square covers exactly one of the four
grey squares. Thus we cannot fit five 2 × 2 tiles into the 5 × 5 square.

The next possible S satisfying the equation is 10, implying N = 20. A possible
tiling is shown below.

Therefore the smallest square that can be made with equal numbers of each type
of tiles has a side length of 10cm.

Copyright © Canadian Mathematical Society, 2019

384/ MathemAttic

MA9. The letters a, b, c, d, e and f represent single digits and each letter rep-
resents a different digit. They satisfy the following equations:

a + b = d, b + c = e, d + e = f.

Find all possible solutions for the values of a, b, c, d, e and f .

Originally problem B8 from 2018 UK Junior Mathematical Olympiad.
We received two submissions both of which were correct and complete. We present
the solution by Richard Hess, modified by the editor.
Given a + b = d, b + c = e, d + e = f , we can combine equations to produce
f = a + c + 2b. Given a solution (a, b, c, d, e, f ) that satisfies these equations,
we can interchange a and d with c and e, respectively, to find a second solution
(c, b, a, e, d, f ). Thus, we need only search for solutions where a < c. As all digits
are unique in the interval 1 to 9, the smallest f can be is 7 (a = 2, b = 1, c = 3).
Below are the 8 solutions to this problem:
(2, 1, 4, 3, 5, 8) (4, 1, 2, 5, 3, 8)
(2, 1, 5, 3, 6, 9) (5, 1, 2, 6, 3, 9)
(1, 2, 4, 3, 6, 9) (4, 2, 1, 6, 3, 9)
(1, 3, 2, 4, 5, 9) (2, 3, 1, 5, 4, 9)

MA10. An arithmetic and a geometric sequence, both consisting of only

positive integral terms, share the same first two terms. Show that each term of
the geometric sequence is also a term of the arithmetic sequence.
Originally Problem J12, proposed by Colin Springer, from Mathematical Mayhem,
1 (1), p.19.
We received 5 submissions, of which all were correct and complete. We present
the solution by Jacob Miles, modified by the editor.
Let our arithmetic sequence be given by

a, a + d, a + 2d, . . .

and our geometric sequence be given by

α, αr, αr2 , . . .

We show by proof by induction that each term αrn of our geometric sequence is a
term of our arithmetic sequence. We are given that a = α and a + d = αr. Hence,
the base cases of n = 0, 1 are satisfied. Let n = k. Assume αrk is a term of our
arithmetic sequence, i.e. αrk = a + md for some m ∈ N. We consider the case of
n = k + 1 as follows
αrk+1 = αrk r

= (a + md)r
= ar + rmd.

Crux Mathematicorum, Vol. 45(7), September 2019

 MathemAttic /385

Given that ar = a + d, the above becomes

αrk+1 = a + d + rmd
= a + (rm + 1)d.

The above expression of αrk+1 is a member of the arithmetic sequence if rm + 1 ∈

N. As m ∈ N it suffices to prove that r ∈ N. It is clear that r > 0 and that
r ∈ Q, for if either were not the case, we would have that ar = a + d 6∈ N which is
a contradiction. We assume that r ∈ Q but r 6∈ N. It follows that r = xy for some
x, y ∈ N where y 6= 1. Without loss of generality, assume gcd(x, y) = 1. As each
term of our geometric sequence is a positive integer, we have that arn ∈ N for all
n ∈ N. Hence Å ãn
x axn
arn = a = n .
y y
We have that y n |axn . However, since gcd(x, y) = 1, it follows that gcd(xn , y n ) = 1.
This implies y n |a. However, as y 6= 1 we can choose a sufficiently large n such that
y n > a causing y n - a. This would imply that arn 6∈ N. By proof by contradiction
we conclude that r ∈ N and that every term in our geometric sequence is in our
arithmetic sequence.

Copyright © Canadian Mathematical Society, 2019

386/ Problem Solving Vignettes

PROBLEM SOLVING
VIGNETTES
No.7
Shawn Godin
Careful Counting
This month we will look at problem B4 from the 2018 Canadian Open Mathematics
Challenge administered by the CMS. You can check out past contests on the CMS
webpage at cms.math.ca/Competitions/COMC.
Determine the number of 5-tuples of integers (x1 , x2 , x3 , x4 , x5 ) so that
(a) xi ≥ i for 1 ≤ i ≤ 5;
5
X
(b) xi = 25.
i=1

Solution 1: We will look to see if we can find any patterns by considering possible
solutions in an orderly manner. Suppose that we fix x1 = 1, x2 = 2, and x3 = 3.
If we want the five numbers to add to 25, then x4 + x5 = 25 − (1 + 2 + 3) = 19.
Recall that we also need x4 ≥ 4 and x5 ≥ 5. Putting this together, we get the
following 11 5-tuples:

(1, 2, 3, 4, 15), (1, 2, 3, 5, 14), (1, 2, 3, 6, 13), . . . , (1, 2, 3, 14, 5).

Next we will examine what happens when we allow x3 to take on different values.
We will keep x1 = 1 and x2 = 2 and let x3 = 4. Using the same idea as in the
first case we get 10 new 5-tuples:

(1, 2, 4, 4, 14), (1, 2, 4, 5, 13), (1, 2, 4, 6, 12), . . . , (1, 2, 4, 13, 5).

Thus, if we fix x1 = 1 and x2 = 2, the total number of 5-tuples is

11
X
11 + 10 + 9 + · · · + 1 = i = 66.
i=1

Now, consider what happens if we fix x1 = 1 and let x2 = 3. If we go through the

same process again, we get 10 + 9 + 8 + · · · + 1 = 55 more 5-tuples. So if we fix
x1 = 1 and let all other values vary we get

11 X
X i
(11 + 10 + 9 + · · · + 1) + (10 + 9 + 8 + · · · + 1) + · · · + (2 + 1) + 1 = j
i=1 j=1

Crux Mathematicorum, Vol. 45(7), September 2019

 Shawn Godin /387

5-tuples. Thus for the total problem, we will need to let x1 vary. Letting x1 = 2
and thinking through the process we get
10 X
X i
(10 + 9 + 8 + · · · + 1) + (9 + 8 + 7 + · · · + 1) + · · · + (2 + 1) + 1 = j
i=1 j=1

new 5-tuples. Hence looking at all possible solutions we must have

[(11 + 10 + 9 + · · · + 1) + · · · + (2 + 1) + 1]+
11 X
X j
i X
[(10 + 9 + 8 + · · · + 1) + · · · + (2 + 1) + 1]+ · · · + [1] = k. (1)
i=1 j=1 k=1

In Vignette #5 [2019: 45(5), p. 236-240] we introduced and proved the following

formulas:
n
X n(n + 1)
1 + 2 + 3 + ··· + n = i= , (2)
i=1
2
n
X n(n + 1)(2n + 1)
1 2 + 2 2 + 3 2 + · · · + n2 = i2 = . (3)
i=1
6

We will add one more which will be of use to us in our solution:

n
X n2 (n + 1)2
13 + 23 + 33 + · · · + n3 = i3 = . (4)
i=1
2

Enjoy practicing your induction by proving that the formula holds for all n.
Going back to (1), using (2), (3), and (4) we get
11 X j
i X 11 X
i
X X j(j + 1)
k=
i=1 j=1 k=1 i=1 j=1
2
11 i
1 XX 2
= (j + j)
2 i=1 j=1
11 Å
1 X i(i + 1)(2i + 1) i(i + 1)
ã
= +
2 i=1 6 2
11
1X 3
= (i + 3i2 + 2i)
6 i=1
1 112 · 122 11 · 12 · 23 11 · 12
Å ã
= +3· +2·
6 4 6 2
= 1001.

Therefore there are 1001 5-tuples that satisfy the conditions in the problem. 2

Copyright © Canadian Mathematical Society, 2019

388/ Problem Solving Vignettes

Solution 2: We will look at the problem from another point of view. Suppose
we wanted, for case of simplicity, to find all 3-tuples of non-negative integers
(x1 , x2 , x3 ) such that x1 + x2 + x3 = 5. This is a simplification of our problem
by considering only 3 numbers, having a smaller sum and letting them all be any
non-negative integer. For this problem we could list out all the possibilities or
count them by carefully looking at cases.
Case 1: Two of the numbers are the same (there is no way they can all be the
same). There are three ways that this can happen: (0, 0, 5), (1, 1, 3), and (2, 2, 1).
For each of these cases there are 3! 2! = 3 ways to arrange the numbers giving 3·3 = 9
3-tuples: (0, 0, 5), (0, 5, 0), (5, 0, 0), (1, 1, 3) (1, 3, 1), (3, 1, 1), (2, 2, 1), (2, 1, 2), and
(1, 2, 2).
Case 2: None of the numbers are the same. There are only two ways that this
can happen: (0, 1, 4) and (0, 2, 3). For each of these cases there are 3! = 6 ways
to arrange the numbers giving 2 · 6 = 12 more 3-tuples: (0, 1, 4), (0, 4, 1), (1, 0, 4),
(1, 4, 0), (4, 0, 1), (4, 1, 0), (0, 2, 3), (0, 3, 2), (2, 0, 3), (2, 3, 0), (3, 0, 2), and (3, 2, 0).
Therefore there are 9 + 12 = 21 3-tuples in total. We can use this method on our
problem, but there will be many more cases to look at. You may (or may not!)
want to see if you can identify all cases and get the correct total of 1001.
Still looking at the simplified problem, suppose we represent any particular 3-tuple
with a collection of stars and bars (the name usually associated with this method).
We will use five stars, since the total is 5 and two bars to separate them into three
groups. Thus the 3-tuple (2, 1, 2) would be represented by ∗ ∗ | ∗ | ∗ ∗. All stars to
the left of the first bar represent x1 , the stars between the bars represent x2 and
the stars to the right of the second bar represents x3 . Similarly ∗|| ∗ ∗ ∗ ∗ would
represent (1, 0, 4) and ∗ ∗ ∗ ∗ ∗|| would represent (5, 0, 0).
Every 3-tuple can be represented by a unique permutation of 5 stars and 2 bars.
Similarly, every permutation of 5 stars and 2 bars represents a unique 3-tuple.
There is a one-to-one correspondence between the 3-tuples and the permutations
of 5 stars and 2 bars. Since the total number of permutations of 5 stars and 2 bars
7!
is 5!2! = 21, we solved our simplified problem in a much more efficient manner.
If we return to the original problem, all permutations of 25 stars and 4 bars would
give all possible 5-tuples of non-negative integers that sum to 25. This is not quite
what we are after, but if we let xi = i + yi , for 1 ≤ i ≤ 5 then (y1 , y2 , y3 , y4 , y5 )
is a 5-tuple of non-negative integers that add to 25 − (1 + 2 + 3 + 4 + 5) = 10
and there is a one-to-one correspondence between the 5-tuples (y1 , y2 , y3 , y4 , y5 )
and the 5-tuples that we are after. That is, for example, since (3, 1, 4, 2, 0) is a
collection of yi s, then (3 + 1, 1 + 2, 4 + 3, 2 + 4, 0 + 5) = (4, 3, 7, 6, 5) is an allowed
solution to the original problem. The number of possible 5-tuples (y1 , y2 , y3 , y4 , y5 )
is the same as the number of permutations of 10 stars and 4 bars or

14!
= 1001.
10!4!
2

Crux Mathematicorum, Vol. 45(7), September 2019

 Shawn Godin /389

The stars and bars method yields a solution much quicker. We can summarize it
as follows: the number of distinct n-tuples of non-negative integers whose sum is
s is
(s + n − 1)!
.
s!(n − 1)!
This can be generalized to the following statement: the number of distinct n-tuples
of integers, (x1 , x2 , . . . , xn ) whose sum is s, where xi ≤ mi for 1 ≤ i ≤ n is

(s + (n − 1) − ni=1 mi )!
P
.
(s − ni=1 mi )!(n − 1)!
P

We will explore other counting techniques in future columns.

Copyright © Canadian Mathematical Society, 2019

390/ Teaching Problems

TEACHING PROBLEMS
No.4
John McLoughlin
Handshakes with a Twist

The handshake problem is perhaps the most flexible problem in my experience

working with K-12 teachers. The basic example of counting handshakes lends itself
to modeling the results with people as the participants. For instance, four people
can each shake hands with one another. The total of 6 handshakes is reached this
way at any level. The case of five people can be a new problem, or an extension
can be readily grasped using the idea of a fifth person arriving in the group, thus,
adding 4 handshakes to give a total of 10 handshakes. Before proceeding further,
it is assumed that the reader is convinced that a total of 10 handshakes would
occur if all people in a group of five shook hands with one another.

Assuming that makes sense be aware of how you would obtain the result. There
are at least three viable avenues including the idea of adding additional people
to simpler cases, as outlined above. You may recognize that each person in the
group must shake 4 hands suggesting that there may be 5 × 4 = 20 handshakes
required. Of course, this is not 10, and so something must be awry. Yes, each
handshake involves two people and so the number of handshakes in total would
equal 20 divided by 2, or 10. Do you see that in general for n people there would
be
n(n − 1)/2 handshakes required? Mathematically this is equivalent to n C2 or
n
2 or “n choose 2”, namely, the number of ways of selecting 2 people from a group
of n people.

My experience suggests that a diagrammatic approach may offer another valuable

way of representing the problem. You are invited to construct a diagram with five
vertices, each representing a person. Joining all possible vertices with segments
will illustrate that there are 10 possible ways of connecting two of the five people.
These segments represent the handshakes. Following our discussion of some of
the basic handshake counting principles, we are ready to tackle an unorthodox
handshake problem. The use of a diagram may be helpful in considering the core
problem of our discussion as stated below. Teachers may wish to have eight people
represent the characters in the problem.

