Crux Mathematicorum

VOLUME 41, NO. 2 February / Février 2015

Editorial Board
Editor-in-Chief Kseniya Garaschuk University of British Columbia

Contest Corner Editor Robert Bilinski Collège Montmorency

Olympiad Corner Editor Carmen Bruni University of Waterloo
Book Reviews Editor Robert Bilinski Collège Montmorency
Articles Editor Robert Dawson Saint Mary’s University

Problems Editors Edward Barbeau University of Toronto

Chris Fisher University of Regina
Anna Kuczynska University of the Fraser Valley
Edward Wang Wilfrid Laurier University
Dennis D. A. Epple Berlin, Germany
Magdalena Georgescu University of Toronto

Assistant Editors Chip Curtis Missouri Southern State University

Lino Demasi Ottawa, ON
Allen O’Hara University of Western Ontario

Guest Editors Alejandro Erickson Durham University

Joseph Horan University of Victoria
Amanda Malloch University of Victoria
Mallory Flynn University of British Columbia

Editor-at-Large Bill Sands University of Calgary

Managing Editor Denise Charron Canadian Mathematical Society

Copyright c Canadian Mathematical Society, 2015

46

IN THIS ISSUE / DANS CE NUMÉRO

47 Editorial Kseniya Garaschuk

48 The Contest Corner: No. 32 Robert Dawson
48 Problems: CC156–CC160
51 Solutions: CC106–CC110
55 The Olympiad Corner: No. 330 Carmen Bruni
55 Problems: OC216–OC220
57 Solutions: OC156–OC160
60 Extending a Tetrahedron I. Sharygin
65 Graphs and Edge Colouring David A. Pike
71 Problems: 3960, 4011–4020
75 Solutions: 2832b, 3911–3920
91 Solvers and proposers index

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

Crux Mathematicorum, Vol. 41(2), February 2015

 47

EDITORIAL
As I was reviewing my last editorial, I thought of how mathematics is taught
across the globe. Surely, math is math everywhere, but students’ and teachers’
approaches to it vary culture to culture. I have experienced this first-hand: I
moved to Canada when I was 18 having gone through the school system and even
one year of university in Belarus (yes, I was born in the USSR). For the sake of
strengthening my English, I decided to start university from scratch in Canada.
And now that I work in mathematical education, I often compare my two freshman
years. Belorussian first-year calculus was Canadian third-year analysis: rigorous,
technical, precise. We could find limits only using  and δ, we dealt with functions
purely analytically and we memorized a lot of proofs. In Canada, my experience
was the complete opposite: all we did was direct computations, graph functions
and generalize patterns. The former was theoretical, the latter — practical. Which
one is better?
Clearly, you need both. My theoretical experience prepared me for doing math
thoroughly, with great attention to detail and with forethought; my practical ex-
perience built up my intuition for math and taught me to always look for and make
connections between various representations of the same mathematical object. But
habits are persistent and we tend to stick to what we are more comfortable with.
So my Canadian students resist the theoretical side of math calling it “dry” and
“irrelevant”, while my Belorussian types often object to graphing and describing
math in words since it is “watered-down” and “not mathy”. You can’t please
everyone.
At Crux, it is also clear that people (both amongst our subscribers and within
the Editorial Board) have different preferred methods and tend to be faithful
to their favourite approaches. Whatever your taste, you will find something to
your liking on the pages of our journal. And if not, then clearly we are missing
your submissions! So send them along to [email protected] for problem
proposals and to [email protected] for articles.
Kseniya Garaschuk

Copyright c Canadian Mathematical Society, 2015

48/ THE CONTEST CORNER

THE CONTEST CORNER

No. 32
Robert Dawson
Les problèmes présentés dans cette section ont déjà été présentés dans le cadre d’un
concours mathématique de niveau secondaire ou de premier cycle universitaire, ou en
ont été inspirés. Nous invitons les lecteurs à présenter leurs solutions, commentaires et
généralisations pour n’importe quel problème. S’il vous plaı̂t vous référer aux règles de
soumission à l’endos de la couverture ou en ligne.

Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au rédacteur au plus tard le 1 avril 2016 ; toutefois, les solutions reçues après cette date
seront aussi examinées jusqu’au moment de la publication.

La rédaction souhaite remercier André Ladouceur, Ottawa, ON, d’avoir traduit les
problèmes.

