Volume/tome 49, issue/numéro 6

June/juin 2023
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 2023. 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 2023. 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 Grant McLoughlin University of New Brunswick

Shawn Godin Cairine Wilson Secondary School
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 Toronto, Canada
Magdalena Georgescu Toronto, Canada
Chip Curtis Missouri Southern State University
Philip McCartney Northern Kentucky University
Guest Editors Yagub Aliyev ADA University, Baku, Azerbaijan
Ana Duff Ontario Tech University
Mateusz Buczek Warsaw, Poland
Andrew McEachern York University
Vasile Radu Birchmount Park Collegiate Institute
Chi Hoi Yip University of British Columbia
Didier Pinchon Toulouse, France
Matt Olechnowicz University of Toronto
Translators Rolland Gaudet Université de Saint-Boniface
Frédéric Morneau-Guérin Université TÉLUQ
Editor-at-Large Bill Sands University of Calgary
 IN THIS ISSUE / DANS CE NUMÉRO

287 MathemAttic: No. 46

287 Problems: MA226–MA230
289 Solutions: MA201–MA205
293 Teaching Problems: No. 22 John Grant McLoughlin
296 From the bookshelf of . . . Ruwan C. Karunanayaka
299 Mathemagical Puzzles: No. 5 Tyler Somer
305 Mathematics from the Web: No. 11
306 Olympiad Corner: No. 414
306 Problems: OC636–OC640
308 Solutions: OC611–OC615
313 Hales-Jewett theorem through examples and exercises (II)
Veselin Jungić
322 Problems: 4851–4860
326 Solutions: 4801–4810

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
 MathemAttic /287

MATHEMATTIC
No. 46
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 August 15, 2023.

MA226. The numbers a, b and c are in arithmetic sequence. The numbers

b, c and d are in geometric sequence. If a = 1.27, d = 3.68 and c is positive,
determine c.

MA227. Find and prove the general formula for the square root of the
product of four consecutive integers plus 1.

MA228. Two circles with radii r1 and of r2 are a distance d apart from each
other. A point P is to be placed on the line connecting the centers of the two
circles so that the tangent lines to the circles go through the point forming angles
φ and ψ as shown below. How far from the center of the left circle should one
place the point P so that φ = ψ? Write your answer in terms of r1 , r2 , and d.

r2
r1
P
φ ψ

MA229. Determine the largest real number t such that the two polynomials
x4 + tx2 + 1 and x3 + tx + 1 have a common root.

MA230. Proposed by Titu Zvonaru, Comăneşti, Romania.

Let ABC be an isosceles triangle with AB = AC and AD, BE, and CF be its
altitudes. A circle of diameter CE intersects the lines BC and CF at M and N ,
respectively. The line M N intersects the altitude AD at P . Prove that DP = M E.

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

Copyright © Canadian Mathematical Society, 2023

288/ MathemAttic

Les problèmes 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 15 août 2023.

MA226. Les nombres a, b et c sont en progression arithmétique, tandis que

les nombres b, c et d sont en progression géométrique. Si a = 1.27, d = 3.68 et c
est positif, déterminer c.

MA227. Déterminer et démontrer comment, par une formule générale, cal-

culer la racine carrée du produit de quatre entiers consécutifs, auquel on a ajouté
1.

MA228. Deux cercles de rayon r1 et r2 respectivement sont situés à une

distance d l’un de l’autre. On place un point P sur la droite liant les centres des
deux cercles de sorte que les droites tangentes aux cercles passant par le point
forment des angles φ et ψ tel qu’illustré ci-dessous. À quelle distance du centre du
cercle de droite doit-on placer le point P afin que φ = ψ? Exprimez votre réponse
en fonction de r1 , r2 et d.

r2
r1
P
φ ψ

MA229. Déterminer le plus grand nombre réel t, tel que les deux polynômes
x4 + tx2 + 1 et x3 + tx + 1 ont une racine en commun.

MA230. Proposé par Titu Zvonaru, Comăneşti, Romania.

Soit ABC un triangle isocèle tel que AB = AC et soient AD, BE et CF ses
hauteurs. Un cercle de diamètre CE rencontre les lignes BC et CF en M et N
respectivement. De plus, la ligne M N rencontre la hauteur AD en P . Démontrer
que DP = M E.

Crux Mathematicorum, Vol. 49(6), June 2023

 MathemAttic /289

MATHEMATTIC
SOLUTIONS
Statements of the problems in this section originally appear in 2023: 49(1), p. 5–6.

MA201. The figure below consists of 4 congruent squares. Find the angle θ.

Originally question 10 from the 34th University of Alabama High School Mathe-
matics Tournament: Team Competition, 2015.
We received 16 submissions of which 15 were correct and complete. We present
the solution by Brian Bradie.
Because the squares are congruent, the steeper line has a slope of 2, which implies
that the angle of inclination, θ1 , of the steeper line satisfies tan θ1 = 2. On the
other hand, the shallower line has a slope of 13 , so the angle of inclination, θ2 , of
the shallower line satisfies tan θ2 = 31 . Thus,

1 2 − 31 π
θ = θ1 − θ2 = tan−1 2 − tan−1 = tan−1 = tan−1 1 = .
3 1 + 2 · 13 4

MA202. Two players, A and B, play a game with a fair six-sided die. The
goal is to roll a 2 or a 5: whoever does so first wins the game. The players take
turns rolling the die, with player A going first. They keep rolling until someone
rolls a 2 or a 5. What is the probability that player A wins the game?
Originally question 3 from the 2009 Fifth Annual Kansas Collegiate Mathematics
Competition.
We received 8 solutions. We present the solution by Aravind Mahadevan, lightly
edited.
Let PA and PB denote the probabilities of A and B winning the game respectively.
If A misses getting a 2 or 5 on his first turn, then B has the same likelihood of
winning the game that A had at the beginning of the game. Therefore PB = 23 PA .
Since PA + PB = 1, we obtain
2 5
1 = PA + PA = PA .
3 3

Copyright © Canadian Mathematical Society, 2023

290/ MathemAttic

Thus the probability of A winning equals 53 .

MA203. Proposed by Digby Smith, Calgary, AB.

Suppose that a, b, c, d are positive real numbers such that

a2 + b2 = c2 + d2 with a + b > c + d.

Show that
c4 + d4 > a4 + b4 .

We received 10 submissions, all of them were complete and correct. We present

the solution by Ivan Hadinata, modified by the editor.
Let a, b, c, d ∈ R+ satisfying a2 + b2 = c2 + d2 and a + b > c + d. Then,

a2k + b2k < c2k + d2k

for k ∈ {2, 3}.

To prove this, note that we are given that a2 + b2 = c2 + d2 and a + b > c + d.
Thus,

2ab = (a + b)2 − a2 − b2 > (c + d)2 − c2 − d2 = 2cd =⇒ ab > cd.

Therefore a4 + b4 = (a2 + b2 )2 − 2a2 b2 < (c2 + d2 )2 − 2c2 d2 = c4 + d4 , as the

original problem wants to prove. And then,

a6 + b6 = (a4 + b4 − a2 b2 )(a2 + b2 ) < (c4 + d4 − c2 d2 )(c2 + d2 ) = c6 + d6 .

MA204. Proposed by Alaric Pow Ian-Jun, Singapore.

Find the exact value of
» √ » √ » √
28 + 12 5 + 28 − 12 5 + 2 43 − 30 2.

We received 17 submissions, 16 of them were complete. We present the solution

by Matteo Vitali.
p √
It is well known that (a + b)2 = a2 + 2ab + b2 = |a + b|. Thus, to get rid of
the first square root, we can search for a and b such that:
® 2
a + b2 = 28
√
2ab = 12 5
In particular, this is the simplest system to solve for a and b. All the others of the
form: ® 2
a + b2 = α
2ab = β

Crux Mathematicorum, Vol. 49(6), June 2023

 MathemAttic /291

√
where α + β = 28 + 12 5 and α ≥ 0 will lead (with more calculations) to the same
value of a + b. So, we have:
180

2
 4 2  2 2
 2 + b = 28  b − 28b + 180 = 0  (b − 18) · (b − 10) = 0


b √ √ √
⇐⇒ ⇐⇒
 6 5 a = 6 5 a = 6 5
a =

b b
b
There are four solutions:
√ √ √ √ √ √ √ √
{( 10, 18); (− 10, − 18); ( 18, 10); (− 18, − 10)}
p √ √ √ √ √
Thus: 28 + 12 5 = | 10 + 18| = 10 + 18.

In the second square root the double product is negative. So, a and b have different
p √ √ √ √ √
signs: 28 − 12 5 = | 10 − 18| = 18 − 10.

We can repeat the same process for the third square root:
450

2
 4 2
 2 + b = 43  b − 43b + 450 = 0
® 2 
a + b2 = 43

b √ √
√ ⇐⇒ ⇐⇒
2ab = −30 2  15 2  a = − 15 2
a = −

b
b
Again a and b are:
√ √ √ √ » √ √
{(5, −3 2); (−5, 3 2); (3 2, −5); (−3 2, 5)} =⇒ 43 − 30 2 = 5 − 3 2

In the end, we can say:

» √ » √ » √
28 + 12 5 + 28 − 12 5 + 2 43 − 30 2
√ √ √ √ √
= 18 + 10 + 18 − 10 + 2 · (5 − 3 2) = 10.

MA205. Suppose a circle is inscribed in an equilateral triangle with side

length two metres. Another circle is inscribed in the upper corner as shown below.
Find the area A between the smaller circle and the corner of the triangle.
A

Copyright © Canadian Mathematical Society, 2023

292/ MathemAttic

Originally question 12 from the 35th University of Alabama High School Mathe-
matics Tournament: Team Competition, 2016.
We received 9 submissions, 8 of which were correct. We present the solution by
Ivan Hadinata.
X

A
P2

O2

P1 Q1

O1

Y Z

Notice the above picture. Let ∆XY Z be the given equilateral triangle. Let us
name Γ1 and Γ2 respectively the bigger and smaller circles as drawn in the figure.
Suppose that Γ1 , Γ2 are respectively centered at O1 , O2 and touching side XY
respectively at P1 and P2 . Γ1 touches XZ at Q1 . Also, let B be the region
constrained by segment XP1 , segment XQ1 , and minor arc P1 Q1 of Γ1 .
Observe that there exists a homothety H with center X and factor P2 O2
P1 O1 that sends
Γ1 to Γ2 , O1 to O2 , and B to A. Now we have XP1 = 1, and let P2 O2 = r. By
using trigonometric ratios, we get XO2 = 2r, P1 O1 = √13 , and XO1 = √23 . We
√
3 3−π
have that the area of B is 9 ; and

2 1 1
√ = XO2 + O1 O2 = 3r + √ =⇒ r = √ .
3 3 3 3
Ä ä2
P 2 O2
Thus, the area of A equals P 1 O1
times the area of B, that is

ã2 √ 1
√
!2 √ √
3 3−π 3 3−π 3 3−π
Å
P2 O2 3 3
· = · = .
P1 O1 9 √1 9 81
3

1/3
Editor’s Comments. The scaling factor is immediate if you use XP XP1 = 1 . The
2

area of B was √calculated by taking one third of the difference between the large
triangle (area 3) and its incircle (area π/3).