Mr. and Mrs. Smith were at a party with three other married couples.
Since some of the guests were not acquainted with one another, various
handshakes took place. No one shook hands with his or her spouse, and
of course, no one shook their own hand! After all of the introductions
had been made, Mrs. Smith asked the other seven people how many
hands each shook. Surprisingly, they all gave different answers. How
many hands did Mr. Smith shake?

Crux Mathematicorum, Vol. 45(7), September 2019

 John McLoughlin /391

A detailed discussion of this problem appears in Combinatorial Explorations, a

publication in the ATOM Series authored by Richard Hoshino and John Grant
McLoughlin. “This problem is fascinating because it does not appear solvable. It
is difficult to imagine that there is enough information here. However, we have all
the information we need! Before reading any further, stop and attempt to solve
this problem on your own.”

Take some time and consider a diagram and/or a logical approach that makes it
plausible to address the problem. It is not anticipated that you will necessarily
solve the problem, as few of my students do in fact without some further guidance.
In any case, play with the problem so that you may understand it better.

Here I will share two approaches, the first being a classroom approach and the
second being the written approach.

Classroom Approach: Modeling the Problem

This approach requires 8 volunteers who are arranged into 4 couples, one of which
is designated as Mr. and Mrs. Smith. Any names for the others are not a concern.
It is easiest to place them in pairs square dancing, as if they are the four directions
(N , S, E, W ) on a compass.

Consider the important fact that Mrs. Smith received seven different answers to
the number of hands shaken by the others in the group. In fact, there are only
seven possible answers as the absence of a handshake with one’s spouse limited
the number of handshakes to a maximum of 6. That is, there must be people who
accounted for each of 0, 1, 2, 3, 4, 5, and 6 handshakes. That is essential to getting
started. Make sure that makes sense to you. The eighth person, Mrs. Smith, is
not included on that list.

Here I ask one of the volunteers other than Mr. Smith to be the person shaking 6
hands. This person steps forward and proceeds to shake all possible hands (as in
all people other than the spouse), thus, giving us a person with 6 handshakes. Do
you see now that the spouse of this individual is the only person who could shake
0 hands? Hence, both members of that couple have completed their handshakes.

We proceed to identify another person other than Mr. Smith to be the person
who will shake 5 hands. Observe that this person will have already counted 1
handshake and now can shake hands with Mr. and Mrs. Smith as well as with
another couple. The spouse of the person shaking 5 hands will become the only
person who can shake only 1 hand. Hence, these people have finished with shaking
hands.

Finally, there is a volunteer who shakes more hands and becomes the person with
4 handshakes while having a spouse with only 2 handshakes. This leaves Mr. and
Mrs. Smith each having shaken 3 hands. We are done with the handshakes and
can definitively answer the question. How many hands did Mr. Smith shake? The
answer is 3.

Copyright © Canadian Mathematical Society, 2019

392/ Teaching Problems

Written Approach: Using a Diagram

Let us model this problem graphically using a diagram. This can be thought of as
a graph with 8 vertices labeled A, B, C, D, E, F , G, and H. Suppose that A is
married to B, C is married to D, E is married to F , and G is married to H. Note
that in the classroom model, Mr. and Mrs. Smith were positioned first. In this
diagrammatic approach, their location will not be apparent until the completion
of the process. That is, the coupling in pairs is required, but not the naming of
any of the pairs.
Each vertex represents a person at the party. Two vertices will be joined if those
two people shook hands. Again note that since no one shakes their own hand,
or the hand of their spouse, a person can shake at most six hands. Thus, every
person at the party shook at least 0 hands and at most 6 hands.
Consider the person who shook 6 hands. Let us assume this person is denoted as
A. Hence, A shook hands with everyone at the party except for B. Represent this
by drawing an edge from A to each of the other vertices in the graph, except for B.
Thus, every person other than B has shaken at least one hand. Since someone at
the party shook 0 hands, this implies that B must have been the person who shook
0 hands. This information is represented in the diagram below. In this diagram,
the oval shape around A and B signifies that this couple has finished performing
all of their handshakes.

Now consider the person who shook 5 hands. Assume this person is C. Then C
must shake hands with each of E, F , G, and H. This shows that everyone (other
than B and D) shook at least two hands. Therefore, it follows that D must have
been the person who shook exactly 1 hand, as illustrated.

Crux Mathematicorum, Vol. 45(7), September 2019

 John McLoughlin /393

Now consider the person who shook 4 hands. Assume this person is E. Since E
has already shaken two hands, E must shake hands with both G and H. Drawing
edges from E to G and E to H we see that F must be the person who shook 2
hands. Further, both G and H have shaken at least three hands. The resulting
diagram is shown:

Note that G and H are married and so they do not shake hands. Thus, both G and
H shook three hands. We have now indicated all the handshakes that took place
at this party. However, we need to identify Mr. and Mrs. Smith. If Mrs. Smith
is any of A to F , two of the individuals would have replied to her question that
they shook exactly three hands. That is a contradiction because all seven replies
were different. Therefore, Mrs. Smith must be either G or H. Thus, Mrs. Smith
shook three hands. Likewise, Mr. Smith as her spouse shook exactly 3 hands.

Questions and thoughts for consideration with the problem

i) Considering the classroom approach, why could we not have let Mr. Smith
be the volunteer who shook extra hands?
Suppose that he began by shaking 6 hands. It would have then been impos-
sible for any of the people other than Mrs. Smith to have had 0 handshakes,
thus, violating the conditions of the problem. (Do you see that we needed
someone else to shake no hands?) It is more difficult to consider other cases
along the way, but you may wish to ponder those also.
ii) How does one know that Mrs. Smith had to shake an odd number of hands?
iii) Observe in the concluding diagram that all couples shook a total of 6 hands.
Is this a coincidence or can you explain why this must be the case?

Some problems to try

Here are a few handshake questions to consider.
1. Suppose that twenty people attended a party, and everyone shook hands with
exactly three guests. How many handshakes took place?
2. Mr. and Mrs. Smith were at a party with ten other married couples. Various

Copyright © Canadian Mathematical Society, 2019

394/ Teaching Problems

handshakes took place. No one shook hands with their spouse, and of course, no
one shook their own hand! After all the introductions had been made, Mrs. Smith
asked the other people how many hands they shook. Surprisingly, they all gave a
different answer. How many hands did Mr. Smith shake?
3. Everyone at a meeting shook hands with one another. Shortly after the meeting
commenced, the chronically late character known as Tar D. arrived. Tar only
managed to shake hands with some of the people present. In total, there were 59
handshakes. How many hands did Tar D. shake?
4. At a party attended by n people, various handshakes took place. Just for fun,
each person shouted out the number of hands they shook. Explain why there must
have been at least two people who shouted out the same number.

Reference
Hoshino, R. & Grant McLoughlin, J. (2005). Combinatorial Explorations. Ottawa:
Canadian Mathematical Society.

Crux Mathematicorum, Vol. 45(7), September 2019

 OLYMPIAD CORNER /395

OLYMPIAD CORNER
No. 375
The problems in this section have appeared in a regional or national Olympiad.

Click here to submit solutions, comments and generalizations to any

problem in this section

To facilitate their consideration, solutions should be received by October 30, 2019.

OC441. Let f : [0, ∞) → (0, ∞) be a continuous function.

(a) Prove that there exists a natural number n0 such that for any natural number
n > n0 there exists a unique real number xn > 0 for which
Z xn
n f (t) dt = 1;
0
(b) Prove that the sequence (nxn )n≥1 is convergent and find its limit.

OC442. Let H = {1, 2, . . . , n}. Are there two disjoint subsets A and B such
that A∪B = H and such that the sum of the elements in A is equal to the product
of the elements in B if (a) n = 2016? (b) n = 2017?

OC443. In a triangle ABC, the foot of the altitude drawn from A is T and
the angle bisector of ∠B intersects side AC at D. If ∠BDA = 45◦ , find ∠DT C.

OC444. We have n2 empty boxes, each with a square bottom. The height
and the width of each box are natural numbers in the set {1, 2, . . . , n}. Each box
differs from any other box in at least one of these two dimensions. We are allowed
to insert a box into another if each dimension of the first box is smaller than the
corresponding dimension of the second box and at least one of the dimensions is
at least units less that the corresponding larger box dimension. In this way, we
can create a sequence of boxes inserted into each other in the same orientation
(i.e. the first box is inside the second, the second box is inside the third, etc.). We
store each sequence of boxes on a shelf with each shelf holding one set of nested
boxes. Determine the smallest number of shelves needed to store all the n2 boxes.

OC445. There are 100 diamonds in a pile, of which 50 are genuine and 50
are fake. We invited a distinguished expert, who can recognize which diamonds
are genuine. Each time we show him three diamonds, he chooses two of them
and (truthfully) tells whether they are both genuine, one genuine or none genuine.
Establish if we can guarantee to spot all the genuine diamonds no matter how the
expert chooses the judged pair.

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

Copyright © Canadian Mathematical Society, 2019

396/ OLYMPIAD CORNER

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 30 octobre 2019.
La rédaction remercie Valérie Lapointe, Carignan, QC, d’avoir traduit les problèmes.

OC441. Soit f : [0, ∞) → (0, ∞) une fonction continue.

(a) Prouvez qu’il existe un nombre naturel n0 tel que pour tout nombre n > n0 ,
il existe un unique nombre réel xn > 0 pour lequel
Z xn
n f (t) dt = 1;
0
(b) Prouvez que la suite (nxn )n≥1 est convergente et trouvez le résultat de sa
limite.

OC442. Soit H = {1, 2, . . . , n}. Existe-t-il deux sous-ensembles disjoints A

et B tels que A ∪ B = H et tels que la somme des éléments dans A est égale au
produit des éléments dans B si (a) n = 2016? (b) n = 2017?

OC443. Dans un triangle ABC, l’extrémité de la hauteur issue de A est T et

la bissectrice de ∠B intercepte le côté AC en D. Si ∠BDA = 45◦ , trouvez ∠DT C.

OC444. On a n2 boı̂tes vides, chacune à fond carré. La hauteur et la largeur

de chaque boı̂te est un nombre naturel de l’ensemble {1, 2, . . . , n}. Chaque boı̂te
est différente d’une autre sur au moins une des deux dimensions. On peut entrer
une boı̂te dans une autre si les deux dimensions sont plus petites et qu’au moins
une des deux dimensions est au moins deux unité plus petite. On peut ainsi créer
une suite de boı̂tes à l’intérieur d’une autre (i.e. la première boı̂te est à l’intérieur
de la deuxième, la deuxième boı̂te est à l’intérieur de la troisième, etc.). On range
une telle suite de boı̂te sur une étagère. Déterminez le plus petit nombre d’étagères
nécessaires pour ranger toutes les n2 boı̂tes.

OC445. Il y a 100 diamants dans une pile dans laquelle 50 sont véritables
et 50 sont faux. On invite un expert qui peut reconnaı̂tre quels diamants sont
véritables. À chaque fois qu’on lui montre trois diamants, il en choisit deux et dit
(honnêtement) s’ils sont soit tous les deux véritables, si un seul l’est ou si aucun
ne l’est. Déterminez si on peut garantir de trouver tous les diamants véritables
peu importe la façon dont l’expert choisit la paire jugée.

Crux Mathematicorum, Vol. 45(7), September 2019

 OLYMPIAD CORNER /397

OLYMPIAD CORNER
SOLUTIONS
Statements of the problems in this section originally appear in 2019: 45(1), p. 17–18;
45(2): 69–71.

OC411. Show that for all integers k > 1 there is a positive integer m less than
k 2 such that 2m − m is divisible by k.
Originally 2017 Hungary Math Olympiad, 3rd Problem, 3rd Category, Final Round.
We received no submissions for this problem.

OC412. Find all the functions f : R → R such that for all real numbers x, y

f (y − xy) = f (x)y + (x − 1)2 f (y).

Originally 2017 Czech-Slovakia Math Olympiad, 3rd Problem, Final Round.

We received 5 submissions of which 4 were correct. We present the solution by
Sundara Narasimhan.
We evaluate the relation at x = 0 and y = 1 to find 0 = f (0).
We evaluate the relation at x = 1 and y = 1 to find f (0) = f (1).
We evaluate the relation at x = x and y = 1, and use f (0) = f (1) = 0 to find
f (1 − x) = f (x).
We make the substitution 1−x = t in the original relation, and use f (1−x) = f (x)
to get for any t ∈ R and y ∈ R

f (yt) = f (t)y + t2 f (y).

We interchange y and t to get f (ty) = f (y)t + y 2 f (t). Since f (ty) = f (yt), we find
that for any t ∈ R and y ∈ R

(t2 − t)f (y) = (y 2 − y)f (t).

We take t = 2 in the last relation to find that for any y ∈ R

f (2) 2
f (y) = (y − y).
2

Therefore, the solutions of our functional equation must be of the form

f (x) = c(x2 − x),

Copyright © Canadian Mathematical Society, 2019

398/ OLYMPIAD CORNER

for some real constant c. In fact, we can check that any function of this form is a
solution of the original relation. We established that the set of all functions that
satisfy the original relation are f (x) = c(x2 − x), with c being a real constant.

OC413. To each sequence consisting of n zeros and n ones is assigned a number

which is the number of largest segments with the same digits in it (for example,
the sequence 00111001 has 4 such segments 00, 111, 00, 1). For each n, add all
the numbers assigned to each sequence. Prove that the resulting sum is equal to
Ç å
2n
(n + 1) .
n

Originally 2017 Czech-Slovakia Math Olympiad, 4th Problem, Final Round.