CC156. Décrire et réaliser un croquis précis de la région qui représente l’en-

semble
{(x, y, z) : |x| + |y| 6 1, |y| + |z| 6 1, |z| + |x| 6 1}.

CC157. Étant donné une matrice 5 × 5 dont chacun des nombres est un 0 ou
un 1, démontrer qu’il doit exister une sous-matrice 2 × 2 (c’est-à-dire l’intersection
de la réunion de deux rangées avec la réunion de deux colonnes) dont tous les
nombres sont soit 0, soit 1.

CC158. On considère un point mobile A sur la partie positive de l’axe des

abscisses, un point mobile B sur la partie positive de l’axe des ordonnées et l’origine
O de manière que le triangle ABO ait toujours une aire de 4. Déterminer l’équation
d’une courbe, définie dans le premier quadrant, qui est tangente à chacun des
segments AB.

CC159. La disposition de huit tuiles carrées, ci-dessous à gauche, peut être

divisée en deux groupes congruents de quatre tuiles, comme sur la droite. (On
remarque qu’un groupe est le reflet de l’autre dans un miroir, ce qui est permis.)

 
    →    
     

Déterminer une façon de diviser la disposition suivante de 100 tuiles en deux

Crux Mathematicorum, Vol. 41(2), February 2015

 THE CONTEST CORNER /49

groupes congruents de 50 tuiles, ou démontrer qu’il est impossible de le réaliser.


  
    
.. .. ..
. . .
   ···    (19 tuiles dans la 10e rangée)

CC160. Déterminer tous les triplets (f, g, h) de fonctions continues à valeurs

réelles définies sur R telles pour tout nombre réel x,

f (g(x)) = g(h(x)) = h(f (x)) = x.

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

CC156. Describe and accurately sketch the region

{(x, y, z) : |x| + |y| 6 1, |y| + |z| 6 1, |z| + |x| 6 1}.

CC157. Show that if a 5 × 5 matrix is filled with zeros and ones, there must
always be a 2 × 2 submatrix (that is, the intersection of the union of two rows with
the union of two columns) consisting entirely of zeros or entirely of ones.

CC158. Suppose movable points A, B lie on the positive x-axis and y-axis,
respectively, in such a way that 4ABO, where O is the origin, always has area 4.
Find an equation for a curve in the first quadrant which is tangent to each of the
line segments AB.

CC159. The following pattern of eight square tiles can be divided into two
congruent sets of four tiles as shown. (Note that one set is the mirror image of
the other — this is legal.)

 
    →    
     

Copyright c Canadian Mathematical Society, 2015

50/ THE CONTEST CORNER

Find a way to divide the following pattern of 100 tiles into two congruent sets of
fifty tiles, or show it cannot be done.


  
    
.. .. ..
. . .
   ···    (19 tiles in the 10th row)

CC160. Find all triples of continuous functions f, g, h : R 7→ R such that,

for all x ∈ R,
f (g(x)) = g(h(x)) = h(f (x)) = x .

Three really magic squares

Having received his yearly salary in silver coins, the royal Mathematician arranged
the coins into vertical stacks and placed them on a 3 × 3 square so that the
numbers representing the amount of coins in each stack formed a magic square,
that is a square such that the sum of the numbers along every row, column and
diagonal of the square is the same. Some stacks came out being quite tall, but
none were higher than 300 coins tall.

The King liked the arrangement but lamented over the fact that all the numbers
came out being composite. “If your majesty gives me 9 more coins, I will add
one to each stack; the magic square property will be preserved but all the new
numbers will be prime”, replied the Mathematician. The King nearly agreed, but
was interrupted by the Joker, who took away one coin from each stack and the
new numbers all became prime (and the square, of course, remained a magic
square).

What was the original magic square composed by the Mathematician?

From Kvant, 1981 (9), p.31.

Crux Mathematicorum, Vol. 41(2), February 2015

 THE CONTEST CORNER /51

CONTEST CORNER
SOLUTIONS
Les énoncés des problèmes dans cette section apparaissent initialement dans 2014: 40(2),
p. 51–52.