Crux Mathematicorum, Vol. 49(6), June 2023

 John Grant McLoughlin /293

TEACHING PROBLEMS
No. 22
John Grant McLoughlin
Promoting Geometric Probability as a Method of
Solution
Last week in a school I found myself with two middle school students who wanted
a mathematical challenge. The following instruction was given to them. “Write
down a number between 0 and 10. The number does not have to be an integer.”
The question then posed was What is the probability that these two numbers will
be within 2 of each other?
This was the starting point for a discussion. We first made some predictions. Then
Susan (the teacher) and I also wrote down numbers. This gave us six different
possible pairings to consider with respect to how many would meet the requirement
of not differing by more than two.
The idea of drawing a picture to represent the problem was then suggested by
me. The intention here was to introduce geometric probability. That is, the
probability could be found by visually representing all possibilities and determining
the fraction of that area which would meet the requirements. My suggestion was
to make a grid of some sort and then figure a way of working out the probability.
Time was given to figure a plan. One of the students noted that the area is going
to be found by using lines related to differences of 2. This person suggested a grid
with spacing by 2, as shown below. That is, the axes would go from 0 to 10 with
intervals of 2. Further it was observed that the values differing by at most 2 would
be represented by the shaded region. The axes were labelled with the names of
the two students. Hence, it can be seen that if Edward selected 2, the number
Chris selected could be as small as 0 or as large as 4.
10

6
Chris

2 4 6 8 10
Edward

Copyright © Canadian Mathematical Society, 2023

294/ Teaching Problems

What would be the desired probability? Initially the number of squares was
counted as 9 including the five full squares and four pairs of half-squares. Care
then had to be taken to recognize that only 25 squares were in the grid rather
9
than the 100 as may be suggested by 10 × 10. The desired probability was 25 or
0.36. We considered this answer and revisited the predictions. A difference of 2
may suggest that there is a window of 4 for any guess as being 2 above or below
would satisfy the requirement. However, numbers less than 2 or greater than 8
would not give such wide windows as the boundaries of 0 and 10 would limit the
acceptable ranges. So an answer a bit less than 4 out of 10 or 0.40 would seem
reasonable.
Alternative approach
A reasoned approach without a picture could be employed here. When a number
6
is initially selected it would be expected to be between 2 and 8 in 10 of the cases.
4
Such selections would offer the full range of 4 or a 10 chance of selecting the second
4
number that differs by at most 2. In the remaining 10 of the cases, there will be a
restricted range of sorts to consider. For example, if a number is selected between
0 and 2 it will be limited in the lower direction by the 0. Likewise the numbers
from 8 through 10 will be restricted on the high side. That is, 4 out of 10 numbers
will be limited on average to 1 rather than 2 in one direction, thus, making for
3
only a 10 chance of getting a second number within 2.
The desired probability becomes
6 4 4 3 36
× + × = .
10 10 10 10 100

Personally, as one who is comfortable with such explanations this is fine. However,
as a teacher it has become clear to me that the visual representation of the prob-
ability has made much more sense to students generally. Geometric probability
was not part of my school math experience. Rather it was first encountered by
me in the 1980’s through a piece that appeared in an NCTM publication entitled
Mathematical Challenges II Plus Six. The problem discussed by Philip Smith is
shared here.
Two witches enjoy meeting each other over a cauldron of tea. Both
witches have serious shortcomings, however. First, each witch is poorly
organized and arrives at the meeting place randomly between midnight
and 1:00 a.m. Second, each is notoriously evil-tempered and becomes
outraged on having to wait 15 minutes or longer for her companion.
Thus, the following temper-saving arrangement has been agreed on:
when either witch has waited fifteen minutes - or when one o’clock
arrives and she is still alone - she disappears at once, not returning
until the next night. Here is our problem: On a given night, what is
the probability that the two witches meet?
It is the so called “witches problem” that has been adapted by me as a teaching
problem using two people arriving at a coffee shop with similar time constraints

Crux Mathematicorum, Vol. 49(6), June 2023

 John Grant McLoughlin /295

to those of the witches.

Concluding comments
Visual representations can be used in other ways to solve probability problems.
It remains a helpful idea for solving probability problems if one can represent a
sample space, or all possibilities, in an area model and then identify the portion
that satisfies the requirements as the desired probability.
Readers are encouraged to solve the problem with the witches using both geo-
metric probability and a reasoned approach resulting in a sum of products, as
shown earlier. Further, one may wish to play with the initial problem by consid-
ering different selection boundaries than 0 to 10 for the numbers or considering
a difference of no more than 1 or 3, for example. Facility with visual representa-
tions of probabilistic situations will broaden one’s mathematical problem solving
repertoire.

Copyright © Canadian Mathematical Society, 2023

296/ From the bookshelf of . . .

From the bookshelf of . . .

Ruwan C. Karunanayaka
This MathemAttic feature brings attention to books of potential interest to the
readers. Some of these will be reviews whereas others will be hearty recommenda-
tions from the contributors. If you have a book related to mathematics that would
be of interest to secondary school students and/or teachers, feel welcome to send
along a submission to [email protected]. Publishers are also welcome to
send along books for possible review.
Struck by Lightning: The Curious World of Probabilities
by Jeffrey S. Rosenthal
ISBN 978-0006394952, 288 pages
Published by Harper Perennial, Year 2006.

Struck by Lightning: The Curious World of Probabilities by Jeffrey S. Rosenthal

is a captivating exploration of the world of probability and its profound influence
on our daily lives. The book is an invaluable resource that can be used to enrich
the learning experience in a classroom setting, particularly for secondary school
teachers and students.
Rosenthal’s book is a journey into the heart of probabil-
ity, illustrating how it fills every aspect of our lives. He
uses relatable examples to explain complex concepts,
making the subject matter accessible to readers of all
backgrounds. For instance, he discusses the role of
probability in medical studies, weather forecasts, and
even the spam emails we receive. These examples could
serve as excellent teaching tools, providing real-world
context to abstract mathematical concepts.
One of the most memorable parts of the book is when
Rosenthal discusses the butterfly effect in the context
of weather forecasting. He explains how weather, like
coins and cards, is a chaotic system where very small
changes today can cause significant differences tomorrow. This could be an excel-
lent way to introduce students to the concept of chaos theory and its implications.
Rosenthal also examines the world of spam emails, explaining how probability the-
ory helps computers separate spam from genuine messages. This example could be
used to teach students about the practical applications of probability in technology
and cybersecurity.
Rosenthal’s book is filled with case studies that provide a deeper understanding
of the role of probability in various aspects of life. For instance, he discusses the
principle of ignoring the extremely improbable, explaining how events of extremely
small probability should generally be ignored when making decisions. This princi-

Crux Mathematicorum, Vol. 49(6), June 2023

 Ruwan C. Karunanayaka /297

ple is illustrated with an example of fearing an accident while standing under the
CN Tower. This could be a great way to teach students about risk assessment and
the role of probability in decision-making.
Another interesting case study in the book is the discussion on the correlation
and causation. Rosenthal uses the example of a study that found that medical-
school class presidents lived an average of 2.4 years less than other medical school
graduates. He explains that just because being class president is correlated with
shorter life expectancy does not mean that it causes shorter life expectancy. This
could be a great way to teach students about the difference between correlation
and causation, a concept that is often misunderstood.
In the chapter “What Are The Odds Of That?”, Rosenthal discusses the concept
of games of chance. He suggests that to win, one needs three ingredients: a well-
studied strategy that wins on average, repetition of the strategy, and patience
for the Law of Large Numbers to eventually lead to victory. This concept can
be applied in the classroom to teach students about strategic thinking and the
importance of patience and consistency in achieving long-term goals.
One of the most intriguing examples Rosenthal provides involves the game of poker.
He explains the concept of “drawing to an Inside Straight” versus “drawing to an
Outside Straight”. If you’re dealt a 5, 6, 8, and 9, you’re drawing to an Inside
Straight, and your chances of getting a 7 to complete the straight are quite small.
However, if you’re dealt a 5, 6, 7, and 8, you’re drawing to an Outside Straight,
and your chances of getting a 4 or 9 to complete the straight are twice as good.
This example can be used to teach students about conditional probability and the
importance of strategic thinking in games of chance.
Rosenthal also discusses the game of five-card stud poker. If you have three Queens
in your first four cards, and your opponent has two 5’s and one 4 showing, plus
one secret face-down card, you might assume that your three Queens beat three
5’s. However, your opponent’s heavy betting might suggest that their face-down
card could be another 5. This scenario can be used to teach students about the
concept of uncertainty and the role of probability in decision-making.
Rosenthal also explains why casinos always win. The secret lies in the Law of
Large Numbers. Even if a player has a strategy that wins on average, the casino
relies on players playing repeatedly. Over time, the Law of Large Numbers ensures
that the casino, which always has a slight edge, will come out on top. This can be
a valuable lesson for students in understanding how seemingly fair games can be
subtly biased and the long-term effects of probability.
Rosenthal also discusses the concept of true randomness, explaining how tradi-
tional science believed that randomness is caused purely by ignorance. He uses
the example of predicting the movement of a baseball or a planet, explaining how
physics can predict these movements precisely. This could be a great way to
introduce students to the concept of randomness and its role in science.
The book also digs into the famous “birthday problem”. The birthday problem is

Copyright © Canadian Mathematical Society, 2023

298/ From the bookshelf of . . .

a well-known probability puzzle that goes like this: In a group of just 23 people,
there’s a 50% chance that at least two people have the same birthday. Despite
there being 365 days in a year, it only takes a group of 23 for there to be a 50-50
chance of a shared birthday, and a group of 70 for there to be a 99.9% chance.
This counter-intuitive problem can be a great way to spark students’ interest in
probability.
“Struck by Lightning: The Curious World of Probabilities” by Jeffrey S. Rosenthal
is a treasure trove of real-world examples and engaging stories that make the
abstract concept of probability accessible and interesting. It’s a great resource for
secondary school teachers and students alike, providing numerous opportunities
for classroom discussions and activities.

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

This book is a recommendation from the bookshelf of

Dr. Ruwan C. Karunanayaka. Dr. Karunanayaka is an
Assistant Professor in Statistics at the University of the
Fraser Valley (UFV). He brings a unique approach to
teaching, employing practical examples to clarify theo-
retical ideas in statistics. Dr. Karunanayaka believes in
the importance of understanding real-world applications
to enhance learning in his field.

Crux Mathematicorum, Vol. 49(6), June 2023

 Tyler Somer /299

MATHEMAGICAL PUZZLES
No. 5
Tyler Somer
Molten Gold - V
This is the concluding article in this series. We started the series by looking
at seemingly-impossible packing puzzles. Many of the puzzles include pieces that
have measurements which can be related to the numbers in the Fibonacci Sequence.
Readers unfamiliar with the Fibonacci Sequence and its association to the Golden
Ratio have a wonderful journey of mathematical discovery ahead of them.
More recently, we have looked at tray-packing puzzles in which a tile seemingly
melts into the tray, but only after the pieces are rearranged. For rectangular
rearrangements, the pieces are turned 90◦ , thus changing their horizontal and
vertical components. Equivalently, turning over rectangular pieces by reflecting
them through a 45◦ axis creates the same swap of their components. When Bill
Cutler created the non-rectangular parallelogram Pentominoes-MB puzzle (see the
previous article in this series [2023: 49(4), 192-194]), he gave the hint to turn the
pieces over. Notice that to fill a non-rectangular parallelogram, the acute angles
of the pieces must be oriented the same way. When these pieces are flipped, and
the acute angles are properly realigned, the lateral components – we cannot call
them horizontal and vertical components – are similarly swapped.
Modern puzzle designers continue to seek out innovative ways of hiding the trick
of such melting-tile puzzles in two dimensions and, analogously, melting-block
puzzles in three dimensions. In fact, Bill Cutler states that the “MB” suffix of
his Pentominoes-MB puzzle is a direct reference to the “Melting Block” puzzle,
first created by Thomas H. O’Beirne of Scotland. This class of puzzle apparently
originated with O’Beirne’s Melting Block. Let us take a closer look at the original.
Thomas H. O’Beirne was a professor of Computer Studies at the University of
Glasgow. He wrote an ongoing column, Puzzles and Paradoxes, as a long-time
contributor to the British magazine The New Scientist. Many of these columns
were compiled in book form in 1965, with the unsurprising title: Puzzles and Para-
doxes (New York, Oxford University Press). According to personal correspondence
with Bill Cutler, the original Melting Block puzzle was designed by O’Beirne in
the 1970s. There seems to be, however, very little literature about the puzzle. I
am once again thankful to Bill Cutler for providing information on the original
Melting Block puzzle, which I can happily share with readers here:
The Melting Block uses pieces which are based on a rectangular solid “brick” of
dimensions 19 × 29 × 44. When 27 of these bricks are stacked in a 3 × 3 × 3
arrangement, with all the bricks oriented the same way, the result is a rectangular
solid of size 57 × 87 × 132. However, if you take 28 of these same bricks and stack
them in a 2×2×7 arrangement, with the smallest dimension being stacked 7 wide,

Copyright © Canadian Mathematical Society, 2023

300/ Mathemagical Puzzles

then the resulting rectangular solid has size 58 × 88 × 133.