We received one submission. We present the solution of Kathleen Lewis.
The total number of distinct sequences of n zeroes and n ones is 2n


n . The number
of largest same-digit segments of such sequence has a range between 2 and 2n. The
minimum number of 2 is displayed by two sequences that have all zeros together
and all ones together
(000 . . . 111 . . . , 111 . . . 000 . . . ).
The maximum number is displayed by two sequences that have alternating zeros
and ones
(101010 . . . , 010101 . . . ).
For a natural number j between 2 and 2n, let Nj be the number of sequences that
have exactly j largest same-digit segments.
First we show that for any j, Nj = N2n+2−j , in other words N2 , N3 , . . . , N2n+1 ,
N2n+2 are symmetrical about n + 1. In fact we can calculate Nj .
Case 1. Assume j is even, i.e. j = 2k for some natural number k.
Since the sequence has j same-digit blocks, k of these are blocks of zeroes and the
remaining k are blocks of ones. The sequence is uniquely determined by the points
where we cut the original list of n zeros and the original list of n ones. k − 1 cuts
need to be made to obtain k blocks, and these cuts are selected from n − 1 links
between the original

n zeros. Therefore the original sequence of n zeros can be cut
in k blocks in n−1
k−1 ways. Similarly for ones. Hence
Ç åÇ å
n−1 n−1
Nj = 2 .
k−1 k−1
The number 2 was added to the above expression to account for whether the
sequence starts with zero or one.
Since j is even, it follows that 2n + 2 − j = 2(n + 1 − k) is even, and
Ç åÇ å
n−1 n−1
N2n+2−j = 2
n−k n−k

Crux Mathematicorum, Vol. 45(7), September 2019

 OLYMPIAD CORNER /399

Using properties of binomial coefficients

Ç åÇ å
n−1 n−1
N2n+2−j = 2 = Nj .
(n − 1) − (n − k) (n − 1) − (n − k)

Case 2. Assume j is odd, i.e. j = 2k + 1 for some natural number k.

Since the sequence has j same-digit blocks, k + 1 of these are blocks of zeroes and
the remaining k are blocks of ones, or vice versa k blocks of zeros and k + 1 blocks
of ones. Using arguments that we invoked at case 1 we show that
Ç åÇ å
n−1 n−1
Nj = 2 .
k k−1

Since j is odd, it follows that 2n + 2 − j = 2(n − k) + 1 is odd, and

Ç åÇ å
n−1 n−1
N2n+2−j = 2
n−k−1 n−k

Using properties of binomial coefficients

Ç åÇ å
n−1 n−1
N2n+2−j = 2 = Nj .
(n − 1) − (n − k − 1) (n − 1) − (n − k)

Now we can proceed to calculate the required sum

S = 2N2 + 3N3 + · · · + (2n − 1)N2n−1 + (2n)N2n .

Because Nj = N2n+2−j , we have

2S = (2N2 + (2n)N2n ) + (3N3 + (2n − 1)N2n−1 ) + · · · + ((2n)N2n + 2N2 )

= (2 + 2n)N2 + (3 + 2n − 1)N3 + · · · + (2n − 1 + 2)N2n−1 + (2n + 2)N2n
= 2(n + 1)(N2 + N3 + · · · + N2n−1 + N2n ).

However, N2 + N3 + · · ·
+ N2n−1 + N2n is the total number of
sequences of n zeros

and n ones, namely 2nn . Therefore, the sum S = (n + 1) n .
2n

An interesting interpretation of this result is that the average number of largest

same-digit segments in a sequence of n zeros and n ones is n + 1. And this is
mainly due to the fact that the distribution of the number of largest same-digit
segments is symmetrical about n + 1.

OC414. Find all prime numbers p and all positive integers a and m such that
a ≤ 5p2 and (p − 1)! + a = pm .
Originally 2017 Bulgaria Math Olympiad, 4th Problem, Grade 9-12, Final Round.
We received only one incomplete submission, which we do not present here.

Copyright © Canadian Mathematical Society, 2019

400/ OLYMPIAD CORNER

OC415. Let n be a positive integer and let f (x) be a polynomial of degree

n with real coefficients and n distinct positive real roots. Is it possible for some
integer k ≥ 2 and for a real number a that the polynomial

x(x + 1)(x + 2)(x + 4)f (x) + a

is the k-th power of a polynomial with real coefficients?

Originally 2017 Bulgaria Math Olympiad, 5th Problem, Grade 9-12, Final Round.
We received no submissions for this problem.

OC416. Given an acute nonisosceles triangle ABC with altitudes CD, AE,
BF . Points E 0 and F 0 are symmetrical to E and F with respect to points A and
−−→
B, respectively. Take a point C1 on the ray CD such that DC1 = 3CD. Prove
that ∠E 0 C1 F 0 = ∠ACB.
Originally 2017 Bulgaria Math Olympiad, 6th Problem, Grade 9-12, Final Round.
We received 3 submissions and we present 2 of them.
Solution 1, by Oliver Geupel.
We drop the constraint that triangle ABC is acute and nonisosceles, and prove
the result for an arbitrary triangle ABC. Moreover, we prove the stronger result
that the triangles ABC and E 0 F 0 C1 are similar.
We work in the complex plane. We use lower-case letters to denote the complex-
number representations of geometrical points denoted by corresponding upper-case
letters. For example a is the complex number assigned to point A. We assume
without loss of generality that the points A, B, and C are on the unit circle.
First we recall the result that the foot of the perpendicular from an arbitrary point
P to the chord XY of the unit circle is the point specified by the complex number
1
(p + x + y − xy p̄).
2
Hence,

1 1 1
Å ã Å ã
ab bc ca 
d= a+b+c− , e= a+b+c− , f= a+b+c− .
2 c 2 a 2 b

Moreover, since points E 0 and F 0 are symmetrical to E and F with respect to

points A and B, respectively

1 1
Å ã
bc ca 
e0 = a + (a − e) = 3a − b − c + , f 0 = b + (b − f ) = 3b − c − a + .
2 a 2 b

Also,
2ab
c1 = d + 3(d − c) = 2a + 2b − c − .
c

Crux Mathematicorum, Vol. 45(7), September 2019

 OLYMPIAD CORNER /401

Next, we compute a(c − b)(e0 − c1 ) and b(c − a)(f 0 − c1 ) to find that

a(c − b)(e0 − c1 ) = b(c − a)(f 0 − c1 )

5 2 1 1 a2 b2
= (a b + ab2 ) + (bc2 − cb2 ) + (ac2 − a2 c) − 3abc − 2 .
2 2 2 c
Thus,
e 0 − c1 b(c − a) (1/a) − (1/c) ā − c̄
0
= = = .
f − c1 a(c − b) (1/b) − (1/c) b̄ − c̄
This equality of complex numbers implies,
C1 E 0 CA
= and ∠E 0 C1 F 0 = ∠ACB.
C1 F 0 CB

This completes the proof.

Solution 2, by Andrea Fanchini.

We use Conway triangle notations: S stands for twice the area of 4ABC, SA =
S cot ∠BAC, SB = S cot ∠ABC, and SC = S cot ∠ACB.
We use barycentric coordinates with reference to the triangle ABC:

D(SB : SA : 0), E(0 : SC : SB ), F (SC : 0 : SA )

E 0 (−2a2 : SC : SB ), F 0 (SC : −2b2 : SA ).

Since the point C1 divides the segment CD in the ratio (−4 : 3), it follows that

CC1 /C1 D = (−4)/3 and C1 (4SB : 4SA : −3c2 ).

Copyright © Canadian Mathematical Society, 2019

402/ OLYMPIAD CORNER

Therefore, the lines C1 E 0 , and C1 F 0 are

C1 E 0 : (SA SB + 3S 2 )x + 2(SB
2
+ 3S 2 )y + 4(a2 SA + S 2 )z = 0,

C1 F 0 : 2(SA
2
+ 3S 2 )x + (SA SB + 3S 2 )y + 4(b2 SB + S 2 )z = 0,
and the intersection points of these lines with the line AB are

E 00 = C1 E 0 ∩ AB = 2(SB
2
+ 3S 2 ) : −(SA SB + 3S 2 ) : 0 ,

F 00 = C1 F 0 ∩ AB = SA SB + 3S 2 : −2(SA
2
+ 3S 2 ) : 0 .

We calculate

∠E 0 C1 F 0 = ∠F 00 C1 D + ∠E 00 C1 D
F 00 D E 00 D E 00 F 00 · C1 D
= arctan + arctan = arctan ,
C1 D C1 D C1 D2 − E 00 D · F 00 D
where
3 (a2 SA + S 2 )(2b2 + SC ) + (b2 SB + S 2 )(2a2 + SC )


00 00
E F = ,
c(2a2 + SC )(2b2 + SC )

3S 3(a2 SA + S 2 ) 3(b2 SB + S 2 )
C1 D = , E 00 D = , F 00 D =
c c(2a2 + SC ) c(2b2 + SC )

Therefore,

S (a2 SA + S 2 )(2b2 + SC ) + (b2 SB + S 2 )(2a2 + SC )

0 0
∠E C1 F = arctan 2 2
S (2a + SC )(2b2 + SC ) − (a2 SA + S 2 )(b2 SB + S 2 )
SC (8SC S 2 + 3c2 S 2 + a2 b2 c2 )
= arctan
S(8SC S 2 + 3c2 S 2 + a2 b2 c2 )
SC
= arctan
S
= ∠ACB.

OC417. Point M is the midpoint of side BC of a triangle ABC in which

AB = AC. Point D is the orthogonal projection of M onto side AB. Circle ω
is inscribed in triangle ACD and tangent to segments AD and AC at K and L,
respectively. Lines tangent to ω which pass through M intersect line KL at X
and Y , where points X, K, L and Y lie on KL in this order. Prove that points
M , D, X and Y are concyclic.
Originally 2017 Poland Math Olympiad, 5th Problem, Final Round.
We received no submissions for this problem.

Crux Mathematicorum, Vol. 45(7), September 2019

 OLYMPIAD CORNER /403

OC418. Three sequences (a0 , a1 , . . . , an ), (b0 , b1 , . . . , bn ), (c0 , c1 , . . . , c2n ) of

nonnegative real numbers are given such that for all 0 ≤ i, j ≤ n we have ai bj ≤
(ci+j )2 . Prove that
n n 2n
!2
X X X
ai · bj ≤ ck .
i=0 j=0 k=0

Originally 2017 Poland Math Olympiad, 6th Problem, Final Round.

We received no submissions for this problem.

OC419. Prove that there exist infinitely many positive integers m such that
there exist m consecutive perfect squares with sum m3 . Determine one solution
with m > 1.
Originally 2017 Germany Math Olympiad, 6th Problem, Final Round.
We received 6 correct submissions. We present a solution that follows the sub-
missions of the Problem Solving Group of Missouri State University and David
Manes. At the end, we include a list of examples by Dominique Mouchet.
We start by computing the difference between m3 and the sum of m arbitrary
consecutive perfect squares:
a+m−1 a+m−1 a−1
!
X X X
3 2 3 2 2
m − i =m − i − i
i=a i=1 i=1
(a + m − 1)(a + m)(2a + 2m − 1) (a − 1)a(2a − 1)
Å ã
= m3 − −
6 6
m 4m2 − 6am + 3m − 6a2 + 6a − 1


= .
6
Therefore, for any pair of positive integers (a, m) that satisfy

f (a, m) = 4m2 − 6am + 3m − 6a2 + 6a − 1 = 0,

we have m consecutive squares summing to m3 .

Let a0 = 1 and m0 = 1 and recursively define for any n ≥ 1

an = 11an−1 + 16mn−1 − 5 (1)

mn = 24an−1 + 35mn−1 − 12.

We prove by induction that f (an , mn ) = 0, an ≥ 1, mn ≥ 1 for all integers n ≥ 1,

and mi 6= mj for all integers i 6= j.
First it is easy to check f (a0 , m0 ) = f (1, 1) = 0. Assuming that f (an−1 , mn−1 ) =
0, a routine, but tedious, calculation yields f (an , mn ) = 0. Second, a0 , m0 ≥
1 and assuming an−1 , mn−1 ≥ 1, it follows that an ≥ 11 + 16 − 5 ≥ 1 and

Copyright © Canadian Mathematical Society, 2019

404/ OLYMPIAD CORNER

mn ≥ 24 + 35 − 12 ≥ 1. Finally, the sequence of mn is strictly increasing since

mn ≥ 24 + 35mn−1 − 12 = 35mn−1 + 12 and the mn are positive. Therefore, the
mn sequence leads to infinitely many positive integers with the required property.
An alternative way to find the recursive solution (1), is to write the equation
f (a, m) = 0 in an equivalent form

3(m + 2a − 1)2 − 11m2 = 1.

The substitution x = m + 2a − 1 yields a Pell equation 3x2 − 11m2 = 1. For

an arbitrary Pell equation cx2 − dy 2 = 1, the Pell resolvent is defined to be
u2 − cdv 2 = 1. Therefore, the Pell resolvent for 3x2 − 11m2 = 1 is u2 − 33v 2 = 1
with fundamental solution (u1 , v1 ) = (23, 4). Let u0 = 1 and v0 = 0. The general
solution (un , vn ) for the Pell resolvent is recursively given for n ≥ 1 by

un+1 = u1 un + cdv1 vn = 23un + 132vn (2)

vn+1 = v1 un + u1 vn = 4un + 23vn .