CC106. At each summit of a regular tetrahedron of side length 3, we cut off

a pyramid such that the cut-off surface makes an equilateral triangle. The four
equilateral triangles thus obtained have all different dimensions. What is the total
length of the edges of the solid thus truncated? Provide a proof.
Originally problem 29 from Demi-finale du Concours Maxi de Mathématiques de
Belgique 2008.
We received two incomplete submissions to this problem, neither of which ade-
quately proved that the four tetrahedral corners that are removed from the original
tetrahedron must be regular. We present an editor’s solution.
Let A, B, C, and D be the vertices of the given tetrahedron, and let P, Q, and R
be the points on DA, DB, and DC, respectively, such that P QR is the equilateral
triangle formed by cutting off the pyramid containing the vertex D. Suppose that
the polyhedron has been labeled so that DP ≥ DQ ≥ DR. We will first prove
that these segments must, in fact, be equal. Compare triangles DP R and DQR.
We have RP = RQ and the angles at D are both 60◦ . Because we assume that
RD is no larger than either DP or DQ, the angles at P and Q must be acute.
From the sine law (applied to both triangles) we have

sin 60◦ sin 60◦

sin ∠DQR = DR = DR = sin ∠DP R,
RQ RP

from which we conclude that the two triangles are congruent, whence DQ = DP .
Focusing now on the 60◦ angle P DQ, we note that the length of the segment P Q
increases monotonically as the lengths DP = DQ increase, so there will be exactly
one position of P and Q for which P Q = QR (= RP ), namely where the lengths
DP, DQ, DR are all equal. Because the angles at D are all 60◦ , the three faces at
D are equilateral triangles, and the tetrahedron DP QR is therefore regular.
Returning to the problem, because all edges of a regular tetrahedron have the same
length, DP + DQ + DR = P Q + QR + RP and we conclude that the truncation
at the vertex D does not change the sum of the edge lengths. Of course, the same
can be said about the truncation at the other vertices, so the total length of the
edges of the truncated tetrahedron must equal 18.

CC107. In a right triangle ABC with right angle at B and BC = 1, we place

D on side AC such that AD = AB = 21 . What is the length of DC?

Copyright c Canadian Mathematical Society, 2015

52/ THE CONTEST CORNER

Originally problem 6 from Demi-finale du Concours Maxi de Mathématiques de

Belgique 2002.
There were eight solution submitted for this question, all essentially the same.
By the Pythagorean Theorem we have
… √
p 1 5
AC = BC 2 + AB 2 = 1+ =
4 2
and √ √
5 1 5−1
DC = AC − AD = − = .
2 2 2

CC108. In an orthonormal system, the line with equation y = 5x crosses

the parabola with equation y = x2 in point A. The perpendicular to OA at O
intersects the parabola at B. What is the area of triangle AOB?
Originally problem 20 from Demi-finale du Concours Maxi de Mathématiques de
Belgique 2009.
We received six correct solutions, and one incorrect solution. We present the
solution of Titu Zvonaru.
It is easy to deduce that A(5, 25). The slope of OB is −1/5. Solving the system
y = − 51 x, y = x2 we obtain B(− 51 , 25
1
).
√ √ » √
1 1 26
It follows that OA = 52 + 252 = 5 26, OB = 52 + 252 = 25 . Hence the
area of the triangle is AOB = OA·OB2
26
= 10 = 13
5 .

CC109. Let E be the set of reals x for which the two sides of the following
equality are defined:
sin kx
cot 8x − cot 27x = .
sin 8x sin 27x
If this equality holds for all the elements of E, what is the value of k?
Originally problem 21 from Demi-finale du Concours Maxi de Mathématiques de
Belgique 2009.
We received seven submitted solutions to this problem, one of which was incorrect
and five were incomplete. We present the only correct solution by Paolo Perfetti
modified by the editor.
mπ mπ
Note first that E = {x ∈ R|x 6= 8 and x 6= 27 for any m ∈ Z. For x ∈ E, the
given equality is equivalent to

sin 8x · sin 27x(cot 8x − cot 27x) = sin kx. (1)

We shall prove that the only value of k for which (1) holds for all x ∈ E is k = 19.

Crux Mathematicorum, Vol. 41(2), February 2015

 THE CONTEST CORNER /53

Since

sin 8x · sin 27x(cot 8x − cot 27x) = sin 27x cos 8x − cos 27x sin 8x
= sin (27x − 8x) = sin 19x,

k = 19 satisfies (1).
Next, suppose (1) holds for all x ∈ E and some k ∈ Z with k 6= 19.
If k = −19, then from (1) we have 2 sin 19x = 0 for all x ∈ E, which is false (for
π
example, if x = 38 , then x ∈ E, but sin 19x = sin π2 = 1 6= 0). Hence k 6= −19.
From (1), we also have
Å ã Å ã
19 − k 19 + k
2 sin x cos x = 0. (2)
2 2
19−k 19−k 2mπ 19+k



Since sin 2 x = 0 if and only if 2 x = mπ or x = 19−k and cos 2 x =0
if and only if 19+k 1
2 x = (m + 2 )π or x = (2m+1)π
19+k some m ∈ Z, there must be
for
some x ∈ E that does not satisfy (2). (To be more precise, the set of all x such
2mπ
that x = 19−k or x = (2m+1)π
19+k for some m ∈ Z is countable while E is clearly
uncountable.) This is a contradiction and our proof is complete.