Figures 1a and 1b: The original Melting Block puzzle

To determine the dimensions reported above, we can construct a system of three
variables. Figure 1 provides a satisfactory visual aid. Consider that a < b < c, and
notice that all three variables change attitude between the 3 × 3 × 3 and 2 × 2 × 7
arrangements. This provides the system:

7a = 3c + 1;
2b = 3a + 1;
2c = 3b + 1.

The reader can either solve the system as an exercise, or simply verify the above
report.

Figure 2: The 3 × 3 × 3 dissection of the original Melting Block puzzle

Once the basic brick is defined, a meaningful puzzle must be designed. Merely
using 28 unit bricks is neither interesting nor challenging. O’Beirne chose a simple
dissection of the 3 × 3 × 3 arrangement, as shown in Figure 2. By slicing the large
block in a 2 : 1 ratio in each orthogonal direction, eight unique pieces are formed.
The simplicity of this dissection is quite elegant.
(Aa, Bb, Cc) Piece dimensions
(a, b, c) (19, 29, 44)
(a, b, 2c) (19, 29, 88)
(a, 2b, c) (19, 58, 44)
(a, 2b, 2c) (19, 58, 88)
(2a, b, c) (38, 29, 44)
(2a, b, 2c) (38, 29, 88)
(2a, 2b, c) (38, 58, 44)
(2a, 2b, 2c) (38, 58, 88)

Table 1: The eight pieces of the original Melting Block.

Crux Mathematicorum, Vol. 49(6), June 2023

 Tyler Somer /301

Taking note of Figure 2, pairs of adjacent pieces can be swapped in a single layer.
Alternatively, an entire layer can be rotated and/or reflected. By various com-
binations of these swaps, rotations, and reflections in the piece layers, some 93
“basic” solutions to the 3 × 3 × 3 box packing are possible. The entire box packing
can be subsequently rotated and reflected 8-fold, resulting in 744 apparent packing
solutions. With a ninth piece added – necessarily a copy of the unit brick – the
2 × 2 × 7 packing becomes the problem to solve. Cutler reassures us, dear reader,
that this second packing is a little more interesting and challenging to solve. In-
deed, there is but one fundamental solution to the 2 × 2 × 7 packing. It is left to
the reader to determine why there are 48 apparent packing solutions to this final
challenge.

For the woodworkers among the readership, there are many construction options.
One is to make 28 copies of the unit brick and create the larger pieces by gluing up
the bits appropriately. Another is to make each piece from a single slab of wood,
with no gluing. A single type of wood could be used for all the pieces, or up to nine
species of wood can create an exotic look, with or without gluing. Taken to the
extreme, gluing bricks made of 28 distinct wood species would create a countless
variety of exotic combinations. A Red Stone version of the puzzle exists with the
ninth piece painted red, and the other eight pieces a single species of wood. A
suitable basic measurement must be chosen, perhaps 1 millimetre or some fraction
of an inch, depending on the tools and equipment in your shop. A caliper is a
necessity. And, of course, the box itself must be sized accordingly, to give enough
play without being sloppy.

For the designers, there are four optional pieces that can be considered, each with
3a = 57 as the measure in that one dimension. Note that the puzzle’s design does
limit both 2b = 58 and 2c = 88 as the maximum measures in those respective
dimensions. As a practical matter, though, the pieces that are 57 × 58 × 44 and
57 × 58 × 88 should be removed from consideration. The almost-square profile of
either piece can be problematic: even if the piece(s) and the box are very precise,
the solver might not notice that such an almost-square piece could be in the wrong
orientation. Removing these two pieces leaves us with only ten useful pieces in this
universe of possibilities. The question arises: must there always be some piece –
any piece, not just the unit brick – that would be repeated in the puzzle? Or, is it
possible to have some set of unique non-unit pieces fill the 3 × 3 × 3 arrangement,
then have the addition of a single unit brick satisfy the 2 × 2 × 7 arrangement?

The next pair of arrangements to try might be 4 × 4 × 4 and 3 × 3 × 7, since

64 = 63 + 1. Indeed, Bill Cutler and colleague John Rausch did just that. Some
of their early results were presented at a public gathering in February of 2016,
while a formal compilation and summary of their extensive computer analysis was
published in September of 2018. I include some highlights here.

Copyright © Canadian Mathematical Society, 2023

302/ Mathemagical Puzzles

Figures 3a and 3b: The Cutler-Rausch Melting Block 4

Notice that the 3 × 3 × 7 arrangement uses 63 unit bricks, and so it is presented
before the larger 4 × 4 × 4 arrangement. Similar to the original Melting Block
puzzle, we can construct a system of three variables in which a < b < c. Again, all
three variables change attitude between the two arrangements, so providing the
system:

7a + 1 = 4c;
3b + 1 = 4a;
3c + 1 = 4b.

Solving, we get a = 37, b = 49, c = 65; this gives the measurements of the unit
brick. The box must accommodate the 4×4×4 arrangement as size 148×196×260,
as well as the smaller 147 × 195 × 259 size of the 3 × 3 × 7 arrangement.
To create possible puzzle pieces, Cutler and Rausch discounted zig-zag and L-
shaped pieces, leaving only rectangular blocks. The limits on the measurements
are from a to 4a, b to 3b, and c to 3c in these respective dimensions. The numerical
values of these limits are, thus, one of {37, 74, 111, 148} by one of {49, 98, 147} by
one of {65, 130, 195}. This provides 36 possible blocks to consider. For those who
might consider duplicating the work of Cutler and Rausch, here are some computer
time requirements to expect:
For the 3 × 3 × 7 box, allowing the unit brick as a piece:
• 2.5 hours of run-time.
• Over 110 million basic solutions; this excludes rotations and reflections.
• Over 435 thousand piece combinations supplied the 110 million solutions.
• Just under 15 thousand piece combinations provide a unique solution.
For the 4 × 4 × 4 box, allowing a second copy of the unit brick:
• 12 hours of run-time.
• Over 385 million basic solutions.
• Over 987 thousand piece combinations supplied the 385 million solutions.
• Just under 59 thousand piece combinations provide a unique solution.

Crux Mathematicorum, Vol. 49(6), June 2023

 Tyler Somer /303

Comparing the unique solutions among the above results:

• 204 sets of pieces solve both boxes uniquely.
These sets are defined as giving doubly-unique solutions, and they can be split in
two categories:
• 106 sets of pieces use one unit brick in the 3 × 3 × 7 box, and a second unit
brick in the 4 × 4 × 4 box.
• 98 sets of pieces use no unit brick in the 3 × 3 × 7 box, but do use one in the
4 × 4 × 4 box.
Cutler and Rausch concentrated the remainder of their analysis on these last 98
results, thus concluding that a single unit brick should appear only in the “+ 1”
packing of the 4 × 4 × 4 arrangement. They found that only 29 of the 36 pieces
are used in these 98 piece sets.
They also found that a “duality” of the pieces – thus also a duality of the solutions
– exists. This duality exists by virtue of swapping the b and c dimensions of some
block. For example, the 111 × 49 × 130 block can also be represented by its
components (3a, b, 2c). Its dual is formed by the swap (3a, 2b, c), which is the
111 × 98 × 65 block. This duality is possible since both the b and c dimensions can
be at most 3 wide; whereas the a dimension can be up to 4 wide, so it cannot be
considered for a duality. Some pieces are self-dual; (4a, b, c) = 148 × 49 × 65, for
example, since it is only 1 wide in both the b and c dimensions. When considering
piece duality, the 98 sets of pieces, above, form 49 pairs of dual sets.
Cutler and Rausch also chose to dismiss the 148 × 147 × 65 piece, since its nearly-
square cross-section is not desired. Note that the larger nearly-square pieces,
148×147×130 and 148×147×195, were among the seven unused pieces, previously
removed from the analysis. Removing this eighth piece reduced the 98 sets with
doubly-unique solutions to 76. The piece dual of 148×147×65 is 148×49×195, and
this piece can be used in doubly-unique solutions. The 76 piece sets that remain
after this sifting include, therefore, 22 piece sets that use the 148 × 49 × 195 block
(but not the 22 piece sets that use its nearly-square dual), and 54 piece sets that
can be matched up as 27 dual pairs. The 152 unique packing problems range from
10 + 1 to 13 + 1 pieces.
As a conclusion to their analysis, Cutler and Rausch proposed a “multipuzzle”
using the 28 pieces and the list of 76 desired piece sets with doubly-unique packing
solutions. Among these, one piece set was selected as the Cutler Rausch Unique
Melting Block – CRUMB – for a puzzle design competition. A few multipuzzle
kits were produced in wood.
At the time of this writing, a metal version, named Melting Block 4, is available
for a mere $149 US from the Cubic Dissection online shop. It is not a multipuzzle
kit. It is a single puzzle designed for collectors. The box is walnut. The first 11
pieces are polished aluminum, and the + 1 unit is brass. Each CNC-machined and
hand-finished block is accurate to within two one-thousandths of an inch. The

Copyright © Canadian Mathematical Society, 2023

304/ Mathemagical Puzzles

polished blocks are stunning, with the appearance of a gold bar among silver bars.
This is several steps beyond the painted, yet more affordable, Red Stone version
available from a number of woodworkers.

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

When he was teaching, Tyler often had mechanical puzzles

in his classroom. As a freelancer, Tyler has worked with
numerous inventors and co-designers to bring dozens of
table-top solo-logic puzzle kits to market. He continues to
design puzzles, and he spends a good deal of time in his
woodshop, building his own and others’ puzzle designs.

Crux Mathematicorum, Vol. 49(6), June 2023

 Mathematics from the Web /305

MATHEMATICS FROM THE

WEB
No. 11
This column features short reviews of mathematical items from the internet that
will be of interest to high school and elementary students and teachers. You can
forward your own short reviews to [email protected].

Magic Squares of Squares (are PROBABLY impossible)

https://youtu.be/Kdsj84UdeYg
This video from Numberphile looks at 3 × 3 magic squares whose entries are all
perfect squares. This is a sequel to an earlier video called “The Parker Square”,
where mathematical author and comedian Matt Parker tries, unsuccessfully, to
create said magic square. He shows how some deep mathematics was used on this
problem and, even then, there still is no result (although it looks like there may
not be any). Interestingly, magic squares of squares are known for higher order
squares. The video shows one of order 4 created by Euler (who else?). This is an
interesting video to watch for the number theoretical hobbist and expert alike.

The Beauty of Quadratic Equations by Suhrid Saha

https://media.pims.math.ca/pi in sky/pi22.pdf
This short note looks at the development of the quadratic formula from a histor-
ical perspective. The author uses Al-Khwarizmi’s geometric method for solving
quadratic equations and uses it to develop the quadratic formula visually. It is a
nice piece of history focusing on a pivotal piece of mathematics in the high school
curriculum.

Mathematical Concepts Illustrated by Hamid Naderi Yeganeh

http://www.ams.org/publicoutreach/math-imagery/yeganeh
This page from the American Mathematical Society website features the artwork
of Hamid Naderi Yeganeh. Hamid uses mathematical functions to create stunning
images and each image is accompanied by the functions that generated it. In the
artists own words:
One of my goals is to create very beautiful images by using mathematical
concepts such as trigonometric functions, exponential function, regular
polygons, line segments, etc. I create images by running my program
on a Linux operating system.

Copyright © Canadian Mathematical Society, 2023

306/ OLYMPIAD CORNER

OLYMPIAD CORNER
No. 414
The problems in this section appeared in a regional or national mathematical Olympiad.

Click here to submit solutions, comments and generalizations to any

problem in this section

To facilitate their consideration, solutions should be received by August 15, 2023.

OC636. Determine all the pairs (p, n) of a prime number p and a positive
p
integer n for which npn +1
+1
is an integer.

OC637. For any positive integer x, we set

g(x) = the largest odd divisor of x,

® x x
2 + g(x) if x is even;
f (x) = x+1
2 2 if x is odd.

Consider the sequence (xn )n∈N defined by x1 = 1, xn+1 = f (xn ). Show that
the integer 2018 appears in this sequence, determine the last integer n such that
xn = 2018, and determine whether n with the property xn = 2018 is unique.

OC638. Find all the real numbers x such that

1 1 1
+ = {x} +
[x] [2x] 3

where [x] denotes the integer part of x and {x} = x − [x]. For example [2.5] = 2,
{2.5} = 0.5 and [−1.7] = −2, {−1.7} = 0.3.