Note that vn is always an even integer and un is odd for each integer n ≥ 0.
The general solution (xn , mn ) for 3x2 − 11m2 = 1 in terms of the solution of the
resolvent is given by

xn = x0 un + dm0 vn = 2un + 11vn (3)

mn = m0 un + cx0 vn = un + 6vn .

Observe that xn is an even integer and mn is an odd integer for any n ≥ 0. As

a result, for xn and mn defined by (2), the equation xn = mn + 2a − 1 admits
an integer solution a. Moreover, the recursive formulas (2) that define the un and
vn sequences can be used to derive the recursive formulas (1) for the an and mn
sequences.
We end by listing several examples.

n mn an Sum

0 1 1 12 = 13

1 47 22 222 + 232 + · · · + 682 = 473

| {z }
47 terms

2 2161 989 9892 + 9902 + · · · + 31492 = 21613

| {z }
2161 terms

3 99359 45450 454502 + 454512 + · · · + 1448082 = 993593

| {z }
99359 terms

4 4568353 2089689 20896892 + 20896902 + · · · + 66580412 = 45683533

| {z }
4568353 terms

5 210044879 96080222 960802222 + 960802232 + · · · + 3061251002 = 2100448793

| {z }
210044879 terms

Crux Mathematicorum, Vol. 45(7), September 2019

 OLYMPIAD CORNER /405

OC420. General Tilly and the Duke of Wallenstein play “Divide and rule!”
(Divide et impera!). To this end, they arrange N tin soldiers in M companies and
command them by turns. Both of them must give a command and execute it in
their turn.
Only two commands are possible: The command “Divide!” chooses one company
and divides it into two companies, where the commander is free to choose their
size, the only condition being that both companies must contain at least one tin
soldier. On the other hand, the command “Rule!” removes exactly one tin soldier
from each company.
The game is lost if in your turn you can’t give a command without losing a com-
pany. Wallenstein starts to command.
(a) Can he force Tilly to lose if they start with 7 companies of 7 tin soldiers each?
(b) Who loses if they start with M ≥ 1 companies consisting of n1 ≥ 1, n2 ≥ 1,
. . . , nM ≥ 1 (n1 + n2 + · · · + nM = N ) tin soldiers?
Originally 2017 Germany Math Olympiad, 3rd Problem, Final Round.
We received 1 submission. We present the solution by Jeremy Mirmina.
We discuss the winning strategy of the game based on the parities (odd/even)
of the number of tin soldiers N , the number of companies M , and the difference
I = N − M.
First, notice the following. When Move 1 (”Divide!”) is played N remains the
same, M decreases by one, and I decreases by one and switches parity. When Move
2 (”Rule!”) is played N decreases by M , M remains the same, and I decreases by
M.
In the next table we summarise the changes in the parities of I, N , and M after
Move 1 or Move 2 are played. We assume that the game did not end, and that
Move 1 and Move 2 can be played.
Move 1 Move 2
I N M new I new N new M new I new N new M
odd odd even even odd odd odd odd even
odd even odd even even even even odd odd
even even even odd even odd even even even
even odd odd odd odd even odd even odd
Second, notice that if there is at least one company with exactly one soldier the
two players can only use Move 1 and for exactly I times. This is because Move 1
will be applied ni − 1 times to split a company with ni soldiers into ni companies,
each with only one soldier.
Third, notice that the game ends when all companies have exactly one soldier.
Case 1. Assume I is odd, i.e. either N is even and M is odd or M is even and
N is odd. Then the first player (Wallenstein) has a winning strategy.

Copyright © Canadian Mathematical Society, 2019

406/ OLYMPIAD CORNER

Because N 6= M , he can choose a company with more than one soldier. He plays
Move 1, and splits this company into one soldier and the rest. At this point, none
of the two players can use Move 2, as it will make them lose. Move 1 is played
exactly I = N − M times until I = 0 and the game ends. The player who starts
with I odd will continue to play with I odd and his opponent will play with I
even. Eventually his opponent will receive the configuration with I = 0 and will
lose. For this reason a player who moves with an even I, never wants to change it
into odd on his opponent’s turn.

Case 2. Assume I is even, and both N and M are odd. Then the second player
(Tilly) has a winning strategy.
Based on the table above, the first player starts with an even I and regardless of
his move he changes the parity of I to odd on his opponent’s turn. Hence, the
second player has always a winning strategy (see Case 1).
This answers part (a) of the problem, since N = 7 × 7 = 49 is an odd number
of soldiers and M = 7 is an odd number of companies. Wallenstein cannot force
Tilly to loose, and Tilly, the second player, has a winning strategy.

Case 3. Assume I is even, and both N and M are even.

In this case, neither of the two players is interested in playing Move 1, which
results in a winning configuration for the opponent. They play Move 2 for as long
as they can. In fact, Move 2 can be played m − 1 times, where m = min{ni : i =
1, 2, . . . , M } is the size of the company with the lowest number of soldiers. If m
is even then the first player (Wallenstein) has a winning strategy, and if m is odd
then the second player (Tilly) has a winning strategy.
The conclusions of cases 1, 2, and 3 answer part (b).

Crux Mathematicorum, Vol. 45(7), September 2019

 Michel Bataille /407

FOCUS ON...
No. 37
Michel Bataille
Geometry with Complex Numbers (II)
Introduction
In this second part, we continue to present various interventions of the complex
numbers in geometry problems. We begin with regular polygons, an obvious do-
main of application. Then, we will consider similarities, either direct or opposite,
as they can be simply represented using complex numbers, and we conclude with
a look at areas.

Complex numbers and regular polygons

It is well-known that the n-th roots of a nonzero complex number are the af-
fixes of the vertices of a regular n-gon. In particular, the n-th roots of unity
exp(2kπi/n), k = 0, 1, . . . , n − 1, correspond to an n-gon inscribed in the unit cir-
cle Γ, with centre O and radius 1. Here is a first illustration, a problem proposed
in the December 2017 issue of Mathematics Magazine:
Let n be an integer, n ≥ 2. Let A1 A2 A3 · · · A2n+1 be a regular polygon
with 2n + 1 sides. Let P be the intersection of the segments A2 An+2
and A3 An+3 . Prove that

(A1 P )2 = (A2 A3 )2 + (A3 P )2 .

We may suppose thatÄ for kä= 1, 2, . . . , 2n + 1, Ak is the point with complex affix
2πi
wk−1 where w = exp 2n+1 . Let p be the affix of P . We readily obtain

(A2 A3 )2 = |w2 − w|2 = |w − 1|2 = (w − 1)(w − 1) = 2 − w − w = 2 − w − w2n .

and

(A1 P )2 − (A3 P )2 = |p − 1|2 − |p − w2 |2

= (p − 1)(p − 1) − (p − w2 )(p − w2n−1 )
= pw2n−1 − p + pw2 − p.

Now, from the equations of the lines A2 An+2 and A3 An+3 , we deduce that pwn +
pw = 1 + wn+1 and pwn−1 + pw2 = 1 + wn+1 . It follows that

pw2n−1 − p = wn (pwn−1 + pw2 ) − wn+1 (pwn + pw) = wn + 1 − (wn+1 + w)

1
pw2 − p = (pwn−1 + pw2 ) − (pwn + pw) = 1 + wn+1 − (w2n + wn )
w

Copyright © Canadian Mathematical Society, 2019

408/ Focus On... Geometry with Complex Numbers (II)

and by addition,

(A1 P )2 − (A3 P )2 = 2 − w − w2n = (A2 A3 )2 .

As a second example, we prove a vectorial result about the projections of a point

on the sidelines of a regular polygon:
Let the consecutive vertices of a regular n-gon be denoted A0 , . . . , An−1 ,
in order, and let An = A0 . Let Bk be the projection of a point M onto
the line Ak Ak+1 . Show that
n−1
X −−−→ n −−→
M Bk = M O.
2
k=0

Again, we suppose that the affix of Ak is wk with w = exp(2πi/n). Let m be the

affix of M and let Ck be the midpoint of Ak Ak+1 .
P −−−→
n−1 −−→
We remark that M Ck = nM O; for example because
k=0

n−1 Ç å n−1
!
X wk + wk+1 X
k 1 − wn
−m = w − nm = − nm = −nm
2 1−w
k=0 k=0

and −−→ −−−−−→

−−−→ (OM · Ak Ak+1 ) −−−−−→
Ck Bk = −−−−−→ Ak Ak+1 .
kAk Ak+1 k2
−−−→ −−−−−→
With the notations mk , δk = wk+1 − wk for the complex affixes of Ck Bk , Ak Ak+1 ,
respectively, the latter gives

δk 1 (mδk + mδk ) 1
mk = (Re(mδk )) = · = (m − mw2k+1 ).
δk δk 2 δk 2
Since n−1
X 1 − (wn )2
w2k+1 = w · = 0,
1 − w2
k=0
n−1 n
we obtain mk = · m, that is,
P
k=0 2
n−1
X −−−→ n −−→
Ck Bk = OM .
2
k=0

The desired result then follows from

n−1
X −−−→ n−1 X −−−→ n−1 X −−−→ −−→ n −−→
M Bk = M Ck + Ck Bk = nM O + OM .
2
k=0 k=0 k=0

(The established result will be used later, in the paragraph devoted to areas.)

Crux Mathematicorum, Vol. 45(7), September 2019

 Michel Bataille /409

Complex numbers and similarities

A similarity with factor k > 0 is a transformation of the plane such that the
distance between the images of M, N is kM N for any points M, N of the plane.
The similarities with factor 1 are the isometries while those of factor k 6= 1 are of
the form H ◦ I = I ◦ H where H is a homothety with factor k and centre Ω and I
is an isometry such that I(Ω) = Ω. The similarity is direct or opposite according
as I is a displacement or a reflection in a line (that is, according as it preserves
orientation or not). Embedding all this in the complex plane, it can be showed
that a direct (resp. opposite) similarity transforms a point M with affix m into the
point M 0 with affix m0 = am + b (resp. m0 = am + b) for some complex numbers
a 6= 0, b independent of m. To illustrate the power of this complex representation,
we consider two examples, the first one being adapted from an exercise set at the
French final high-school exam long ago:
Let OAB be a triangle and suppose that points P and Q are such that
the triangles OA0 B 0 , AB 0 P, BQA0 are directly similar to the triangles
OAB, ABO, BOA, respectively. Show that O is the midpoint of P Q.
Taking O as the origin and using the corresponding lower-case letter for the affix
of any other point, the hypotheses imply that for some complex numbers u, u1 , u2 ,
we have a0 = ua, b0 = ub, b0 = u1 b + p, a = u1 a + p, a0 = u2 a + q, b = u2 b + q (for
example, the latter because some direct similarity transforms A into A0 , B into B
and O into Q). The elimination of u1 and u2 leads to p(a − b) = ab(u − 1) and
q(b − a) = ab(u − 1). The desired relation p + q = 0 follows.
As a second example, we offer a variant of solution to problem 3401 [2009 : 42,
44 ; 2010 : 49]:
Let ABCDE be a convex pentagon such that ∠BAC = ∠EAD and
∠BCA = ∠EDA, and let the lines CB and DE intersect in the point
F . Prove that the midpoints of CD, BE, and AF are collinear.

As remarked in the featured geometric solution, what matters is the fact that the
triangles ABC and AED are oppositely similar. Therefore, taking the point A
as the origin, there exists a complex number ω such that e = ωb and d = ωc.
Let U, V, W be the midpoints of CD, BE, AF , respectively. Their affixes are u =
1 1 1
2 (c + ωc), v = 2 (b + ωb), and w = 2 · f.

Copyright © Canadian Mathematical Society, 2019

410/ Focus On... Geometry with Complex Numbers (II)

The equations of the lines BC and DE are readily obtained:

z(c − b) − z(c − b) = (bc − bc)

and
ωz(c − b) − ωz(c − b) = |ω|2 (bc − bc).
This said, U, V, W are collinear if and only if

(u − w)(v − w) = (u − w)(v − w)

or
(c + ωc − f )(b + ωb − f ) = (c + ωc − f )(b + ωb − f ).
Expanding and arranging, this condition can be written as

f (b − c) − f (b − c) + bc − bc = ωf (b − c) − ωf (b − c) + |ω|2 (bc − bc).

This certainly holds since both sides are 0 (the left-hand side because F is on BC
and the right one because F is on DE). The conclusion follows.

Complex numbers and area

We start with the following known expression of the area [ABCD] of a quadrilateral
ABCD:
1 −→ −−→
[ABCD] = ± AC · BD · sin(∠(AC, BD))
2
where the sign is + if and only if the quadrilateral is positively oriented, and
−→ −−→ −→ −−→
∠(AC, BD) is the directed angle from AC to BD. If a, b, c, d are the affixes of the
vertices, we obtain the following formula:
1
[ABCD] = ± Im((d − b)(c − a)),
2
of which, for convenience, we repeat the proof.
−→ −−→
Let α = arg(d − b) and β = arg(c − a). Then, sin(∠(AC, BD)) = sin(α − β), hence
1
[ABCD] = ± |c − a| · |d − b| sin(α − β)
2
1
= ± |c − a| · |d − b|Im(ei(α−β) )
2
1
= ± Im(|d − b|eiα · |c − a|e−iβ ),
2
that is,
1
[ABCD] = ± Im ((d − b)(c − a)) .
2
Note that taking d = c, we obtain a formula for the area of ∆ABC:
1
[ABC] = ± Im ((c − b)(c − a)) .
2

Crux Mathematicorum, Vol. 45(7), September 2019

 Michel Bataille /411

We will see these results at work in two examples. First, we again consider the
projections Bk of a point M on the sides of a regular n-gon A0 A1 . . . An−1 (with
A0 = An ) and we suppose that the point Bk lies on the segment Ak Ak+1 for
k = 0, 1, . . . , n − 1 (see the second problem of the second paragraph). We will
show that S1 = S2 where

n−1
X n−1
X
S1 = [M Ak Bk ] and S2 = [M Ak+1 Bk ].
k=0 k=0

We use the notations used earlier and denote by bk and ck the affixes of Bk
and Ck . Observing that ∆M Ak Bk and ∆M Ak+1 Bk have opposite orientations,
proving that S1 − S2 = 0 amounts to proving that

n−1
X
Im[(m − bk )(wk − bk ) + (m − bk )(wk+1 − bk )] = 0.
k=0

Now, we have

(m − bk )(wk − bk ) + (m − bk )(wk+1 − bk ) = 2(m − bk )(ck − bk )

= 2(m − bk )(ck − m) + 2|m − bk |2

and we remark that (m − bk )ck is a real number (since OCk is parallel to M Bk ).