CC110. What is the number of real solutions to the equation:

|1 + x − |x − |1 − x||| = | − x − |x − 1||.

Originally problem 26 from Demi-finale du Concours Maxi de Mathématiques de

Belgique 2009.
We have received four correct solutions and one incorrect submission. We present
the solution by Henry Ricardo.
We compute the left-hand side (LHS) and the right-hand side (RHS) on three
intervals that cover the real number line.
Case 1. Suppose that 0 ≤ x ≤ 1. Then

RHS = | − x − (1 − x)| = | − x − 1 + x| = 1.

When x ∈ [−, 21 ],

|1 + x − |x − (1 − x)|| = |1 + x − (1 − 2x)| = 3x

and when x ∈ ( 21 , 1],

|1 + x − |2x − 1|| = |1 + x − (2x − 1)| = |2 − x| = 2 − x

so that ß
3x if 0 ≤ x ≤ 21 ,
LHS =
2−x if 12 < x ≤ 1.

Copyright c Canadian Mathematical Society, 2015

54/ THE CONTEST CORNER

1
Thus LHS = RHS when either 3x = 1 or 2 − x = 1, which implies x = 3 and
x = 1 for x ∈ [0, 1].
Case 2. If x > 1, we have

LHS = |1 + x − |x − (x − 1)|| = |1 + x − 1| = x

and
RHS = | − x − (x − 1)| = | − 2x + 1| = 2x − 1.
But x > 1 implies that (2x − 1) − x = x − 1 > 0, so RHS > LHS and there are no
solutions to the equation in the interval (1, ∞).
Case 3. Finally, for x ∈ (−∞, 0), RHS = 1 and LHS = |3x| = −3x, so LHS =
RHS if and only if −3x = 1, which implies x = − 31 .
1 1
Thus the only solutions of the given equation are x = − , 1, .
3 3

Math Quotes

The solution of problems is one of the lowest forms of mathematical research.

Yet, its educational value cannot be overestimated. It is the ladder by which
the mind ascends into higher fields of original research and investigation. Many
dormant minds have been aroused into activity through the mastery of a single
problem.

Benjamin Franklin Finkel.

Crux Mathematicorum, Vol. 41(2), February 2015

 THE OLYMPIAD CORNER /55

THE OLYMPIAD CORNER

No. 330
Carmen Bruni
Les problèmes présentés dans cette section ont déjà été présentés dans le cadre d’une
olympiade mathématique régionale ou nationale. Nous invitons les lecteurs à présenter
leurs solutions, commentaires et généralisations pour n’importe quel problème. S’il vous
plaı̂t vous référer aux règles de soumission à l’endos de la couverture ou en ligne.
Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au rédacteur au plus tard le 1 avril 2016 ; toutefois, les solutions reçues après cette date
seront aussi examinées jusqu’au moment de la publication.
La rédaction souhaite remercier Rolland Gaudet, professeur titulaire à la retraite à l’Uni-
versité de Saint-Boniface, d’avoir traduit les problèmes.

OC216. Soit p = n2 + 1 un nombre premier. Déterminer toutes les solutions

entières à l’équation suivante :

x2 − (n2 + 1)y 2 = n2 .

OC217. Soit G le centroı̈de du triangle rectangle ABC où ∠BCA = 90◦ . Soit
P le point sur le rayon AG tel que ∠CP A = ∠CAB, et soit Q le point sur le rayon
BG tel que ∠CQB = ∠ABC. Démontrer que les cercles circonscrits de AQG et
BP G se rencontrent à un point sur le côté AB.

OC218. Déterminer toute fonction f : N → N satisfaisant

f (mn) = lcm(m, n) · gcd(f (m), f (n))

pour tous les entiers positifs m et n.