OC639. For the curve sin(x) + sin(y) = 1 lying in the first quadrant, find the
constant α such that
d2 y
lim xα 2 .
x→0 dx
exists and is nonzero.

OC640. An equiangular hexagon has side lengths 1, 1, a, 1, 1, a in that order.

Given that there exists a circle that intersects the hexagon at 12 distinct points,
we have M < a < N for some real numbers M and N . Determine the minimum
N
possible value of the ratio M .

Crux Mathematicorum, Vol. 49(6), June 2023

 OLYMPIAD CORNER /307

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 15 août 2023.

OC636. Déterminer toutes les paires (p, n) telles que p est un nombre premier,
p
+1
n est un entier positif, et npn +1 est entier.

OC637. Pour tout entier positif x, on pose

g(x) = le plus grand diviseur impair de x,

® x x
2 + g(x) si x est pair;
f (x) = x+1
2 2 si x est impair.

Soit alors la suite (xn )n∈N définie par x1 = 1, xn+1 = f (xn ). Démontrer que
l’entier 2018 fait partie de cette suite, déterminer le plus grand entier n tel que
xn = 2018, puis déterminer si n tel que xn = 2018 est unique ou non.

OC638. Soit [x] la partie entière de x et {x} = x − [x]; par exemple, [2.5] = 2,
{2.5} = 0.5 et [−1.7] = −2, {−1.7} = 0.3. Déterminer tous les nombres réels x
tels que
1 1 1
+ = {x} + .
[x] [2x] 3

OC639. Pour la partie de la courbe sin(x) + sin(y) = 1 se situant dans le

premier quadrant, déterminer la constante α telle que

d2 y
lim xα
x→0 dx2
existe et est non nulle.

OC640. Les côtés d’un hexagone équiangulaire sont 1, 1, a, 1, 1, a, dans cet

ordre. Or, il existe un cercle qui rencontre cet hexagone en 12 points distincts, d’où
il en découle que M < a < N pour certains nombres réels M et N . Déterminer le
N
minimum du ratio M .

Copyright © Canadian Mathematical Society, 2023

308/ OLYMPIAD CORNER

OLYMPIAD CORNER
SOLUTIONS
Statements of the problems in this section originally appear in 2023: 49(1), p. 26–27.

OC611. Determine whether the following product is rational:

∞ Å
1
Y ã
1− .
2022k!
k=0

Originally Problem 1 from the 2022, Harvard MIT Invitational Competition.

We received 5 submissions, all of which were correct and complete. We present
the solution by Oliver Geupel.
The answer is no. To prove this, it is enough to show that, for every integer b ≥ 2,
the number
∞
Y
1 − b−k!


p =1/(b − 1) +
k=1
−1 −2 −3


= b +b +b + ...
+ (1 − b−1! − b−2! + b−1!−2! − b−3! + b−1!−3! + b−2!−3! − b−1!−2!−3! − b−4! ± . . . )

is irrational.
Let S be the set of positive integers that can be written as a sum of one or more
factorials of distinct positive integers, that is

S = {1!, 2!, 1! + 2!, 3!, 1! + 3!, 2! + 3!, 1! + 2! + 3!, 4!, 1! + 4!, 2! + 4!, 1! + 2! + 4!, . . .}.
For every positive integer n, we have
n
X
k! ≤ n · n! < (n + 1)!.
k=1

Hence, the sum representations of the members of S are unique up to the order
of terms. Therefore, the digits in base-b notation B = (1.d1 d2 d3 d4 . . . ) of p are
either 0 or 1 or 2. Furthermore, the digit dn at position of b−n is 1 if and only if
n ∈ N \ S.
Since the expansion B contains infinitely many occurrences of the digit 2, it is
infinite. For every positive number n, the consecutive integers
n
X
1+ k!, . . . , (n + 1)! − 1
k=1

Crux Mathematicorum, Vol. 49(6), June 2023

 OLYMPIAD CORNER /309

are not elements of S. Thus, B contains arbitrarily long (finite) segments that
consist of consecutive digits 1. Consequently, the infinite representation B is not
periodic. This shows that p is irrational.

OC612. Find all real numbers x such that

9 27 219
x9 + x6 + x3 − x + = 0.
8 64 512

Originally Problem 32 from the 2011 Harvard-MIT November Tournament, Guts

Round.
We received 12 submissions, of which 10 were correct and complete. We present
a typical solution.
We multiply both sides of the equation by 512 to eliminate the denominators

512x9 + (9 × 64)x6 + (27 × 8)x3 − 512x + 219 = 0,

and arrange the terms to get

(8x3 + 3)3 = 512x − 192,

 3 3  3
83 x3 + = 83 x − ,
8 8
 3 3
 3
x3 + =x− .
8 8
The real solutions of the equation above are the same as the real solutions of the
equation
1
3  33
x3 + = x − .
8 8

1
Consider the function f : R → R, f (x) = x3 + 38 then f −1 (x) = x − 38 3 . We
have to solve the equation f (x) = f −1 (x). Since f (x) and f −1 (x) are symmetrical
about the x axis, we get that f (x) = x gives us the required solutions. Note that
f (x) = x is equivalent to
3
x3 − x − =0 or (2x − 1)(4x2 + 2x − 3) = 0.
8
√ √
1 −1 + 13 −1 − 13
Hence the solutions are x = 2, x= , and x = .
4 4

OC613. Circles ω and Ω meet at points A and B. Let M be the midpoint

of the arc AB of circle ω (M lies inside Ω). A chord M P of circle ω intersects Ω
at Q (Q lies inside ω). Let lP be the tangent line to ω at P , and let lQ be the
tangent line to Ω at Q. Prove that the circumcircle of the triangle formed by the
lines lP , lQ , and AB is tangent to Ω.

Copyright © Canadian Mathematical Society, 2023

310/ OLYMPIAD CORNER

Originally Problem 5 from the 2014 Asian Pacific Mathematics Olympiad.

We received 5 submissions, of which 3 were correct and complete. We present the
solution by Oliver Geupel.
Let C = `Q ∩ AB, D = AB ∩ `P , E = `P ∩ `Q , F = AB ∩ M P , let G be the
midpoint of the line segment M P , and let O be the centre of ω. Also let c = DE,
d = EC, e = CD, and let tC , tD , and tE denote the lengths of the tangent segment
from C, D, and E, respectively, to Ω.

By the observation that ∠DP F = ∠GOM = ∠P F D, we have F D = P D. By the

power of point D with respect to circles Ω and ω, we obtain
√
tD = AD · BD = P D = F D.

For the triangle CDE and the transversal M G it holds by Menelaus’s theorem,
C P D QE
with unsigned distances, that FF D · P E · QC = 1. Hence,

dtD − etE − ctC = CE · F D − CD · QE − DE · QC

= CE(F C + CD) − CD(QC + CE) − DE · QC
= CE · F C − (CD + DE)QC
= F C(QC + CE) − (F C + CD + DE)QC
= F C · QE − P E · QC = 0.

A variation of the Power Theorem of Casey states that the circle through C, D,
and E is tangent to the circle Ω if and only if dtD ± etE ± ctC = 0 Hence the result.

Crux Mathematicorum, Vol. 49(6), June 2023

 OLYMPIAD CORNER /311

The variation of the Power Theorem of Casey is discussed together with a proof in
Johnson, Roger A. Advanced Euclidean Geometry, Dover Publications, Inc., 1960,
pp. 89-90, Theorem 117.

OC614. In a circus, there are n clowns who dress and paint themselves up
using a selection of 12 distinct colours. Each clown is required to use at least
five different colours. One day, the ringmaster of the circus orders that no two
clowns have exactly the same set of colours and no more than 20 clowns may use
any one particular colour. Find the largest number n of clowns so as to make the
ringmaster’s order possible.
Originally Problem 5 from the 2006 Asian Pacific Mathematics Olympiad.
We received 5 submissions, all of which were correct and complete. We present
the solution by Theo Koupelis.
The maximum number of times a distinct colour is used is 12 × 20, which must
be greater than the minimum number of colours used, which is 5 × n. Therefore,
n ≤ 48. It is possible to construct 48 distinct sets, each with 5 colours. For
example, if we number the colours as 1, 2, 3, . . . , 12, each of the 8 sets

{1, 2, 3, 4, 5, 6}, {3, 4, 5, 6, 7, 8, }, {5, 6, 7, 8, 9, 10}, {7, 8, 9, 10, 11, 12},

{9, 10, 11, 12, 1, 2}, {11, 12, 1, 2, 3, 4}, {1, 2, 3, 4, 7, 8}, {3, 4, 7, 8, 9, 10}
has 6 distinct subsets, each subset consisting of 5 colours; all 48 subsets are clearly
distinct. Therefore, nmax = 48.

OC615. Let ABC be a triangle. A circle intersects side BC at points U

and V , side CA at points W and X, and side AB at points Y and Z. The
points U, V, W, X, Y, Z lie on the circle in that order. Suppose that AY = BZ and
BU = CV . Prove that CW = AX.
Originally Problem 2 from the 2016 Australian Mathematical Olympiad.
We received 11 submissions, all of which were correct and complete. We present
a typical solution.

Copyright © Canadian Mathematical Society, 2023

312/ OLYMPIAD CORNER

Using the given equalities and the power of point with respect to the circle we
have

(AX)(AW ) = (AY )(AZ) = (BZ)(BY ) = (BU )(BV ) = (CV )(CU ) = (CW )(CX).

Thus
(AX)2 + (AX)(XW ) = (CW )2 + (CW )(XW ),
which implies (AX − CW )(AX + CW + XW ) = 0 and AX = CW .

Crux Mathematicorum, Vol. 49(6), June 2023

 Veselin Jungić /313

Hales-Jewett theorem through

examples and exercises: Part II
Veselin Jungić
“The ambition should always be to play an elegant game.”

Edson Arantes do Nascimento Pelé, 1940-2022

1 Introduction

In the first part of this two-part note we introduced the so-called combinatorial
lines.

Definition 1. Let A be an alphabet, let n ∈ N, and let τ ∈ An∗ be a root. A

combinatorial line in An rooted in τ is the set of words Lτ = {τa : a ∈ A}.

For reader’s convenience here are definitions of terms used in the definition of a
combinatorial line. In the rest of this note, for a natural number n we will denote
the set {1, 2, . . . , n} by [1, n].

1. For m ∈ N, any set A such that |A| = m is called an alphabet on m symbols.

2. Let A be an alphabet on m symbols. For n ∈ N, any function w : [1, n] → A

is called a word of length n on the alphabet A. If w(i) = ai , i ∈ [1, n], then
we write w = a1 a2 · · · an .

3. The set of all words of length n on the alphabet A is denoted by An and

called the n-dimensional cube on the alphabet A.

4. Let A be an alphabet (on m symbols) and let ∗ be a symbol such that ∗ 6∈ A.

We consider the alphabet A∗ = A ∪ {∗}. Any word on the alphabet A∗ , i.e.,
any element of (A∗ )n = An∗ , for some n ∈ N, that contains the symbol ∗ is
called a root.

5. For a root τ ∈ An∗ and a symbol a ∈ A we define the word τa ∈ An in the

following way. For i ∈ [1, n]
ß
τ (i) if τ (i) 6= ∗,
τa (i) =
a if τ (i) = ∗.

In this note our goal is to motivate and state the claim of the Hales-Jewett theorem
and to demonstrate a few of its applications.

Copyright © Canadian Mathematical Society, 2023

314/ Hales-Jewett theorem through examples and exercises: Part II

2 Generalized Tic-Tac-Toe Game

Recall the game of Tic-Tac-Toe: two players take turns claiming the spaces in a
3 × 3 grid with the goal to claim a row, a column, or a diagonal.
A group of undergraduate students created a
web game called Quad-Tac-Toe [1]. This is a
game of a player vs. AI on a 4 × 4 × 4 cube
Q(4, 3) = {(x, y, z) : x, y, z ∈ [1, 4]}. The player
and AI are given two different colours and tasked
to colour one point in Q(4, 3) at each turn. Who
first completes a monochromatic line wins.
Here, a line means a set of four collinear points, either horizontally, vertically, or
diagonally.
We illustrate the fact that each of the combinatorial lines in [1, 4]3 corresponds to
a winning position in the Quad-Tac-Toe† game by the following example.