It follows that
n−1 n−1
!
X X
Im[(m − bk )(wk − bk ) + (m − bk )(wk+1 − bk )] = −2 · Im m (m − bk ) .
k=0 k=0

n−1
n P −−−→
n−1
n −−→
But we have (m − bk ) = · m (from M Bk = proved earlier) so that
P
2 2 MO
k=0 k=0

n−1
X
−2 · Im(m (m − bk )) = −nIm(mm) = 0
k=0

and consequently S1 − S2 = 0.

Our second example is adapted from a problem set in the Mathematical Gazette
in 2017:

Squares are described externally on the sides of a convex quadrilateral

ABCD. Prove that the line segments joining the centres of opposite
squares
» are perpendicular and that the length of each line segment
is 2S + 12 (x2 + y 2 ), where S is the area of ABCD and x = AC,
y = BD.

Copyright © Canadian Mathematical Society, 2019

412/ Focus On... Geometry with Complex Numbers (II)

Assuming that ABCD is clockwise oriented and with obvious notations, we have
b−m = i(a−m), hence (1−i)m = b−ia. Similarly, (1−i)n = c−ib, (1−i)p = d−ic
and (1 − i)q = a − id. It follows that q − n = −i(p − m) so that N Q = P M and
N Q ⊥ P M . Furthermore,

2M P 2 = |(1 − i)(p − m)|2 = [d − b − i(c − a)] · [d − b + i(c − a)] = α + iβ

where α = |d − b|2 + |c − a|2 = x2 + y 2 and

β = (d − b)(c − a) − (d − b)(c − a) = 2iIm[((d − b)(c − a)] = −4i[ABCD].

Finally, 2M P 2 = x2 + y 2 + 4S, as desired.

As usual, we end the number with a couple of exercises.

Exercises
1. Let C be a point distinct from the vertices of a triangle OAB. Suppose
that ∆OCD and ∆CAE are directly similar to ∆OAB. Prove that CDBE is
a parallelogram.
2. Use complex numbers to solve problem 3898: On the extension of the side AB
of the regular pentagon ABCDE, let the points F and G be placed in the order
F, A, B, G so that AG = BF = AC. Compare the area of triangle F GD to the
area of pentagon ABCDE.

Crux Mathematicorum, Vol. 45(7), September 2019

 Problems /413

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 October 30, 2019.

4461. Proposed by Marian Dinca, Leonard Giugiuc and Daniel Sitaru.

w−u
Let u, v and w be distinct complex numbers such that is not a real number.
v−u
Consider a complex number z = αu + βv + γw, where α, β, γ > 0 are real numbers
such that α + β + γ = 1. Prove that
(|z − v| + |w − u|)2 + (|z − w| + |u − v|)2 > (|z − u| + |v − w|)2 .

4462. Proposed by George Apostolopoulos.

Let a, b, c be the lengths of the sides of triangle ABC with inradius r and circum-
radius R. Show that
√
a2 b2 c2 3 6R »
+ + ≤ R(R − r).
b+c a+c a+b 4r

4463. Proposed by Max A. Alekseyev

For all integers n > m ≥ 0, prove that
n Ç å
k 2n + 1
X
(−1) (2k + 1)2m+1 = 0.
n−k
k=0

4464. Proposed by Borislav Mirchev and Leonard Giugiuc.

Let ABC be a triangle with external angle bisectors k, l and m to angles A, B and
C, respectively. Projections of A on l and m are L and P , respectively. Similarly,
projections of B on m and k are N and K and projections of C on k and l are Q
and M . Show that the points M, N, P, Q, K and L are concyclic.

4465. Proposed by Nguyen Viet Hung.

Let ABC be a triangle with centroid G and medians ma , mb , mc . Rays AG,BG,CG
intersect the circumcircle at A1 ,B1 ,C1 respectively. Prove that
Area[A1 B1 C1 ] (a2 + b2 + c2 )3
= .
Area[ABC] (8ma mb mc )2

Copyright © Canadian Mathematical Society, 2019

414/ Problems

4466. Proposed by Arsalan Wares.

Let A be a regular hexagon with vertices Ak , k = 1, 2, . . . , 6. There are two
congruent overlapping squares inside A. Each of the squares shares one vertex
with A and two vertices of each square lie on opposite sides of hexagon A as in
the figure:

Find the exact area of the shaded region, if the length of each side of hexagon A
is 2.

4467. Proposed by Paul Bracken.

Show that for x > 0,
1 x
arctan x · arctan > 2
.
x 2(x + 1)
(Ed.: Take a look at the problem 4327.)

4468. Proposed by Florin Stanescu.

Let f : [0, 1] → R be a differentiable function such that f 0 is continuous and
f (0) + f 0 (0) = f (1). Show that there exists c ∈ (0, 1) such that
Z c
c
f (c) = f (x)dx.
2 0

4469. Proposed by Leonard Giugiuc and Dan-Stefan Marinescu.

Let ABC be a triangle and let P be an interior point of ABC. Denote by Ra ,
Rb , Rc the circumradii of the triangles P BC, P CA and P AB, respectively. Prove
that Ra Rb Rc ≥ P A · P B · P C.

4470. Proposed by Leonard Giugiuc and Diana Trailescu.

Let a, b and c be three distinct complex numbers such that |a| = |b| = |c| = 1 and
|a + b + c| ≤ 1. Prove that |a2 + bc| ≥ |b + c|.

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

Crux Mathematicorum, Vol. 45(7), September 2019

 Problems /415

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

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

Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au plus tard le 30 octobre 2019.
La rédaction souhaite remercier Rolland Gaudet, professeur titulaire à la retraite à
l’Université de Saint-Boniface, d’avoir traduit les problèmes.

4461. Proposé par Marian Dinca, Leonard Giugiuc et Daniel Sitaru.

w−u
Soient u, v et w des nombres complexes distincts tels que n’est pas un
v−u
nombre réel. Considérons alors un nombre complexe z = αu + βv + γw, où
α, β, γ > 0 sont des nombres réels tels que α + β + γ = 1. Démontrer que
(|z − v| + |w − u|)2 + (|z − w| + |u − v|)2 > (|z − u| + |v − w|)2 .

4462. Proposé par George Apostolopoulos.

Soient a, b, c les longueurs des côtés du triangle ABC, où r est le rayon du cercle
inscrit et R est le rayon du cercle circonscrit. Démontrer que
√
a2 b2 c2 3 6R »
+ + ≤ R(R − r).
b+c a+c a+b 4r

4463. Proposé par Max A. Alekseyev

Pour entiers n > m ≥ 0, démontrer que
n Ç å
k 2n + 1
X
(−1) (2k + 1)2m+1 = 0.
n−k
k=0

4464. Proposé par Borislav Mirchev et Leonard Giugiuc.

Soit ABC un triangle dont les bissectrices externes des angles A, B et C sont k, l et
m respectivement. Les projections de A vers l et m sont L et P , respectivement. De
façon similaire, les projections de B vers m et k sont N et K respectivement, et les
projections de C vers k et l sont Q et M . Démontrer que les points M, N, P, Q, K
et L sont cocycliques.

4465. Proposé par Nguyen Viet Hung.

Soit ABC un triangle avec centroı̈de G et médianes ma , mb , mc . Les rayons AG,
BG, CG intersectent le cercle circonscrit en A1 ,B1 ,C1 respectivement. Démontrer
que
Area[A1 B1 C1 ] (a2 + b2 + c2 )3
= .
Area[ABC] (8ma mb mc )2

Copyright © Canadian Mathematical Society, 2019

416/ Problems

4466. Proposé par Arsalan Wares.

Soit A un hexagone régulier de sommets Ak , k = 1, 2, . . . , 6. Deux carrés congrus
se chevauchent à l’intérieur de A. Chacun des carrés partage un sommet avec
A et deux sommets de chaque carré se trouvent sur des côtés opposés de A, tel
qu’indiqué ci-bas:

Déterminer la valeur exacte de la surface colorée, si les côtés de A sont de longueur

2.

4467. Proposé par Paul Bracken.

Démontrer que pour x > 0
1 x
arctan x · arctan > 2
.
x 2(x + 1)
(Note: Voir le probléme 4327.)

4468. Proposé par Florin Stanescu.

Soit f : [0, 1] → R une fonction différentiable telle que f 0 est continue et que
f (0) + f 0 (0) = f (1). Démontrer qu’il existe c ∈ (0, 1) tel que
Z c
c
f (c) = f (x)dx.
2 0

4469. Proposé par Leonard Giugiuc et Dan-Stefan Marinescu.

Soit ABC un triangle et soit P un point dans son intérieur. Dénotons par Ra , Rb ,
Rc les rayons des cercles circonscrits des triangles P BC, P CA, P AB, respective-
ment. Démontrer que Ra Rb Rc ≥ P A · P B · P C.

4470. Proposé par Leonard Giugiuc et Diana Trailescu.

Soient a, b et c trois nombres complexes distincts tels que |a| = |b| = |c| = 1 et
|a + b + c| ≤ 1. Démontrer que |a2 + bc| ≥ |b + c|.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /417

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 2019: 45(1), p. 33–37;
45(2), p. 85–89.

4401. Proposed by Ruben Dario and Leonard Giugiuc.

Let D and E be the centres of squares erected externally on the sides AB and
AC, respectively, of an arbitrary triangle ABC, and define F and G to be the
intersections of the line BC with lines perpendicular to ED at D and at E. Prove
that the resulting segments BF and CG are congruent.
We received 11 solutions. We present the solution by Madhav Modak.

Let H be the point on DE such that DH = DF . Since D is the centre of the

square on side AB we have DB ⊥ DA and DB = DA. Further, we are given that
DF ⊥ DH, and so ∠ADH = 90◦ − ∠HDB = ∠BDF . It follows that 4ADH
and 4BDF are congruent. Hence F B = HA. Moreover,

∠AHE = 180◦ − ∠AHD = 180◦ − ∠BF D,

and using the fact that DF ||EG we conclude that ∠AHE = ∠ECG. As before,
since AE ⊥ EC and HE ⊥ EG, we get that ∠AEH = ∠CEG. Using the fact
that AE = EC, we get that 4AEH and 4CEG are congruent, so AH = CG.
Therefore, F B = CG, as desired.

Copyright © Canadian Mathematical Society, 2019

418/ Solutions

4402. Proposed by Peter Y. Woo.

Consider a rectangular carpet ABCD lying on top of floor tiled with 8 square tiles
with side length of 1 foot each (as shown in the diagram).

Suppose AH bisects ∠BAC. Express tan ∠BAH as the sum of a rational number
and the square root of a rational number.
We received 15 submissions, all correct. We present the solution by Jirapat Kaewkam,
enhanced by the editor.
Let CX be perpendicular to the horizontal line l extended from AH with X being
on l. Connect CX (see figure). Since ∠ABC = ∠AXC = 90◦ , we see that A,
B, X, C are concyclic so ∠XCH = ∠BAX = ∠XAC. Hence ∆XCH ∼ ∆XAC
from which it follows that
AX CX
= or (AX)(HX) = (CX)2 = 1,
CX HX
so (HX + 3)(HX) = 1. Solving

(HX)2 + 3(HX) − 1 = 0,

we then obtain
√ …
HX −3 + 13 3 13
tan(∠BAH) = tan(∠XCH) = = HX = =− + .
CX 2 2 4

4403. Proposed by Michel Bataille.

Let m be an integer with m > 1. Evaluate in closed form
n Ç å
k−1 n + 1 k
X
(−1) .
k+1 m+k
k=1

We received 8 submissions, all of which were correct and complete. We present

two solutions.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /419

Solution 1, by the proposer.

Let
n Ç å
X
k−1 n+1 k
Sn = (−1) .
k+1 m+k
k=1

We show that Ç å
1 n+1
Sn = 1 − m+n
.
m−1 m
Using the well-known identity
n Ç å Ç å
X j n+1
=
k k+1
j=k

and changing the order of summation, we obtain

j
n X Ç å
X
k−1 k j
Sn = (−1) .
j=1 k=1
m+k k

j j−1



Taking the relation k k =j k−1 into account yields

n j Ç å
X X 1 j−1
Sn = j (−1)k−1 . (1)
j=1
m+k k−1
k=1

Now we use the closed form

j Ç å
X
k−1 1 j−1 (j − 1)!
(−1) =
m+k k−1 (m + 1)(m + 2) · · · (m + j)
k=1

1
which readily follows from the decomposition of (m+1)(m+2)···(m+j) into partial
fractions. Back to (1), this leads to
n n
X j! X 1
Sn = = m! . (2)
j=1
(m + 1)(m + 2) · · · (m + j) j=1
(j + 1)(j + 2) · · · (j + m)

But we have
1
(j + 1)(j + 2) · · · (j + m)
1 1 1
Å ã
= − .
m − 1 (j + 1)(j + 2) · · · (j + m − 1) (j + 2)(j + 3) · · · (j + m)

So that the last sum in (2) is telescopic and therefore

Ç å
1 1 1 n+1
Å ã
m!
Sn = − = 1− m+n .
m−1 m! (n + 2) · · · (n + m) m−1


m

Copyright © Canadian Mathematical Society, 2019

420/ Solutions

Solution 2, by Paul Bracken, Brian Bradie, Madhav Modak, CR Pranesachar, and

Daniel Văcaru, all done independently.
We have
n å Ç
X n+1k−1 k
(−1)
k+1 m+k
k=1
n Ç å n Ç å
k−1 n + 1 k−1 n + 1 1
X X
= (−1) −m (−1) . (3)
k+1 k+1 m+k
k=1 k=1

Using binomial expansion,

n+1 Ç å n Ç å
n+1
X n+1 k
X
k+1 n + 1
0 = (1 − 1) = (−1) = (−1)
k k+1
k=0 k=−1
n Ç å
k−1 n + 1
X
= −n + (−1) . (4)
k+1
k=1

Again by binomial expansion

n+1 Ç
X n+1
å n Ç å
n+1 k k
X n+1
(1 − x) = (−1) x = 1 − (n + 1)x + (−1)k−1 xk+1 .
k k+1
k=0 k=1

It follows that
n Ç å
m−2 n+1
X n+1
x ((1 − x) − 1 + (n + 1)x) = (−1)k−1 xm+k−1 .
k+1
k=1

Since m > 1, we have

n Ç å Z 1X n Ç å
k−1 n + 1 1 n+1
X
(−1) = (−1)k−1 xm+k−1 dx
k+1 m+k 0 k=1 k + 1
k=1
Z 1
= xm−2 ((1 − x)n+1 − 1 + (n + 1)x) dx.
0
1
(m − 2)!(n + 1)!
Z
The integral xm−2 (1−x)n+1 dx is a beta function with the value .
0 (m + n)!
Therefore the above integral has the value
(m − 2)!(n + 1)! n + 1 1
+ − .
(m + n)! m m−1
Combining this with the result in (4) into (3) we have
n Ç å
k−1 n + 1 (m − 2)!(n + 1)! n + 1 1
Å ã
X k
(−1) =n−m + −
k+1 m+k (m + n)! m m−1
k=1
Ç å
1 n+1
= 1 − m+n
.
m−1 m

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /421

4404. Proposed by Nguyen Viet Hung.

Let x, y and z be integers such that x > 0, z > 0 and x + y > 0. Find all the
solutions to the equation

x4 + y 4 + (x + y)4 = 2(z 2 + 40).

We received 11 submissions, 9 of which were correct and complete. We present

the solution by Brian D. Beasley.
The given equation is equivalent to

(x2 + xy + y 2 )2 = z 2 + 40.

Letting
n = x2 + xy + y 2 ,
we note that n > 0 and
(n + z)(n − z) = 40.
Since z > 0, this implies

(n + z, n − z) = (40, 1), (20, 2), (10, 4), or (8, 5).

But
(n + z) + (n − z) = 2n
must be even, so 2n = 22 or 2n = 14, and hence (n, z) = (11, 9) or (n, z) = (7, 3).
If n = x(x + y) + y 2 = 11 with x > 0 and x + y > 0, then y ∈ {0, ±1, ±2, ±3}. But
none of these values for y will yield an integer value for x.
If n = x(x + y) + y 2 = 7 with x > 0 and x + y > 0, then y ∈ {0, ±1, ±2}. Four
of these five values for y yield a positive integer value for x. Thus there are four
solutions for (x, y, z) to the original equation:

(1, 2, 3), (2, 1, 3), (3, −1, 3), (3, −2, 3).

4405. Proposed by Kadir Altintas and Leonard Giugiuc.

Let ABC be a triangle and let K be a point inside ABC. Suppose that BK
intersects AC in F and CK intersects AB in E. Let M be the midpoint of BE,
N be the midpoint of CF and suppose that M N intersects BK at P . Show that
the midpoints of AF, EK and M P are collinear.
All 7 of the submissions we received were correct, but two of them relied on a
computer. We present the solution by Andrea Fanchini.
We use barycentric coordinates with respect to triangle ABC. Working backwards,
for the points E and F to be

E = CK ∩ AB = (m, 1 − m, 0), F = BK ∩ AC = (1 − n, 0, n)

Copyright © Canadian Mathematical Society, 2019

422/ Solutions

(where m, n are parameters with 0 < m, n < 1), we must have

BK : nx − (1 − n)z = 0, CK : (1 − m)x − my = 0,

and, finally,
K (m(1 − n) : (1 − m)(1 − n) : mn) .
The midpoints of BE and of CF are then

M = (m : 2 − m : 0), N = (1 − n : 0 : 1 + n).

The line M N : (2 − m)(1 + n)x − m(1 + n)y − (1 − n)(2 − m)z = 0 intersects BK

at P , so that

P = m(1 − n2 ) : (2 − m)(1 − n) : mn(1 + n) .

Finally, the midpoints of AF, EK and M P are

MAF = (2 − n : 0 : n),

MEK = (m(mn − 2n + 2) : (1 − m)(mn − 2n + 2) : mn),

and

MM P = (m(mn − n2 − n + 2) : (2 − m)(mn − 2n + 2) : mn(1 + n)).

Because
 

 2−n 0 n 


 m(mn − 2n + 2) (1 − m)(mn − 2n + 2) mn  = 0,

 m(mn − n2 − n + 2) (2 − m)(mn − 2n + 2) mn(1 + n) 

it follows that these midpoints are collinear.

Editor’s comments. Note that there is no need to restrict K to the interior of the
triangle: it could any point in the plane except B or C. In other words, we can
allow m and n to be any real numbers except m 6= 0 and n 6= 1, and the result
continues to hold.
It is interesting to compare our problem with an extended version of Hjelmslev’s
theorem:
When all the points P on one line are related by a similarity to all the
points P 0 on a second line, then the points X dividing the segments
P P 0 in a fixed ratio P X : XP 0 are distinct and collinear or else they
all coincide.
See, for example, F. G.-M., Exercices de géométrie, 4th ed., Theorem 1146d, page
473. Does any reader see an easy direct proof that there exists a similarity that
takes the points A, E, M to the points F, K, P ?

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /423

4406. Proposed by Bill Sands.

Four trees are situated at the corners of a rectangle. You are standing outside
the rectangle, the nearest point of the rectangle being the midpoint of one of its
sides, 1 metre away from you. To you in this position, the four trees appear to be
equally spaced apart.
a) Find the side lengths of the rectangle, assuming that they are positive inte-
gers.
b) Suppose that the rectangle is a square. Find the length of its side.

B C
A

A

A

Q
A A
D
@ A

@ A 1

@A

@A

P

We received 8 correct solutions. We present two different approaches.

Solution 1, by Roy Barbara and C.R. Pranesachar (independently).
Let θ = ∠QP C, so that

2θ = ∠AP B = ∠BP C = ∠CP D

and 3θ = ∠QP D. Let the length of BC be u and the length of CD be v − 1. Then

tan θ = u/(2v) and

u 3 tan θ − tan3 θ 12uv 2 − u3

= tan 3θ = 2 = .
2 1 − 3 tan θ 2(4v 3 − 3u2 v)

Hence 4v 3 − 3u2 v = 12v 2 − u2 , whereupon u2 (3v − 1) = 4v 2 (v − 3).

(a) Let u and v be positive integers. Since 3v − 1 and v 2 are coprime, 3v − 1 must
divide 4(v − 3). Because

2(3v − 1) − 4(v − 3) = 2v + 10 > 0,

we must have 3v − 1 = 4(v − 3) and v = 11. Hence u = v = 11 and the respective

lengths of CD and BC are 10 and 11.
An alternative argument begins by rewriting the foregoing equation as
32
27u2 = 36v 2 − 96v − 32 − .
3v − 1
The right side is positive and 0 6= 4v 2 (v − 3), so that v exceeds 3. Since 3v − 1
divides 32, the only possibility is v = 11.

Copyright © Canadian Mathematical Society, 2019

424/ Solutions

(b) If ABCD is square, then v = u + 1 and so

0 = 4(u + 1)2 (u − 2) − u2 (3u + 2)

= u3 − 2u2 − 12u − 8
= (u + 2)(u2 − 4u − 4).
√
Hence the sidelength of the square is u = 2(1 + 2).

Solution 2, by Daniel Vacaru.

Let the respective lengths of AP , AB and AD be s, t, u. Let θ = ∠BP Q. Since

P QkAB, ∠ABP = θ. Also ∠AP B = 2θ and ∠AP Q = 3θ. By the Law of Sines,
sin 2θ/t = sin θ/s, whereupon cos θ = t/(2s).

Since
1 t3 − 3ts2
= cos 3θ = 4 cos3 θ − 3 cos θ = ,
s 2s3
then (3t + 2)s2 = t3 . Setting s2 = 1 + (u2 /4) yields that

(3t + 2)(4 + u2 ) = 4t3 =⇒ (3t + 2)u2 = 4t3 − 12t − 8.

(a) Let t and u be positive integers. Rewrite the equation as

32
27u2 = 36t2 − 24t − 92 − .
3t + 2

Since 3t+2 divides 32, either t = 2 or t = 10. The first option leads to (t, u) = (2, 0)
which is inadmissible, and the second to (t, u) = (10, 11).

(b) When u = t, the equation becomes

0 = t3 − 2t2 − 12t − 8 = (t + 2)(t2 − 4t − 4),

√
and the value 2(1 + 2) for the sidelength of the square.

4407. Proposed by Mihaela Berindeanu.

Circle C1 lies outside circle C2 and is tangent to it at E. Take arbitrary points B

and D different from E on the common tangent line. Let the second tangent from
B to C1 touch it at M and to C2 touch it at N , while the second tangents from
D to those circles touch them at Q and P , respectively. If the orthocenters of the
−−−→ −−→
triangles M N Q and P N Q are H1 and H2 , prove that H1 H2 = M P .

We received 6 submissions of which 5 were correct. We present the solution by

Oliver Geupel.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /425

Let point M 0 be the antipode of M on C1 . We use directed angles (]) modulo

180◦ . In ∆QM 0 M and its circumcircle we have

]QM M 0 = ]M QM 0 + ]QM 0 M = 90◦ + ]QEM.

Since the points E, M , and N all lie on a common circle centered at B, we have
that ]M EN = 21 ]M BN , so

]BM N = 90◦ + ]M EN.

Hence,

]QM N = ]QM M 0 + ]M 0 M B + ]BM N

= 90◦ + ]QEM + 90◦ + 90◦ + ]M EN
= 90◦ + ]QEN.

Similarly (replacing Q, M, N, B by N, P, Q, D, respectively),

]QP N = 90◦ + ]QEN.

Thus, ]QM N = ]QP N , which implies that the points M , N P , and Q all lie on
a common circle, say, C with center O.
Let O0 be the midpoint of the segment N Q. Let point Q0 be the antipode of Q
on C. The segment O0 O joins the midpoints of two sides of the right triangle
−−→ −−→
N QQ0 . Hence 2O0 O = N Q0 . On the other hand, the lines M Q0 and H1 N are
both perpendicular to M Q, and the lines Q0 N and M H1 are both perpendicular
to N Q. Hence, the quadrilateral H1 M Q0 N is a parallelogram, from which we
−−→ −−−→ −−−→ −−→ −−→ −−→
deduce N Q0 = H1 M . It follows that H1 M = 2O0 O. Analogously, H2 P = 2O0 O.
We conclude that
−−−→ −−→
H1 M = H2 P .

Copyright © Canadian Mathematical Society, 2019

426/ Solutions

Consequently, the quadrilateral H1 M P H2 is a parallelogram, which proves the

desired result.
Editor’s comments. The person who submitted the faulty solution misread the
problem, labeling the figure so that N and Q are on C2 . Interestingly, the result
−−−→ −−→
H1 H2 = M P continues to hold in the modified problem, but because the segment
M P is now a chord of C1 , the proof becomes somewhat easier.

4408. Proposed by Leonard Giugiuc, Dan Stefan Marinescu and Daniel Sitaru.
Let α ∈ (0, 1] ∪ [2, ∞) be a real number and let a, b and c be non-negative real
numbers with a + b + c = 1. Prove that

aα + bα + cα + 1 ≥ (a + b)α + (b + c)α + (c + a)α .

We received 4 submissions. One of the submitted solutions was incomplete. We

present the proof by Ioannis D. Sfikas.
We give a proof based on the following proposition by Leonard Giugiuc. (See
Hlawka’s Inequalities for a class of functions, Romanian Mathematical Magazine,
2016, by Daniel Sitaru and Leonard Giugiuc.)
Proposition: Let f (x) : [0, ∞) → ∞ be a differentiable function such that f (0) = 0
and f 0 (x) is convex. Then for all nonnegative x, y, z ∈ R,

f (x) + f (y) + f (z) + f (x + y + z) ≥ f (x + y) + f (y + z) + f (z + x).

Proof of the current problem: Define f (x) : [0, ∞) → R by f (x) = xα . Then

f (0) = 0 and f 000 (x) = α(α − 1)(α − 2)xα−3 ≥ 0 since (α − 1)(α − 2) ≥ 0 for
α ∈ (0, 1] ∪ [2, ∞).
Hence f 0 (x) is convex. We then have, by the Proposition, that

f (a) + f (b) + f (c) + f (a + b + c) ≥ f (a + b) + f (b + c) + f (c + a)

so
aα + bα + cα + 1 ≥ (a + b)α + (b + c)α + (c + a)α
follows.