Example 1. For a, b ∈ [1, 4], consider the root τ = ∗ a b ∈ [1, 4]3∗ and the
Euclidean line `τ in R3 given by its parametric equations x = t, y = a, z = b,
t ∈ R. Hence, `τ is the line that passes through the point (0, a, b) and is parallel to
the x-axis. Now, the combinatorial line Lτ = {1 a b, 2 a b, 3 a b, 4 a b} corresponds
to the winning position
{(1, a, b), (2, a, b), (3, a, b), (4, a, b)} = {(t, a, b) : t ∈ [1, 4]} ⊆ `τ ∩ Q(4, 3).

Observe that the line x = t, y = 5 − t, z = 1, t ∈ R, contains the winning position

{(1, 4, 1), (2, 3, 1), (3, 2, 1), (4, 1, 1)} = {(t, 5 − t, 1) : t ∈ [1, 4]} that does not corre-
spond to any of the combinatorial lines in [1, 4]3 .
Actually, there are 76 different Euclidean lines containing four points from Q(4, 3),
but only 61 of them correspond to combinatorial lines in the 3-dimensional cube
on alphabet [1, 4]. Hence, not every winning position in the Quad-Tac-Toe game
corresponds to a (monochromatic) combinatorial line. See Exercises 5 and 6 in
Part I.
But what if one considers a k-player game that is played on the “board” given by
Q(m, n) = {(x1 , x2 , . . . , xn ) : x1 , x2 , . . . , xn ∈ [1, m]}?‡ Similar to the Tic-Tac-Toe
game and the Quad-Tac-Toe game, each player is given one of k different colours
and tasked to colour one point in Q(m, n) at each turn. The player who first
completes a monochromatic line wins. Here “a line” means a set of m collinear
points, i.e., a set of m points in Q(m, n) that lie on a line in Rn given by its
parametric equations x1 = a1 + α1 · t, x2 = a2 + α2 · t, . . . , xn = an + αn · t, t ∈ R,
for some fixed real numbers ai , αi , i ∈ [1, n].
† In 1980, this variant of Tic-Tac-Toe was in detail analyzed by Oren Patashnik [8].
‡ Observe that (x1 , x2 , . . . , xn ) ∈ Q(m, n) is a point in Rn with x1 , x2 , . . . , xm ∈ [1, m].
With [1, m]n = {x1 x2 · · · xn : x1 , x2 , . . . , xn ∈ [1, m]} we denote the n-cube on the alphabet
[1, m].

Crux Mathematicorum, Vol. 49(6), June 2023

 Veselin Jungić /315

Graham, Rothschild, and Spencer called the above generalization of Tic-Tac-Toe

a “k-person n dimensional Tic-Tac-Toe m-in a row” game [3].

Proposition 1. Any combinatorial line in [1, m]n corresponds to a winning posi-

tion in a k-person n dimensional Tic-Tac-Toe m-in a row game.

Proof. For given m, n ∈ N, we consider a root τ = a1 a2 . . . an ∈ [1, m]n∗ . Let the

line `τ in Rn be given by its parametric equations

x1 = b1 + α1 · t, x2 = b2 + α2 · t, . . . , xn = bn + αn · t, t ∈ R,

where bi = 0 and αi = 1 if ai = ∗ , and bi = ai and αi = 0 if ai ∈ [1, m].

Lτ `τ
(m) (m) (m) (m)
τm = x1 ··· xn (x1 , . . . , xn )

(i) (i) (i) (i)

τi = x1 · · · xn (x1 , . . . , xn )

(2) (2) (2) (2)

τ2 = x1 · · · xn (x1 , . . . , xn )

(1) (1) (1) (1)

τ1 = x1 · · · xn (x1 , . . . , xn ) Q(m, n)
n
[1, m] Rn

(i) (i) (i)

Recall that for i ∈ [1, m], the word τi = x1 x2 · · · xn ∈ [1, m]n is such that
(i) (i)
xj = i if aj = ∗ and xj = aj if aj ∈ [1, m].
It follows that, by taking t = i in the parametric equation for `τ , the point
(i) (i) (i)
(x1 , x2 , . . . , xn ) ∈ `τ ∩ Q(m, n).
This establishes an injection between the combinatorial line Lτ = {τi : i ∈ [1, m]}
and the set `τ ∩ Q(m, n). Since |Lτ | = m, the combinatorial line Lτ corresponds
to “m-in a row” points in Q(m, n), i.e., to a winning position.

3 The Hales-Jewett Theorem

Suppose that a k-person n dimensional Tic-Tac-Toe m-in a row game ended up

in a tie. Hence, the set Q(m, n) was partitioned into k mutually disjunct parts,
Q(m, n) = P1 ∪ . . . ∪ Pk , Pi ∩ Pj = ∅ if i 6= j, in such a way that none of the parts
contained a set of m collinear points.

Copyright © Canadian Mathematical Society, 2023

316/ Hales-Jewett theorem through examples and exercises: Part II

Define a k-colouring C of the cube [1, m]n as following: for a1 · · · an ∈ [1, m]n ,
C(a1 · · · an ) = i if and only if (a1 , . . . , ak ) ∈ Pi . By Proposition 1 and our as-
sumption that the game ended up in a tie, it follows that the colouring C of the
n-dimensional cube on alphabet [1, m] does not contain a monochromatic combi-
natorial line.
In the spirit of Ramsey theory, this observation leads us to the following question.

Question 1. Let A be an alphabet on m symbols and let An be the n-dimensional

cube on alphabet A, i.e., let An = {a1 a2 · · · an : ai ∈ A, i ∈ [1, n]}. If An is k-
coloured, under which conditions can we be sure that An contains a monochromatic
combinatorial line?
More precisely, let k, m ∈ N and let A be an alphabet on m symbols. Does there
exist an n ∈ N such that whenever An is k-coloured there exists a monochromatic
line?

♣ ♥ ♠ ... ♥
♣ ♣ ♠ ... ♥
♣ ♠ ♠ ♥
♣ ♦ ♠ ... ♥
{♥, ♣, ♠, ♦}n is k–coloured. . . . . . and a red combinatorial line.

A nice answer to Question 1 would be that, for any k, m ∈ N and for a big
enough natural number n, the k-person n dimensional Tic-Tac-Toe m-in a row
game cannot end up in a tie.
This is exactly what Hales and Jewett discussed in their paper Regularity and posi-
tion games, published in 1963 [6]. Their famous Hales-Jewett theorem establishes
that, if the dimension is sufficiently large, a generalized Tic-Tac-Toe game never
ends up in a tie.

Theorem 1 ([6]). Let k, m ∈ N and let A be an alphabet on m symbols. There

exists an n ∈ N such that whenever An is k-coloured there exists a monochromatic
combinatorial line.

Definition 2. The smallest n guaranteed by Theorem 1 is denoted by H(k; m)

and called a Hales-Jewett number.

Exercise 1. Let A = {a, b, c, d}. Can you find a 2-colouring of A2 that does not
contain a monochromatic combinatorial line? If yes, does this contradict the claim
of the Hales-Jewett theorem?

Crux Mathematicorum, Vol. 49(6), June 2023

 Veselin Jungić /317

Exercise 2. Let A = [1, 3] be an alphabet. Check if the 2-colouring depicted in

the figure below yields a monochromatic combinatorial line in A3 . Based on your
observation, what can you tell about the Hales-Jewett number H(2; 3)?
  (3, 3, 3)


(1, 1, 3)


 (3, 3, 2)

 

 (3, 3, 1)

 (3, 2, 1)

 
(1, 1, 1) (2, 1, 1) (3, 1, 1)

Exercise 3. Prove that, for r ≥ 2, HJ(r; 2) = r.

Example 2. Use the Hales-Jewett theorem to prove van der Waerden’s theorem:
If k, l ∈ N then any l-colouring of N contains a k-term monochromatic arithmetic
progression.
Solution. Let k, l ∈ N be given. Let c : N → [1, l] be an l-colouring of the set of
natural numbers. Let N = HJ(l; k). We define an l-colouring of the N -cube [1, k]N
as follows: if x1 x2 · · · xN ∈ [1, k]N then c0 (x1 x2 · · · xN ) = c(x1 + x2 + · · · + xN ).
By the Hales-Jewett theorem there is a c0 -monochromatic combinatorial line Lτ
rooted in a root τ = a1 a2 · · · aN ∈ [1, k]N
∗ . We observe that there is at least one
i ∈ [1, N ] such that ai = ∗.
Let S be the set of all i ∈ [1, N ] such that ai ∈ [1, k], i.e., the symbol ai 6= ∗.
Recall that the combinatorial line Lτ is the set of words of the form, for j ∈ [1, k],
(j) (j) (j) (j) (j)
τj = a1 a2 · · · aN , with ai = ai if i ∈ S and ai = j if i 6∈ S.
Let a = i∈S ai and let d = |[1, N ]\S|, i.e., let d ≥ 1 be the number of times that
P
the symbol ∗ appears in the root τ . Note that, for j ∈ [1, k],
N
(j) (j) (j)
X X X
ai = ai + ai = a + dj.
i=1 i∈S i∈[1,N ]\S

On the other hand,Äc0 (τ1 ) = cä0 (τ2 ) = · · · = c0 (τk ) which together with, for each
PN (j)
j ∈ [1, k], c0 (τj ) = c i=1 ai = c(a + jd), implies

c(a + d) = c(a + 2d) = · · · = c(a + kd).

Hence, the k-term arithmetic progression a+d, a+2d, . . . , a+kd is c-monochromatic.

Copyright © Canadian Mathematical Society, 2023

318/ Hales-Jewett theorem through examples and exercises: Part II

Exercise 4. Use the Hales-Jewett theorem to prove that for any 2-colouring of
natural numbers there is an `-term arithmetic progression a1 , a2 , . . . , a` , not neces-
sarily monochromatic, such that the set A = {2a1 , 2a2 , . . . , 2a` } is monochromatic.

Exercise 5 (Gallai’s theorem for semigroups). Tibor Gallai, a Hungarian mathe-

matician, 1912-1992, was a lifelong friend and collaborator of Paul Erdős. In this
exercise we use the Hales-Jewett theorem to prove Gallai’s theorem for semigroups.
Let (A, •) be a semigroup and let ` ∈ N be given. For any r-colouring of the set A
there are a, b ∈ A such that the set {a, a • b, a • b2 , . . . , a • b`−1 } is monochromatic.
Note: A semigroup (A, •) is an algebraic structure consisting of a set A together
with an associative binary operation •. Commonly, x • y, denotes the result of
applying the semigroup operation to the ordered pair (x, y). For example, both
(N, +) and (N, ·), i.e., the set of natural numbers together with the usual addition
and multiplication, are semigroups. Associativity for all x, y, z ∈ A is expressed as
(x • y) • z = x • (y • z). For k ∈ N and x ∈ A we write

xk = x • (x • (x • (. . . (x • x) . . .)) = x
| • x •{z
. . . x • x} .
| {z }
k k

4 The Hales-Jewett Theorem and the Polymath

Project
Similar to Ramsey’s theorem and van der Waerden’s theorem, the Hales-Jewett
theorem was one of those significant mathematical results that became a source
of inspiration for generations of mathematicians.
Here we briefly reflect on one of the developments, closely related to the Hales-
Jewett theorem, that have enriched the whole 21st century’s mathematical land-
scape.
In 1983, Ronald Graham offered $1,000 for a proof of what he called, a “density
version for the Hales-Jewett theorem” (DHJ):
For all finite A and ε > 0 there exists N (A, ε) such that if N ≥ N (A, ε)
and R ⊆ AN satisfies |R| ≥ ε|AN | then R must contain a combinatorial
line [4].
In simple terms, Graham’s question was if any large enough subset of a cube of
the appropriate dimension must contain a combinatorial line.
ln 1991, Hillel Furstenberg and Yitzhak Katznelson proved Graham’s conjecture
from 1983 by using the ergodic theory techniques [2].
In early 2009, in a series of blog posts, Timothy Gowers invited members of the
mathematical community to jointly search for an elementary proof of DHJ. Gow-
ers’s initial blog titled Is massively collaborative mathematics possible? [5] marks

Crux Mathematicorum, Vol. 49(6), June 2023

 Veselin Jungić /319

the beginning of the Polymath Project, a still ongoing collaboration of mathemati-

cians across the world on a variety of important mathematical problems.
For all of those who study, teach, and do mathematics, here are two of the “ground
rules” of the Polymath Project that Gowers established in 2009 [5]:
“3. When you do research, you are more likely to succeed if you try
out lots of stupid ideas. Similarly, stupid comments are welcome here.
(In the sense in which I am using “stupid,” it means something com-
pletely different from “unintelligent.” It just means not fully thought
through.)
5. Don’t actually use the word “stupid,” except perhaps of yourself.”
The first Polymath Project was a success. In 2012, the first elementary proof of
DHJ, together with a quantitative bound on how large n needs to be, was published
[9]. The collaborators fittingly attributed “D.H.J. Polymath” as the author.

5 Hints and solutions

Exercise 1. There is no monochromatic combinatorial line in this red/black
colouring of the cube A2 :
∗a ∗b ∗c ∗d a∗ b∗ c∗ d∗ ∗∗
aa ab ac ad aa ba ca da aa
ba bb bc bd ab bb cb db bb
ca cb cc cd ac bc cc dc cc
da db dc dd ad bd cd dd dd
The existence of such a colouring does not contradict the Hales-Jewett theorem.
It just shows that HJ(2; 4) > 2.
Exercise 2. There is no monochromatic combinatorial line in this colouring.
Hence, HJ(2; 3) > 2. Note: Observe that {(1, 3, 1), (2, 2, 1), (3, 1, 1)} is a monochro-
matic set that contains three points that are collinear in R3 . Does this set corre-
spond to a combinatorial line in A3 ?
Exercise 3. Let r ≥ 2 and let A = {0, 1} be an alphabet. Let 2 ≤ n < r
and let c : An → [0, r − 1] be an r-colouring defined in the following way: If
w = a1 a2 · · · an then c(w) = |{i ∈ [1, n] : ai = 0}|. Let τ ∈ An∗ be any root. From
c(τ0 ) 6= c(τ1 ), it follows that the combinatorial line Lτ is not c-monochromatic.
Hence, HJ(r; 2) ≥ r.
(i) (i) (i)
Let n ≥ r. For i ∈ [1, n + 1], let the word wi = a1 a2 . . . an be such that
(i) (i)
aj = 0, if j < i, and aj = 1, if j ≥ i. Let c be any r-colouring of the cube
n
A . By the pigeonhole principle, there are p, q ∈ [1, n + 1], p < q, such that
. . . 0} ∗| ∗{z
c(wp ) = c(wq ). We consider the root τ = 0| 0 {z . . . ∗} 1| 1 {z
. . . 1} and observe
p−1 q−p n−q+1
that the combinatorial line Lτ = {τ0 = wq , τ1 = wp } is c-monochromatic.

Copyright © Canadian Mathematical Society, 2023

320/ Hales-Jewett theorem through examples and exercises: Part II

Note: Only in 2014, Neil Hindman and Eric Tressler established that HJ(2; 3) = 4
and obtained, what they called, “the first non-trivial Hales-Jewett number” [7].
Exercise 4. Let N = HJ(2; `) and let c be a 2-colouring of the set of nat-
ural numbers. We define a 2-colouring of the N -cube [1, `]N as follows: for
x1 x2 · · · xN ∈ [1, `]N , c0 (x1 x2 · · · xN ) = c (2x1 · 2x2 · . . . · 2xN ). By the
Hales-Jewett theorem, there is a c0 -monochromatic line Lτ , determined by a root
τ = a1 a2 · · · aN ∈ [1, `]N
∗ . Let S = {i ∈ [1, N ] : ai ∈ [1, `]}. Let a = i∈S ai and
P
let d = |[1, N ]\S|.
(j) (j) (j)
Recall that Lτ = {τ1 , τ2 , . . . , τ` } ⊆ [1, `]N , where, for j ∈ [1, `], τj = a1 a2 · · · aN ,
(j) (j)
with ai = ai , if i ∈ S, and ai = j, if i 6∈ S.
Note that, for any j ∈ [1, `],
N
(j) (j) (j)
X X X X
ai = ai + ai =a+ j = a + jd.
i=1 i∈S i∈[1,N ]\S i∈[1,N ]\S

On
 Pthe other hand, c0 (τ1 ) = c0 (τ2 ) = · · · = c0 (τ` ), which together with c0 (τj ) =
N (j)
c 2 i=1 ai = c 2a+jd , for each j ∈ [1, `], implies that the `-term arithmetic

progression a1 = a + d, a2 = a + 2d, . . . , a` = a + `d is with the required property.

Exercise 5. Let c be an r-colouring of A and let N = HJ(r; `). Let x ∈ A be
fixed. We define an r-colouring of the N -cube [1, `]N as follows: for n1 n2 · · · nN ∈
[1, `]N , c0 (n1 n2 · · · nN ) = c (xn1 +n2 +···+nN ).
By the Hales-Jewett theorem, there is a c0 -monochromatic line Lτ , determined
by a root τ = a1 a2 · · · aN ∈ [1, `]N
∗ . Let S = {i ∈ [1, N ] : ai ∈ [1, `]}. Let
m = i∈S ai and let d = |[1, N ]\S|.
P

(j) (j) (j)

Recall that Lτ = {τ1 , τ2 , . . . , τ` } ⊆ [1, `]N , where, for j ∈ [1, l], τj = a1 a2 · · · aN ,
(j) (j)
with ai = ai , if i ∈ S, and ai = j, if i 6∈ S. Note that for any j ∈ [1, `],
N
(j) (j) (j)
X X X X
ai = ai + ai =a+ j = a + jd.
i=1 i∈S i∈[1,N ]\S i∈[1,N ]\S

On the other handc0P (τ1 ) = c0 (τ2 ) = · · · = c0 (τ` ) which implies that, for each
N
0 τ (i)
= c xm+jd .


j ∈ [1, `], c (τj ) = c x i=1 j

Observe that, since binary operation • is associative, it follows, for each j ∈ [1, `],
xm+jd = | • .{z
|x • x •{z. . . • x} = (x . . • x}) • (x
| • .{z
. . • x})
m+jd m+d (j−1)d

j−1
= | • .{z
(x . . • x}) • ((x
| • .{z
. . • x}) • . . . • (x . . • x})) = xm+d · xd
| • .{z .
m+d d d
| {z }
j−1

Crux Mathematicorum, Vol. 49(6), June 2023

 Veselin Jungić /321

Therefore, for a = xm+d and b = xd , the set {a, a • b, a • b2 , . . . , a • b`−1 } is

c-monochromatic.
Note: Observe that, in the case of the semigroup (N, +), Gallai’s theorem for
semigroups implies van der Waerden’s theorem. Also, observe that the set of all
non-negative integer powers of 2, with the usual multiplication, is a semigroup.

References
[1] J. Andrews, A. Blondal, T. Trinh, and H. Zhang. Quad-Tac-Toe game. In
V. Jungić, editor, Introduction to Ramsey Theory: Students’ Projects. Simon
Fraser University, 2022.
[2] H. Furstenberg and Y. Katznelson. A density version of the Hales-Jewett the-
orem. Annals of Mathematics, 57: 64-119, 1991.
[3] R. L. Graham, B. Rothschild, and J. H. Spencer. Ramsey Theory. John Wiley
and Sons, New York, 2nd edition, 1990.
[4] R.L. Graham. Recent developments in Ramsey theory. In Z. Cisielski and C.
Olech, editors, Proceedings of the International Congress of Mathematicians,
Warsaw, 1983(2): 1555-1569, Amsterdam, 1984. North Holland.
[5] W.T. Gowers. Is massively collaborative mathematics possible? Gowers’s We-
blog, January 2009.
[6] A.W. Hales and R.I. Jewett. Regularity and positional games. Transactions of
the American Mathematical Society, 106(2): 222-229, 1963.
[7] N. Hindman and E. Tressler. The first nontrivial Hales-Jewett number is four.
Ars Combinatoria, 113: 385-390, 2014.
[8] O. Patashnik. 4 × 4 × 4 Tic-Tac-Toe. Mathematics Magazine, 53(4): 202-216,
1980.
[9] D.H.J. Polymath. A new proof of the density Hales-Jewett theorem. Annals of
Mathematics, 175: 1283-1327, 2012.

Copyright © Canadian Mathematical Society, 2023

322/ Problems

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

To facilitate their consideration, solutions should be received by August 15, 2023.

4851. Proposed by Mihaela Berindeanu.

Let Γ be the circumcircle of 4ABC, with circumcenter O and radius R. Point X

is diametrically opposed to A, AX ∩ BC = {P } andAP = 2P X. The tangent to
the circle Γ through X cuts AB in M and AC in N . Show that

2R · M N + OM · CN + ON · BM = M C · ON + BN · OM.

4852. Proposed by Aravind Mahadevan.

In triangle ABC, the bisectors of angles A, B and C meet the sides BC, CA and
AB at D, E and F respectively. If ∠ADC = x, ∠AEB = y and ∠BF C = z,
prove that a sin 2x + b sin 2y + c sin 2z = 0 where a, b and c are the lengths of BC,
CA and AB respectively.

4853. Proposed by Byungjun Lee.

Two congruent ellipses Γ1 and Γ2 with semi-major axis a and semi-minor axis b
are given. The major axis of Γ1 and the minor axis of Γ2 lie on the same line, and
two common internal tangents of Γ1 and Γ2 are perpendicular. Find the area of
the triangle formed by two common internal tangents and one common external
tangent.

Crux Mathematicorum, Vol. 49(6), June 2023

 Problems /323

4854. Proposed by Michel Bataille.

kπ
Let n be a positive integer and let θk = n+1 . For r, s ∈ {1, 2, . . . , n}, evaluate
n
X
(sin θjr + sin θjs )2 .
j=1

4855. Proposed by Ivan Hadinata.

Find all pairs of positive integers (a, b) such that ab − ba = a − b.

4856. Proposed by Titu Zvonaru.

Let ABC be a triangle with ∠A = 30◦ and ∠B = 100◦ . Consider the points D
and E on the sides AC and BC, respectively, such that ∠ABD = ∠DBC and
DE||AB. Find ∠EAC.

4857. Proposed by Toyesh Prakash Sharma.

3
Let a, b, c be positive real numbers such that a + b + c = . Show that
2
3
aa bb + bb cc + cc aa ≥ .
2

4858. Proposed by Anton Mosunov.

Prove that for every positive integer n and for all α1 , α2 , . . . , αn ∈ R, we have
Ã
Z π Y n
n
csc2 (x − αk )dx ≥ 4π.
0 k=1

4859. Proposed by Trinh Quoc Khanh, modified by the Editorial Board.

Given a triangle ABC, a point X on segment AB and a point Y on segment
AC, such that B, X, Y , C are concyclic, let I, J, K be the incenters of triangles
ABC, XBC, and Y BC, respectively. Prove that AI is orthogonal to JK.

4860. Proposed by George Apostolopoulos.

Let ABC be a triangle with ∠A > 90◦ . Let M1 , M2 , . . . , Mn (n ≥ 1) be internal
points on the side BC such that BM1 = M1 M2 = · · · = Mn−1 Mn = Mn C. Prove
that  
2n + 1
AM1 + AM2 + · · · + AMn < n BC.
6(n + 1)

Copyright © Canadian Mathematical Society, 2023

324/ Problems

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

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

Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au plus tard le 15 août 2023.

4851. Soumis par Mihaela Berindeanu.

Soit Γ le cercle circonscrit à 4ABC, de centre O et de rayon R. Le point Xest

diamétralement opposé à A, AX ∩ BC = {P } et AP = 2P X. La tangente au
cercle Γ passant par X coupe AB en M et coupe AC en N . Montrez que

2R · M N + OM · CN + ON · BM = M C · ON + BN · OM.

4852. Soumis par Aravind Mahadevan.

Dans le triangle ABC, les bissectrices des angles A, B et C rencontrent les côtés
BC, CA et AB en D, E et F respectivement. Si ∠ADC = x, ∠AEB = y
et ∠BF C = z, montrez que a sin 2x + b sin 2y + c sin 2z = 0, où a, b et c sont
respectivement les longueurs de BC, CA et AB.