Editor’s comments: It is easy to find a counterexample to show that the proposed

inequality needs not be true if α ∈ (1, 2); e.g., if α = 23 , a = b = 25 , c = 15 , then

cα + bα + cα + 1 ≈ 2(0.4)1.5 + (0.2)1.5 + 1 = 1.5954 . . .

while

(a + b)α + (b + c)α + (c + a)α ≈ (0.8)1.5 + 2(0.6)1.5 = 1.6451 . . . ,

so LHS < RHS.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /427

4409. Proposed by Christian Chiser.

Let A and B be two matrices in M2 (R) such that A2 = O2 and B is invertible.
Prove that the polynomial P = det(xB 2 − AB + BA) has all integer roots.
We received 6 submissions of which 4 were correct and complete. As stated this
problem is false and 4 counterexamples were provided. The solution by Ivko Dim-
itrić featured here provides a general solution for the roots and when such roots
are integer.
The statement is shown to be false by the following counter-example with
ï ò ï ò
1 −1 1 0
A= and B = .
1 −1 0 2

For these matrices, A2 = O2 ,

ï ò
2 x 1
xB − AB + BA =
1 4x

and the roots of P (x) = 4x2 − 1 are non-integers, x = ± 21 .

Nevertheless, it is possible to determine the roots of P in general and examine
when they will be integers.
ï ò
a b
Let A = . From
c d

a2 + bc b(a + d)
ï ò ï ò
0 0
A2 = 2 =
c(a + d) bc + d 0 0

we see that if a + d 6= 0 then b = c = 0, which immediately yields also a = d = 0

and a + d = 0, a contradiction! Thus, tr A = a + d = 0, so d = −a and a2 + bc = 0.
ï ò ï ò
p q a b
Let B = and A = with a2 = −bc. Then we compute
r s c −a
ï òï ò ï ò òï
p q a b a
q b p
−AB + BA = −
r s c −a sc −a r
ï ò
cq − br b(p − s) − 2aq
= .
2ar − c(p − s) br − cq

Further,
ï òï ò ï 2 ò
2 p q p q (p + qr)x q(p + s)x
xB = x =
r s r s r(p + s)x (s2 + qr)x
Hence,

(p2 + qr)x − (br − cq)

ï ò
2 q(p + s)x + b(p − s) − 2aq
xB − AB + BA =
r(p + s)x − c(p − s) + 2ar (s2 + qr)x + (br − cq)

Copyright © Canadian Mathematical Society, 2019

428/ Solutions

and

P (x) = [(p2 + qr)x − (br − cq)] [(s2 + qr)x + (br − cq)]

− [q(p + s)x + b(p − s) − 2aq] [r(p + s)x − c(p − s) + 2ar].

The coefficient of x2 in this quadratic trinomial is

(p2 + qr)(s2 + qr) − qr(p + s)2 = (ps − qr)2 = (det B)2

and the coefficient of x is reduced to

(p2 + qr)(br − cq) − (s2 + qr)(br − cq)

+ q(p + s)[c(p − s) − 2ar] + r(p + s)[2aq − b(p − s)]
= (br − cq)(p2 − s2 ) + (p + s)[cq(p − s) − br(p − s)]
= (br − cq)(p2 − s2 ) − (p + s)(p − s)(br − cq)
= 0.

Finally, the constant term of P (x) equals

−(br − cq)2 + [b(p − s) − 2aq] [c(p − s) − 2ar]

= −(br − cq)2 + bc(p − s)2 − 2a(br + cq)(p − s) + 4a2 qr
= −(br − cq)2 + (br + cq)2 − [a(p − s) + (br + cq)]2 + 4a2 qr
= −[a(p − s) + (br + cq)]2 + 4bcqr + 4a2 qr
= −[a(p − s) + (br + cq)]2 ,

since bc = −a2 . Hence,

P (x) = (ps − qr)2 x2 − [a(p − s) + (br + cq)]2

has roots
a(p − s) + (br + cq)
x=± .
ps − qr
The roots are integer if and only if the quotient on the right hand side is an integer,
in particular, when A, B ∈ M2 (Z) and A2 = O2 , det B = ±1, but in general the
roots are non-integers.

4410. Proposed by Daniel Sitaru.

Prove that π
Z 4 √ √ π
sin 2xdx < 2− .
0 4

We received 6 correct solutions. There were 10 additional solutions that can be

considered weakly correct in that either they obtained a different upper bound for
the integral and then gave a numerical argument that this did not exceed the desired

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /429

bound, or based estimates on a series expansion. There was one incorrect solution.
We present two solutions following different approaches.

Solution 1, by Michel Bataille and Àngel Plaza (independently).

The substitution u = (π/4) − x leads to
Z π/4 √ Z π/4 √
sin 2x dx = cos 2u du.
0 0

From the Cauchy-Schwarz Inequality,

√ √
1 + cos 2x < 2(1 + cos 2x)1/2 = 2 cos x.

Therefore

π
Z π/4 √ Z π/4 √ Z π/4 √
+ sin 2x dx = (1 + cos 2x) dx < 2 cos xdx = 2.
4 0 0 0

The result follows.

Solution 2, Brian Bradie and Daniel Vicaru (independently).

By the Root-Mean-Square (or the Jensen) Inequality,
√ …
1+ sin 2x 1 + sin 2x cos x + sin x  π
< = √ = sin x + .
2 2 2 4

Hence

π
Z π/4 √ Z π/4 √ h  π iπ/4 √
+ sin 2x dx = (1 + sin 2x) dx = 2 − cos x + = 2,
4 0 0 4 0

from which the result follows.

4411. Proposed by Michel Bataille.

Let n be a positive integer. Find the largest constant Cn such that

(xy)n (yz)n (zx)n n−1

n+1
+ n+1 + n+1 ≥ Cn (max(x, y, z))
z x y
1 1 1
holds for all real numbers x, y, z satisfying xyz > 0 and x + y + z = 0.
We received 4 solutions, 3 of which were correct. We present the solution by
Walther Janous.
The two conditions xyz > 0 and x1 + y1 + z1 = 0 imply that two of the three variables
x, y, and z have to be negative. As the inequality is symmetric, we let x > 0.

Copyright © Canadian Mathematical Society, 2019

430/ Solutions

Then, say, y = −s, and z = −t, with s and t positive. But then max{x, y, z} = x,
and x1 − 1s − 1t = 0 leads to
st
x= .
s+t

The inequality under consideration is successively equivalent to

(−sx)n (st)n (−tx)n

n+1
+ n+1 + ≥ Cn · xn−1
(−t) x (−s)n+1

(st)n
Å n
tn
ã
s
− n+1 + n+1 · xn ≥ Cn · xn−1
xn+1 t s
ãn−1
(s + t)n+1 s2n+1 + t2n+1
Å
st
− ≥ Cn ·
st (st) · (s + t)n s+t

(s + t)2n s2n+1 + t2n+1

− ≥ Cn .
(st)n (st)n · (s + t)

As the left-hand expression is homogeneous of degree 0, we may and do let t = 1,

resulting in the inequality

(t + 1)2n t2n+1 + 1
n
− n ≥ Cn
t t · (t + 1)

for t > 0; that is, upon expanding, we have successively

2n Ç å 2n
X 2n X
· tj − tj
j=0
j j=0
≥ Cn ,
tn
2n−1 ÇÇ å å
X 2n
− 1 · tj
j=1
j
≥ Cn ,
tn
n−1 ÇÇ å å Å ã ÇÇ å å
X 2n 1 2n
− 1 · tn−j + n−j + − 1 ≥ Cn .
j=1
j t n

Since w + 1/w ≥ 2 for all w > 0, the left-hand sum attains its least value for t = 1.
Therefore, the best constant Cn has the value

(1 + 1)2n 12n+1 + 1
Cn = − = 22n − 1,
1n 1n · (1 + 1)

and the proof is complete.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /431

4412. Proposed by Mihaela Berindeanu.

Let ABC be an acute triangle with incenter I. If Ia , Ib , Ic are the excenters of
−→ −→ −→ → −
ABC, show that IIa + IIb + IIc = 0 if and only if ABC is equilateral.
We received 10 submissions, all of which were correct, and we present the solution
by Cristóbal Sánchez-Rubio with some details added by the editor.
The desired result follows quickly from three familiar theorems; each is elementary
and easy to prove for an arbitrary triangle ABC (acute or not).
−−→ −−→ −−→ → −
1. For a point P in the plane of triangle Ia Ib Ic , P Ia + P Ib + P Ic = 0 if and
only if P is the centroid of ∆Ia Ib Ic .
2. The incenter of the given triangle ABC is the orthocenter of ∆Ia Ib Ic .
3. A triangle is equilateral if and only if its centroid and orthocenter coincide.
−→ −→ −→ →
−
Consequently, IIa + IIb + IIc = 0 if and only if ∆Ia Ib Ic is equilateral. But if
either triangle Ia Ib Ic or ABC is equilateral, its sides would be parallel to the sides
of the other, whence the other would also be equilateral.

4413. Proposed by Miguel Ochoa Sanchez and Leonard Giugiuc.

Let ABC be a triangle with incenter I and circumcircle ω. The lines AI, BI, CI
intersect ω a second time at M, N, P , respectively. Also suppose that N P in-
tersects AB and AC at E and F , respectively. We define points G, H, J and D
analogously (see the picture). Show that if EF = GH = JD, then triangle ABC
is equilateral.

We received 9 submissions, all correct, and present a composite of similar solutions

from Prithwijit De and Jirapat Kaewkam, done independently.
Let T = AI ∩ N P . Observe that in triangle AIP ,
∠AP T = B/2 = ∠IP T and ∠IAP = A/2 + C/2 = ∠AIP.

Copyright © Canadian Mathematical Society, 2019

432/ Solutions

Thus, ∆AIP is isosceles and P T bisects its vertex angle, so that P T ⊥ AI and
AT = T I = AI/2. Moreover, in ∆AEF the bisector AT of the angle at A is
perpendicular to the base, whence ET = T F = EF/2. Thus

AI sin(A/2) r
EF = 2ET = 2AT tan(A/2) = = ,
cos(A/2) cos(A/2)

where r is the inradius of triangle ABC. Similarly,

r r
GH = and JD = .
cos(C/2) cos(B/2)

It follows that EF = GH = JD if and only if cos(A/2) = cos(B/2) = cos(C/2),

which (because 0 < A2 , B2 , C2 < 90◦ ) is equivalent to A = B = C = 60◦ . In other
words, ∆ABC is equilateral if and only if EF = GH = JD.

4414. Proposed by Konstantin Knop.

Let α and β be a pair of circles that intersect in points P and Q, and let the
diameter AA0 of α lie on the same line as the diameter BB 0 of β such that the
end points lie in the order AB 0 A0 B. Suppose that P B 0 intersects α again at
the point C, that P A0 intersects β again at D, and that the lines AD and BC
intersect at R. Prove that the line QR intersects the segment AB at its midpoint.

We received 2 solutions to this problem. However, both submissions utilized brute

force calculations to achieve the result. We leave the problem open in hopes
to receive a more insightful solution. Please email your submissions directly to
crux-editors@ cms. math. ca .

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /433

4415. Proposed by Titu Zvonaru.

Let ABC be an acute-angled triangle with AB < AC, where AD is the altitude
from A, O is the circumcenter and M and N are the midpoints of the sides BC
and AB, respectively. The line AO intersects the line M N at X. Prove that DX
is parallel to OC.
We received 13 correct solutions. Of those, 7 gave a synthetic argument; 4 used
analytic geometry and 1 used barycentric coordinates.

Solution 1, by Dimitrić Ivko.

Let α = ∠CAB, β = ∠ABC, γ = ∠BCA. Since ∠AOC = 2β and OA = OC,
then ∠XAC = 90◦ − β. Since N M kAC, then ∠N XA = ∠XAC = 90◦ − β. Since
BN = N D, then ∠N DA = 90◦ − ∠N DB = 90◦ − β.
Thus, ∠N XA = ∠N DA, and so AN DX is cyclic. Hence
∠AXD = 180◦ − ∠AN D = ∠BN D = 180◦ − 2β = ∠XOC.
It follows that DXkOC.
(Note from Sushanth Sathish Kumar: Once AN DX is proved to be cyclic, DXkOC
follows from ∠N XD = ∠N AD = 90◦ − β = ∠OCA and N XkAC.)

Solution 2, by Vijaya Prasad Nalluri.

Let BC and AX intersect at E. Since N XkAC, triangles M EX and CEA are
similar, so that XE : AE = M E : EC. Since ADkOM , triangles M OE and

Copyright © Canadian Mathematical Society, 2019

434/ Solutions

DAE are similar, so that AE : OE = DE : M E. Hence XE : OE = DE : EC.

Therefore, triangles XED and OEC are similar, so that ∠DXE = ∠COE. This
equality of alternate angles implies that DXkOC.

Solution 3, by Prithwijit De.

Assign coordinates: O ∼ (0, 0), A ∼ (p, q), B ∼ (−b, k), C ∼ (b, k). Then
M ∼ (0, k) and D ∼ (p, k). The equation of the line AO is y = (q/p)x. Since M N
passes through (0, k) and has the same slope as AC, its equation is

k−q
Å ã
y=k+ x.
b−p

The point X where M N and AO intersect has coordinates

kp(b − p) kq(b − p)
Å ã
, .
qb − pk qb − pk

The slope of the line DX is

kq(b−p)
qb−pk −k kp(k − q) k
= = ,
kp(b−p)
−p pb(k − q) b
qb−pk

which is the slope of AC. The result follows.