4853. Soumis par Byungjun Lee.

Deux ellipses congruentes Γ1 et Γ2 de demi-grand axe a et demi-petit axe b sont

données. Le grand axe de Γ1 et le petit axe de Γ2 se trouvent sur la même ligne
et deux tangentes internes communes de Γ1 et Γ2 sont perpendiculaires. Trouvez
l’aire du triangle formé par deux tangentes internes communes et une tangente
externe commune.

Crux Mathematicorum, Vol. 49(6), June 2023

 Problems /325

4854. Soumis par Michel Bataille.

kπ
Soit n un entier positif et soit θk = n+1 . For r, s ∈ {1, 2, . . . , n}. Évaluez
n
X
(sin θjr + sin θjs )2 .
j=1

4855. Soumis par Ivan Hadinata.

Trouvez toutes les paires d’entiers positifs (a, b) telles que ab − ba = a − b.

4856. Soumis par Titu Zvonaru.

Soit ABC un triangle avec ∠A = 30◦ et ∠B = 100◦ . Considérons les points D et
E sur les côtés AC et BC, respectivement, tels que ∠ABD = ∠DBC et DE||AB.
Trouvez ∠EAC.

4857. Soumis par Toyesh Prakash Sharma.

3
Soient a, b et c des nombres réels positifs tels que a + b + c = . Montrez que
2
3
aa bb + bb cc + cc aa ≥ .
2

4858. Soumis par Anton Mosunov.

Montrez que pour tout entier positif n et pour tout α1 , α2 , . . . , αn ∈ R, on a
Ã
Z π Y n
n
csc2 (x − αk )dx ≥ 4π.
0 k=1

4859. Proposée par Trinh Quoc Khanh, modifié par le comité de rédaction.
Pour un triangle donné ABC, soient X un point situé sur le segment AB et Y un
point situé sur le segment AC, de façon à ce que B, X, Y et C soient cocycliques;
les centres des cercles inscrits des triangles ABC, XBC et Y BC sont alors dénotés
I, J et K, respectivement. Démontrer que AI est orthogonal à JK.

4860. Proposée par George Apostolopoulos.

Soit ABC un triangle tel que ∠A > 90◦ . Soient aussi M1 , M2 , . . . , Mn (n ≥ 1) des
points internes du côté BC, tels que BM1 = M1 M2 = · · · = Mn−1 Mn = Mn C.
Démontrer que
 
2n + 1
AM1 + AM2 + · · · + AMn < n BC.
6(n + 1)

Copyright © Canadian Mathematical Society, 2023

326/ Solutions

SOLUTIONS
No problem is ever permanently closed. The editor is always pleased to consider for
publication new solutions or new insights on past problems.
Statements of the problems in this section originally appear in 2023: 49(1), p. 44–47.

4801. Proposed by Michel Bataille.

Find all functions f : (0, ∞) → R such that

1
Å ã
f x+ = yf (xy + y)
y

for all x, y > 0.

We received 17 submissions, 10 of which are correct. We present 2 different solu-
tions.
Solution 1, by the UCLan Cyprus Problem Solving Group.
t−1 2
For t > 1, letting x = 2 and y = t+1 we have xy + y = 1 and so

2f (1)
f (t) = .
t+1
Now for any r > o, letting x = r/2, y = 2/r we have xy + y = 1 + y > 1 and so

2yf (1) 2f (1) 2f (1)

f (r) = yf (1 + y) = = = .
y+2 1 + 2/y r+1

It follows that there is a constant k = 2f (1) such that f (x) = k/(x + 1) for every
x > 0.
It is easy to check that every such function satisfies the functional equation.

Solution 2, by Theo Koupelis.

Let P (x, y) be the statement that

1
Å ã
P (x, y) : f x+ = yf (xy + y), ∀x, y > 0.
y

We have
1 x+2
Å ã Å ã
x x
P , : f · f (1),
=
x x+1 x x+1
x 2 2 x+2
Å ã Å ã
P , : f (x) = · f .
2 x x x

Crux Mathematicorum, Vol. 49(6), June 2023

 Solutions /327

From the above equations we get

2
f (x) = · f (1),
x+1
with f (1) ∈ R, which clearly satisfies the given equation.

4802. Proposed by Arkan Manva.

Suppose there are 2n students in a class, each starting with a certain number of
sweets, such that the total number is a multiple of n. On each day, the teacher
chooses n students and gives c ∈ N sweets to each of them. Find the minimum
number of days needed such that no matter how the sweets were distributed at
the start, the teacher can leave all the students with an equal number of sweets.
(Note: c can change on a daily basis).
On top of the proposer’s solution, we only received the submission by UCLan
Cyprus Problem Solving Group and it was correct. We present their solution,
with some modifications and clarification by the editor based on the solution of the
proposer.
We show that the minimum number of days needed is 2n − 1.
(1) We first show that 2n − 1 days are sufficient. It will be convenient to assume
that the teacher can also subtract sweets from the students (so c is allowed to be
negative). This will not have any impact on the minimum number of days needed
because instead of subtracting sweets from a set of n students we can actually give
the same number of sweets to the complementary set of n students. We pick a
special child, say with y sweets. Let S be the total number of sweets at the start.
Label the (2n − 1) non-special children by 1, . . . , 2n − 1. For each 1 ≤ i ≤ 2n − 1,
let ai be the number of sweets the child i has. All indices in the following are
taken modulo 2n − 1. For each 1 ≤ j ≤ 2n − 1, consider the interval Ij =
{j, j + 1, . . . , j + n − 1}. Our strategy is ”cyclic”, namely, give xj sweets to each
child with label in Ij on the j-th day, so that eventually all children have y sweets,
the same as the special child. Using this strategy, the number of sweets that child
i has will become
X i
X
a0i := ai + xj = ai + xj .
i∈Ij j=i−n+1
1≤j≤2n−1

We need to guarantee that a0i are all the same, that is, a0i = a0i+1 for each 1 ≤ i ≤
2n − 1. Given that a0i = a0i+1 , we need
i
X i+1
X
ai + xj = ai+1 + xj
j=i−n+1 j=i−n+2

equivalently,
ai + xi−n+1 = ai+1 + xi+1 . (1)

Copyright © Canadian Mathematical Society, 2023

328/ Solutions

Note that gcd(2n − 1, n) = 1. Thus, given x0 , we can use (1) to inductively

determine xn , x2n , etc, and eventually xn(2n−1) , so that a01 = a02 = · · · = a02n−1 .
Note that if we sum over equation (1) over all 1 ≤ i ≤ 2n − 1, we get 0 = 0. This
shows that x0 = xn(2n−1) is well-defined. Thus, it remains to determine x0 so that
a00 = y. We claim that choosing
S
x0 = − an − a2n−1
n
would work. Note that
n−1
X 2n−1
X
a0n = an + xj , a02n−1 = a2n−1 + xj .
j=0 j=n

Thus, if a00 = z, then we have

2n−1
X (2n − 1)z − (S − y)
2z = a0n +a02n−1 = an +a2n−1 +x0 + xj = an +a2n−1 +x0 + ,
j=1
n

which implies that z = y. More explicitly, we can take

S
xi = − ai − ai+n .
n

(2) Next we show that 2n − 2 days are not enough using linear algebra. For the
sake of contradiction, suppose otherwise that 2n − 1 days work.
Consider the vector space V = Q2n . Let 1 be the all-one vector in V . For each
subset A ⊂ {1, 2, . . . , 2n}, let 1A be the indicator vector of A.
Let

U = {u = (a1 , . . . , a2n ) ∈ Z2n : ai ≥ 0, n | (a1 + a2 + · · · + a2n )},

that is, the collection of vectors representing the number of sweets the children
have at the start. Note that U ∪ {1} spans V . Indeed, if v = (v1 , . . . , v2n ) ∈ V ,
then there is a sufficiently large λ ∈ Q such that v + λ · 1 consists of positive
entries, and there is a sufficiently large integer M such that M (v + λ · 1) ∈ U .
Let u ∈ U . By the assumption, there are integers c1 , c2 , . . . , c2n−2 and n-subsets
A1 , A2 , . . . , A2n−2 of {1, 2, . . . , 2n}, such that if ci sweets are given to children with
label in Ai on the i-th day, then eventually all children obtain the same number
of sweets. Alternatively, this means that

u + c1 · 1A1 + · · · + c2n−2 · 1A2n−2 = λ · 1

for some integer λ.

Therefore, [
U⊆ span{1A1 , 1A2 , . . . , 1A2n−2 , 1},
A1 ,A2 ,...,A2n−2

Crux Mathematicorum, Vol. 49(6), June 2023

 Solutions /329

where the union is taken over all collections of n-subsets A1 , A2 , . . . , A2n−2 of

{1, 2, . . . , 2n}. It follows that
[
Q2n = span(U ∪ {1}) ⊆ span{1A1 , 1A2 , . . . , 1A2n−2 , 1},
A1 ,A2 ,...,A2n−2

that is, Q2n can be covered by finitely many subspaces with dimension at most
2n − 1, which is absurd.

Editor’s Comment. We refer to this Math Stackexchange post for a relevant dis-
cussion on covering a vector space by proper subspaces.

4803. Proposed by Nguyen Viet Hung.

Find all non-negative integers a, b, c and pairs (p, q) of prime numbers satisfying

p2a + q 2b = (2c + 1)2 .

We received 27 submissions, only 9 of which were completely correct. We present

Oliver Geupel’s solution.
By inspection, (1, 2, 2, 3, 2) and (2, 1, 2, 2, 3) are solutions for (a, b, c, p, q). We prove
that there are no further solutions. Assume that a, b, c, p, and q satisfy the given
conditions. By inspection, we see that a and b are positive. Since (2c + 1)2 is odd,
exactly one of the primes p and q is even, that is, equal to 2. By the symmetry
(a, p) ↔ (b, q) it is enough to consider the case q = 2. We have

4b = (2c + 1 − pa ) (2c + 1 + pa ) .

Hence there is an integer m where 1 ≤ m < b such that 2c + 1 − pa = 2m and

2c + 1 + pa = 22b−m . Then

2pa = 22b−m − 2m = 2m 4b−m − 1 .

We obtain m = 1 and
pa = 2b−1 − 1 2b−1 + 1 .

Therefore, pa has two divisors that differ by 2. Thus, p = 3, a = 1, b = 2, and

c = 2.
Editor’s Comments. There were two unfortunately common errors: 10 solvers
found only one solution, and 12 solvers overlooked the possibility that a or b is 0
when deducing p and q have opposite parity (consider 20 + 21 ).
Most solvers gave some variation of the elementary argument above; some noted
that (pa , q b , 2c + 1) is a Pythagorean triple. B. Roy was unique in reducing to
Catalan’s equation. S. Dutta pointed out this problem generalizes Problem 8 on
the 1992 Indian National Mathematical Olympiad.

Copyright © Canadian Mathematical Society, 2023

330/ Solutions

4804. Proposed by Jimmy Zhao.

Let ABC be an acute triangle with incenter I and circumcenter O. Let D be a

point on BC such that AI ⊥ ID. Let N be the midpoint of minor arc BC. DA
and DN meet the circumcircle of ABC at E, F respectively. Let EF meet AI at
G. Show that OI ⊥ DG.

We received 8 correct solutions. We present the solution by Michel Bataille.

Let Γ and Γ0 be the circumcircles of 4ABC and 4BIC, respectively. We recall

that N is on the line AI and that Γ0 is centered at N . The point D, which is on
the radical axis BC of Γ and Γ0 , has the same power p = DB · DC with respect
to these two circles. Since DI ⊥ IN , the line DI is tangent to Γ0 at I, hence
p = DI 2 . We also have p = DE · DA = DF · DN . Therefore, if I denotes the
inversion in the circle with center D and radius DI, we have I(I) = I, I(B) = C,
I(A) = E, I(N ) = F .

Since the circle Γ is its own inverse, the point U = I(O) is the foot of the polar
of D with respect to this circle. But the quadrilateral AEF N is inscribed in Γ
and AE,N F intersect at D, hence the lines AN and EF intersect on the polar of
D. It follows that this polar is the perpendicular to DO through G and intersects
DO at U .

Now, ∠DU G = ∠DIG = 90◦ , hence U = I(O) and I = I(I) are on the circle
with diameter DG. Thus, this circle is the inverse of the line IO and therefore
DG ⊥ IO, as desired.

Crux Mathematicorum, Vol. 49(6), June 2023

 Solutions /331

4805. Proposed by Goran Conar.

Let a, b, c > 0 be real numbers such that ab + bc + ca = 4abc. Prove
…
1 1 1 3 4
√ +√ +√ ≥4 .
a
a b
b c
c 3

We received 20 submissions of which 17 were correct and complete. We present

the solution by Oliver Geupel.
Let x = 1/a, y = 1/b, and z = 1/c. Then x + y + z = (ab + bc + ca)/(abc) = 4.
Using the strict convexity of t 7→ et log t = tt for t > 0, we have
 x + y + z (x+y+z)/3 Å ã4/3 …
1 1 1 x y z 4 3 4
√ +√ +√ =x +y +z ≥3 =3 =4 .
a
a b
b c
c 3 3 3

By the equality condition of Jensen’s inequality, here equality holds only if we

have a = b = c = 34 .

4806. Proposed by Arsalan Wares.

The figure shows three congruent, non-overlapping regular hexagons with a vertex
that is common to all three hexagons. Certain vertices of the hexagons are con-
nected as shown and these line segments enclose the shaded triangle. If each side
of each regular hexagon is 1, determine the exact area of the shaded triangle.

We received 26 solutions based on many different ideas and theorems. The follow-
ing solution is by Brian D. Beasley.
We model the hexagons with their common intersection point at the origin, placing
√
the other vertices
√ of the three sides common to that point at (−1, 0), (1/2, 3/2),
and (1/2, − 3/2). Then the sides of the shaded triangle lie on the following lines:
√ å √
√ √
Ç
3 3 3 3
`1 , connecting (−1, 3) and ,− ; its equation is y = − (x+1)+ 3;
2 2 5
Ç √ å √
3 3 3
`2 , connecting − , − and (2, 0); its equation is y = (x − 2);
2 2 7

Copyright © Canadian Mathematical Society, 2023

332/ Solutions

√ √ √ √
`3 , connecting (0, 3) and (−1, − 3); its equation is y = 2 3x + 3.
Ç √ å
12 2 3
Lines `1 and `2 intersect at A = ,− ;
13 13
Ç √ å
9 5 3
lines `2 and `3 intersect at B = − , − ;
13 13
Ç √ å
3 7 3
lines `3 and `1 intersect at C = − , .
13 13
6
Thus the side lengths of triangle ABC are AB = BC = CA = √ , so the triangle
√ Å √ 13
3 6 2 9 3
ã
is equilateral. Hence its area is √ = .
4 13 13

4807. Proposed by Dong Luu.

Let an acute triangle ABC be inscribed in the circle (O). M, N, P are respectively
the midpoints of BC, CA, AB. E is the intersection of ray P O and circle (ABM );
F is the intersection of ray N O and circle (ACM ). Suppose that AE, AF respec-
tively intersect the circle (M EF ) for the second time at the points G, H (G is
different from E, H is different from F ). Prove that GH, EF, BC are concurrent.
We received four submissions, of which two relied on computer calculations that
were basically correct, but included a step where there was division by zero. Our
featured solution is a composite of the work of Theo Koupelis and the proposer.

When AB = AC the configuration is symmetric about the line AM , in which case

the lines GH, EF, BC are parallel (and are concurrent in a point at infinity). So
let us assume that AB 6= AC, so that EF intersects BC in a finite point, call it Q.
Define K to be the point on the line segment AM for which AK = BM . We have

Crux Mathematicorum, Vol. 49(6), June 2023

 Solutions /333

∠KAE = ∠M AE = ∠M BE (inscribed in the circle ABM E), and BE = AE

(because P E is the perpendicular bisector of AB), whence the triangles BM E
and AKE are congruent. It follows that EK = EM ; similarly, F K = F M .
Consequently, EF is the perpendicular bisector of KM .
Furthermore, ∠CM E = ∠M KE (corresponding external angles at M and K),
while ∠M KE = ∠EM K (base angles of the isosceles triangle EM K), which
implies that M E bisects ∠CM K. Similarly, M F bisects ∠KM B. It follows
immediately that
• ∠EM F = 90◦ ,
• EF is a diameter of the circle (M EF ),
• K also lies on this circle, and
• EH ⊥ F H and F G ⊥ EG (angles inscribed in a semicircle), so that EH
and F G are altitudes of ∆AF E, and finally,
• QC is tangent to this circle at M (because the inscribed ∠M KE equals the
angle between QC and the chord M E).
We are now ready to apply Menelaus’ theorem to the triangle AF E to prove that
the line GH also passes through Q. First, we denote by U the point where AM
intersects QE. Referring to the right triangles F EG, F EH, EAU , and F AU , we
see that
GE F E · cos E EU/AE EU AF
= = = · . (1)
HF F E · cos F F U/AF F U AE
Furthermore, in the right triangles AF G and AHE we have cos A = AG AH
AF = AE ,
or
AH AE
= . (2)
GA AF
Finally, because M F and M E are the internal and external bisectors of ∠U M Q
in triangle U M Q, we have
EU QE
= . (3)
FU FQ
Rewriting (1) then substituting (2) and (3) we find that

GE F U AE AH F Q EG
1= · · = · · .
HF EU AF HF QE GA

Because Q is on the tangent BC to circle (M EF ) (and therefore on the line

EF external to the chord F E) while our assumption that the angles of ∆ABC are
acute insures that H and G lie in the interior of the sides AF and AE, respectively,
of ∆AF E, Menelaus’ theorem implies that Q, H, G are collinear. Consequently,
the lines GH, EF, BC are concurrent at Q.

Editor’s comment. Because neither computer argument explicitly distinguished

the case in which AB = AC (and the lines GH, EF, BC are not concurrent in a

Copyright © Canadian Mathematical Society, 2023

334/ Solutions

finite point), it is clear that there was an implicit division by zero somewhere in
the intricate calculations.

4808. Proposed by George Stoica.

Find all x, y, z ∈ N \ {0} with (x, y, z) = 1 and x + y + z | xk + y k + z k for
k ∈ {2, 4, 6}.
We received six solutions. We present the one by C. R. Pranesachar, lightly edited.
We have
x2 + y 2 + z 2 = (x + y + z)2 − 2(xy + xz + yz)
and

x6 + y 6 + z 6 = (x2 + y 2 + z 2 )(x4 + y 4 + z 4 − x2 y 2 − x2 z 2 − x2 y 2 ) + 3x2 y 2 z 2 .

Since x + y + z divides both x2 + y 2 + z 2 and x6 + y 6 + z 6 , we find

(x + y + z)|2(xy + xz + yz) (1)

and
(x + y + z)|3x2 y 2 z 2 . (2)

If a prime p ≥ 5 divides x + y + z, then (2) implies that p divides x2 y 2 z z . Then

p divides one of x, y, or z, say p divides x. From (1) we then get that p divides
yz and thus p divides one of y or z, say y. Since p divides x + y + z it then also
divides z, a contradiction to gcd(x, y, z) = 1. A similar argument show that 4 and
9 cannot divide x + y + z. Since x, y, z ≥ 1, we find x + y + z ∈ {3, 6}.
Assuming x ≤ y ≤ z the only options for (x, y, z) are then (1, 1, 1), (1, 2, 3), and
(1, 1, 4). Clearly (1, 1, 1) is a solution, whereas (1, 2, 3) is not, since 1 + 2 + 3 does
not divide 12 + 22 + 32 = 14. Finally 12 + 12 + 42 = 18, 14 + 14 + 44 = 258, and
16 + 16 + 46 = 4098 are all multiples of 6, thus (1, 1, 4) is a solution as well.

4809. Proposed by Daniel Sitaru.

Let a, b > 0. Find
n ÇZ 1 å−1 ÇZ 1 å−1
1 X xk xk
lim dx dx .
n→∞ n3 0 ax + b 0 bx + a
k=1

We received 9 solutions, one of which was incorrect. We present the solution by

Yunyong Zhang.
For x ∈ [0, 1],

(a + b)x ≤ (ax + b) ≤ (a + b), (a + b)x ≤ (bx + a) ≤ (a + b),

Crux Mathematicorum, Vol. 49(6), June 2023

 Solutions /335

implying that
1 1 1
1 xk xk xk−1 1
Z Z Z
= dx ≤ dx ≤ dx =
(a + b)(k + 1) 0 a+b 0 ax + b 0 a+b (a + b)k
and
1 1 1
1 xk xk xk−1 1
Z Z Z
= dx ≤ dx ≤ dx = .
(a + b)(k + 1) 0 a+b 0 bx + a 0 a+b (a + b)k
Multiplying gives
1
(a + b)2 k 2 ≤ ÄR 1 ä ÄR 1 ä ≤ (a + b)2 (k + 1)2 ,
xk xk
0 ax+b
dx 0 bx+a
dx

and summing then gives

n
(a + b)2 X 1 (a + b)2
(2n3 +3n2 +n) < ÄR 1 ä ÄR 1 ä< (2n3 +9n2 +13n).
6 xk
dx xk
dx 6
k=1 0 ax+b 0 bx+a

We thus have
n ÇZ 1 å−1 ÇZ 1 å−1
1 X xk xk (a + b)2
lim 3
dx dx = .
n→∞ n 0 ax + b
k=1 0 bx + a 3

4810. Proposed by Goran Conar.

Let a1 , a2 , . . . , an > 0 be real numbers such that a21 + a22 + · · · + a2n = 1, n > 1.
Prove that
√
a22 + a23 + · · · + a2n a21 + a23 + · · · + a2n a21 + a22 + · · · + a2n−1 n n
+ +· · ·+ ≥ .
(a2 + a3 + · · · + an )3 (a1 + a3 + · · · + an )3 (a1 + a2 + · · · + an−1 )3 (n − 1)2

We received 18 submissions and 17 of them were all complete and correct. We

present 2 different solutions by the majority of solvers.
Solution 1.
By Cauchy-Schwarz inequality, we have

a22 + · · · + a2n 1
(n−1)(a22 +· · ·+a2n ) > (a2 +· · ·+an )2 =⇒ > .
(a2 + · · · + an )3 (n − 1)(a2 + · · · + an )
Letting
a22 + · · · + a2n a21 + · · · + a2n−1
S= + · · · + ,
(a2 + · · · + an )3 (a1 + · · · + an−1 )3
we have
1 1 1
Å ã
S> + ··· + .
n−1 a2 + · · · + an a1 + · · · + an−1

Copyright © Canadian Mathematical Society, 2023

336/ Solutions

By AM-HM inequality, we get

n2 n2
S> = .
(n − 1)[(a2 + · · · + an ) + · · · + (a1 + · · · + an−1 )] (n − 1) (a1 + · · · + an )
2

Again by Cauchy-Schwarz we have

1 1
n = n(a21 + · · · + a2n ) > (a1 + · · · + an )2 =⇒ >√ .
a1 + · · · + an n

Thus √
n n
S>
(n − 1)2
as required.

Solution 2.
By Cauchy-Schwarz inequality, we have

(a2 + a3 + · · · + an )2 ≤ (n − 1)(a22 + a23 + · · · + a2n ).

It follows from the given condition a21 + a22 + · · · + a2n = 1 that

a22 + a23 + · · · + a2n 1 − a21 1 1

≥ = ·p ,
(a2 + a3 + · · · + an ) 3 [(n − 1)(1 − a21 )]3/2 (n − 1)3/2 1 − a21

with similar expressions for the other terms. Thus, it is sufficient to show that
n √ n √
1 X 1 n n X 1 n n
≥ or ≥√ .
(n − 1)3/2 i=1 1 − a2i (n − 1)2
p p
i=1
1 − a2i n−1

Let f (t) = √1 ,
1−t
where 0 < t < 1. Then f 0 (t) = 1
2(1−t)3/2
> 0, and f 00 (t) =
3
4(1−t)5/2
> 0. Thus, f (t) is an increasing, convex function in (0, 1). Using Jensen’s
inequality for ti = a2i , i = 1, 2, . . . , n, we get
n n √
1 t1 + t2 + · · · + tn 1
Å ã Å ã
X X n n
p = f (ti ) ≥ nf = nf =√ ,
i=1
2
1 − ai i=1
n n n−1
√
which is the desired inequality. Equality occurs when a1 = a2 = · · · = an = 1/ n.

Crux Mathematicorum, Vol. 49(6), June 2023