4416. Proposed by Nguyen Viet Hung.

Let ABC be an acute triangle with orthocentre H. Denote by ra , rb , rc the
exradii opposite the vertices A, B, C, and by r1 , r2 , r3 the inradii of triangles
BHC, CHA, AHB, respectively. Prove that

r1 + r2 + r3 + ra + rb + rc = a + b + c.

We received 9 submissions, all correct, and present the solution by Kee-Wai Lau.
We start with standard formulas for the inradius and an exradius of a triangle
ABC in terms of its circumradius R:

r = R(cos A + cos B + cos C − 1) (1)

and
ra = R(1 + cos B + cos C − cos A). (2)
By (2) and the corresponding expressions for rb and rc we obtain

ra + rb + rc = R(3 + cos A + cos B + cos C). (3)

We now show that

r1 = R(sin B + sin C − cos A − 1). (4)

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /435

Note that the angles of ∆BCH are π2 −C, π2 −B, and B +C, while its circumradius
is
BC a
= = R.
2 sin(B + C) 2 sin A
We get formula (4) by replacing A by B + C, B by π2 − C, and C by π2 − B in (1).
This, together with the corresponding expressions for r2 and r3 , gives us
r1 + r2 + r3 = R(2 sin A + 2 sin B + 2 sin C − cos A − cos B − cos C − 3). (5)
Adding together (3) and (5), we obtain
r1 + r2 + r3 + ra + rb + rc = 2R(sin A + sin B + sin C) = a + b + c,
as desired.

4417. Proposed by Dan Stefan Marinescu, Daniel Sitaru and Leonard Giugiuc.
Let a, b and c be positive real numbers such that abc ≥ 1. Further, let x, y and z
be real numbers such that xy + yz + zz ≥ 3. Prove that
(y 2 + z 2 )a + (z 2 + x2 )b + (x2 + y 2 )c ≥ 6.

We received 4 solutions, 3 of which were correct. We present the solution by

Walther Janous.
We shall prove a more general result, namely:
y 2 + z 2 a + z 2 + x2 b + x2 + y 2 c ≥ 2(xy + yz + zx).

This will follow if we show that Q, defined by

Q = (b + c)x2 + (c + a)y 2 + (a + b)z 2 − 2xy − 2yz − 2zx
is a positive semi-definite quadratic form. But its corresponding symmetric matrix
equals  
b + c −1 −1
M =  −1 c + a −1  .
−1 −1 a + b
We thus have to check only its three principal minors.
• The inequality b + c > 0 is clear.
• Next we have to show that
 
b + c −1 
 −1 c + a > 0,
 

that is
ab + ac + bc + c2 > 1.
By the AM-GM inequality, we get even more:
ab + ac + bc + c2 > ab + ac + bc ≥ 3(abacbc)1/3 = 3(abc)2/3 ≥ 3.

Copyright © Canadian Mathematical Society, 2019

436/ Solutions

• Finally, the inequality det M ≥ 0 has to be verified; that is,

a2 b + ab2 + b2 c + bc2 + c2 a + ca2 + 2abc − 2a − 2b − 2c − 2 ≥ 0,

or equivalently,

a2 b + ab2 + b2 c + bc2 + c2 a + ca2 + 2abc ≥ 2a + 2b + 2c + 2.

Because of abc ≥ 1, it certainly will follow from

a2 b + ab2 + b2 c + bc2 + c2 a + ca2 ≥ 2a + 2b + 2c. (1)

Since
5 2
1 2 5 2 1
Å ã
a b + a2 c + b c + bc2 = (b + c) ·


a + bc ,
6 6 6 6
the left-hand expression in (1) can be written as
5 2 1 5 2 1 5 2 1
Å ã Å ã Å ã
(b + c) · a + bc + (c + a) · b + ca + (a + b) · c + ab .
6 6 6 6 6 6
But the AM-GM inequality yields
Å
5 2 1
ã √
5/6
(b + c) · a + bc ≥ 2 bc · a2 · (bc)1/6
6 6
= 2a5/3 (bc)2/3
= 2a(abc)2/3 ≥ 2a,

and two similar inequalities for the other two summands. This completes
the proof.
Remark: It is a bit disturbing to have a, b, c limited by the constraint abc ≥ 1.
We shall remove it as follows. Let a, b, and c be arbitrary positive real numbers.
Then the three numbers
a b c
a1 = 1/3
, b1 = 1/3
, c1 =
(abc) (abc) (abc)1/3
satisfy the condition
a1 b1 c1 = 1 ≥ 1,
whence by what we have already shown,

(b1 + c1 ) x2 + (c1 + a1 ) y 2 + (a1 + b1 ) z 2 ≥ 2(xy + yz + zx);

that is, there holds even more generally

(b + c)x2 + (c + a)y 2 + (a + b)z 2 ≥ 2(abc)1/3 (xy + yz + zx)

for all positive real numbers a, b, and c and all real numbers x, y, and z. Fur-
thermore, the various applications of the AM-GM inequality show that equality
occurs if and only if a = b = c and x = y = z.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /437

4418. Proposed by Daniel Sitaru.

Consider a convex cyclic quadrilateral with sides a, b, c, d and area S. Prove that

(a + b)5 (b + c)5 (c + d)5 (d + a)5

+ + + ≥ 64S 2 .
c+d d+a a+b b+c

We received 7 correct solutions. We present 5 of them.

We make some preliminary remarks. The formula for the area S of a quadrilateral
with sides a, b, c, d and perimeter 2s = a + b + c + d is
»
S = (s − a)(s − b)(s − c)(s − d) − abcd cos2 θ,

where θ is half the sum of two opposite angles. This is dominated by the area of
a cyclic quadrilateral with the same sides, namely
»
(s − a)(s − b)(s − c)(s − d)
1»
= (b + c + d − a)(c + d + a − b)(d + a + b − c)(a + b + c − d)
4
1 »
= [(a + b)2 − (c − d)2 ][(c + d)2 − (a − b)2 ]
4
1»
= [(a + c)2 − (b − d)2 ][(b + d)2 − (a − c)2 ].
4
The statement of the problem remains true for noncyclic quadrilaterals.

Solution 1, by Oliver Geupel.

Let
(w, x, y, z) = (s − a, s − b, s − c, s − d).
Then
(a + b, b + c, c + d, d + a) = (y + z, z + w, w + x, x + y).
Applying the arithmetic-geometric means inequality twice, we find that

(a + b)5 (b + c)5 (c + d)5 (d + a)5

+ + +
c+d d+a a+b b+c
(y + z)5 (z + w)5 (w + x)5 (x + y)5
= + + +
w+x x+y y+z z+w

≥ 4(y + z)(z + w)(w + x)(x + y)

√ √ √ √
≥ 4(2 yz)(2 zw)(2 wx)(2 xy)

= 64xyzw ≥ 64S 2 .

Equality holds if and only if the quadrilateral is a square.

Copyright © Canadian Mathematical Society, 2019

438/ Solutions

Solution 2, by S̆efket Arslanagić.

By the arithmetic-geometric means inequality,
» »
S ≤ (s − a)(s − b) (s − c)(s − d)
1 1
≤ (2s − a − b)(2s − c − d) = (c + d)(a + b).
4 4

Similarly, S ≤ 41 (b + c)(a + d). Therefore

64S 2 = 4(16S 2 )
≤ 4(a + b)(b + c)(c + a)(d + a)
ò1/4
(a + b)5 (b + c)5 (c + d)5 (d + a)5
ï
=4 · · ·
c+d d+a a+b b+c
(a + b)5 (b + c)5 (c + d)5 (d + a)5
≤ + + + .
c+d d+a a+b b+c

Solution 3, by C.R. Pranesachar.

By the arithmetic-geometric means inequality,

(a + b)5 (c + d)2
+ ≥ 2[(a + b)2 (c + d)2 ]
c+d a+b
≥ 2[(a + b)2 − (c − d)2 ][(c + d)2 − (a − b)2 ]
≥ 32S 2 .

A similar inequality holds for the other two terms of the left side and the result
follows.

Solution 4, by Leonard Giugiuc and Digby Smith, independently. We have:

64S 2 = 64(s − a)(s − b)(s − c)(s − d)

ò4
(s − a) + (s − b) + (s − c) + (s − d)
ï
≤ 64
4
Å ã4
2s
= 64
4
4
= 4s .

From an instance of the Hölder inequality, for positive x, y, z, t, m, n, p, q,

x5 y5 z5 t5
Å ã
+ + + (m + n + p + q)(1 + 1 + 1 + 1)3 ≥ (x + y + z + t)5 ,
m n p q

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /439

applied to

(x, y, z, t; m, n, p, q) = (a + b, b + c, c + d, d + a; c + d, d + a, a + b, b + c),

we find that the left side is not less than

25 (a + b + c + d)5 210 s5
= 8 = 4s4 ≥ 64S 2 .
4 · 2(a + b + c + d)
3 2 s

Solution 5, by Walther Janous.

We prove a more general result: Let p > q > 0 and p + q ≥ 1. Then
(a + b)p (b + c)p (c + d)p (d + a)p
+ + + ≥ 2p−q+2 S (p−q)/2 .
(c + d)q (d + a)q (a + b)q (b + c)q
Applying the arithmetic-geometric means inequality to the denominator yields

(a + b)p (c + d)p (a + b)p+q + (c + d)p+q

+ =
(c + d)q (a + b)q [(a + b)(c + d)]q
[(a + b)p+q + (c + d)p+q ]
≥ 22q · ,
(a + b + c + d)2q
with an analogous inequality for the other two terms on the left side. Using the
convexity of xp+q , we see that the left side is not less than

(a + b)p+q + (b + c)p+q + (c + d)p+q + (d + a)p+q

ï ò
22q
(a + b + c + d)2q
òp+q
22q · 4 (a + b) + (b + c) + (c + d) + (d + a)
ï
≥
(a + b + c + d)2q 4

22q+2 a + b + c + d p+q
ï ò
=
(a + b + c + d)2q 2

= 2q−p+2 (a + b + c + d)p−q .

On the other hand, from the AM-GM inequality [as in Solution 4],

(a + b + c + d)2
S≤
4
whereupon

(a + b + c + d)p−q
ï ò
2p−q+2 S (p−q)/2 ≤ 2p−q+2
22(p−q)
= 2q−p+2 (a + b + c + d)p−q .

The result follows.

Copyright © Canadian Mathematical Society, 2019

440/ Solutions

4419. Proposed by Michel Bataille.

Let ABC be a triangle with ∠BAC = 90◦ . Let D on the hypotenuse BC produced
beyond C be such that CD = CB +BA. The internal bisector of ∠ABC intersects
the line through the midpoints of AB and AC at T . Prove that ∠T CA = ∠CDA.
We received 13 solutions. We present the solution by Mihai Miculiţa and Titu
Zvonaru.

Use a, b and c to denote the lengths of the sides of the triangle. Let M and N be
the midpoints of AB and AC, respectively, and P be the intersection of line M N
with line AD. Note that M N ||BC and M N = a2 ; also, P must be the midpoint
of AD.
Since BT is the bisector of ∠CBA, so ∠CBT = ∠T BA, and M N ||BC gives us
∠M T B = ∠T BC. So 4M T B is isosceles, giving us M T = M B = c/2, and hence
N T = M N − M T = a−c 2 .
a+c
Since N P joins the midpoints of AC and AD, using CD = a+c we get N P = 2 .

Thus
NT · NP = NA · NC ⇐⇒ (a − c)(a + c) = b2 ,
which holds by the Pythagorean Theorem. We deduce that the quadrilateral
P CT A is cyclic, thus ∠T CA = ∠T P A.
Finally, since N P ||CD we have ∠T P A = ∠CDA, allowing us to conclude ∠T CA =
∠CDA.

4420. Proposed by Leonard Giugiuc and Marian Dinca.

Let A0 A1 . . . An−1 , n ≥ 10 be a regular polygon inscribed in a circle of radius r
centered at O. Consider the closed disks ω(Ak ), k = 0, . . . , n − 1 centered at Ak of
radius r. Prove that
n−1
\
ω(Ak ) = {O}.
k=0

We received 4 solutions. Presented is the one by Walther Janous, lightly edited.

We show that the statement holds for all n ≥ 3. First of all, it’s clear that O is an
element of the intersection under consideration. We now distinguish two cases.
If n is even, then the disks ω(A0 ) and ω(An/2 ) are tangent to each other.

Crux Mathematicorum, Vol. 45(7), September 2019

 Solutions /441

If n is odd, let n = 2k + 1. We consider ω(A0 ) and its opposite disks ω(Ak ) and
ω(Ak+1 ). Suppose w.l.o.g. that A0 is north of O and the vertices of the polygon
are ordered counterclockwise. Then the boundary circles of ω(A0 ) and ω(Ak )
intersect in O (the southernmost point of ω(A0 ) and a point P northwest of O.

Thus the intersection ω(A0 ) ∩ ω(Ak ) is contained in the western hemisphere. By

a similar argument the intersection ω(A0 ) ∩ ω(Ak+1 ) is contained in the eastern
hemisphere. Therefore the intersection ω(A0 ) ∩ ω(Ak ) ∩ ω(Ak+1 ) consists of only
O and the claim follows.

Copyright © Canadian Mathematical Society, 2019

 SNAPSHOT

Mathematician Marco Buratti in Val di Rabbi, a subvalley of Val di Sole, proving

that certain graphs can survive at the altitude of 3000 meters (see the T-shirt).