OC219. Pour m et n des entiers positifs donnés, démontrer qu’il existe un

entier c tel que les nombres cm et cn ont le même nombre d’occurences de chaque
chiffre non nul, lorsqu’ils sont exprimés en base 10.

OC220. Soit A1 A2 ...A8 un octagone convexe où tous les côtés sont de même
longueur et où les côtés opposés sont parallèles. Pour chaque i = 1, ..., 8, posons
Bi le point d’intersection des segments Ai Ai+4 et Ai−1 Ai+1 , où Aj+8 = Aj et
Bj+8 = Bj pour tout j. Fournir un nombre i, parmi 1, 2, 3, et 4, satisfaisant

Ai Ai+4 3
6 .
Bi Bi+4 2

Copyright c Canadian Mathematical Society, 2015

56/ THE OLYMPIAD CORNER

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

OC216. Let p = n2 + 1 be a given prime number. Find the set of integer

solutions to the following equation :

x2 − (n2 + 1)y 2 = n2 .

OC217. Let G be the centroid of a right-angled triangle ABC with ∠BCA =

90◦ . Let P be the point on ray AG such that ∠CP A = ∠CAB, and let Q be
the point on ray BG such that ∠CQB = ∠ABC. Prove that the circumcircles of
triangles AQG and BP G meet at a point on side AB.

OC218. Find all functions f : N → N satisfying

f (mn) = lcm(m, n) · gcd(f (m), f (n))

for all positive integers m, n.

OC219. Given positive integers m and n, prove that there is a positive integer
c such that the numbers cm and cn have the same number of occurrences of each
non-zero digit when written in base ten.

OC220. Let A1 A2 ...A8 be a convex octagon such that all of its sides are equal
and its opposite sides are parallel. For each i = 1, ..., 8, define Bi as the intersection
between segments Ai Ai+4 and Ai−1 Ai+1 , where Aj+8 = Aj and Bj+8 = Bj for all
j. Show that some number i, amongst 1, 2, 3, and 4 satisfies
Ai Ai+4 3
6 .
Bi Bi+4 2

Crux Mathematicorum, Vol. 41(2), February 2015

 THE OLYMPIAD CORNER /57

OLYMPIAD SOLUTIONS
Les énoncés des problèmes dans cette section apparaissent initialement dans 2014 : 40(2),
p. 56–57.

OC156. Let ABCD be a tetrahedron. Prove that vertex D, center of ins-

phere and centroid of ABCD are collinear if and only if the areas of triangles
ABD, BCD, CAD are equal.
Originally question 2 from day 1 of the Poland Math Olympiad.
We received four correct solutions to this problem. We present the solution by
Michel Bataille.
Let I and r be the center and the radius of the insphere. Let V(·) and A(·) denote
volume and area, respectively.
1
Since V(IBCD) = 3 · r · A(BCD) (and similarly for triangles CDA, DAB and
ABC), we can use

(A(BCD) : A(CDA) : A(DAB) : A(ABC))

instead of
V(IBCD) : V(ICDA) : V(IDAB) : V(IABC))
for the barycentric coordinates of I relative to (A, B, C, D). It follows that

σI = (A(BCD))A + (A(CDA))B + (A(DAB))C + (A(ABC))D

where σ is the sum of the areas of the faces of ABCD and the bold face letters
represent the coordinates of the points. In particular, we have
−→ −−→ −−→ −−→
σ DI = (A(BCD))DA + (A(CDA))DB + (A(DAB))DC. (1)

Let G be the centroid of ABCD. Then,

4G = A + B + C + D,

from which we deduce

−−→ −−→ −−→ −−→
4DG = DA + DB + DC. (2)
Now, D, I, G are collinear if and only if
−→ −−→
σ DI = λ(4DG) (3)
−−→ −−→ −−→
for some real number λ. Since DA, DB, DC are not coplanar, (1) and (2) show
that (3) occurs if and only if

A(BCD) = A(CDA) = A(DAB).

Copyright c Canadian Mathematical Society, 2015

58/ THE OLYMPIAD CORNER

OC157. Find all functions f : R → R such that

f (f (x)2 + f (y)) = xf (x) + y, ∀x, y ∈ R .

Originally question 4 from Kyrgyzstan National Olympiad.

We received three correct solutions and one incorrect submission. We present the
solution by Henry Ricardo.
Letting x = 0 in the given equation, we have

f (f (0)2 + f (y)) = y, ∀y ∈ R (1)

so f is onto. Thus there exists a ∈ R such that f (a) = 0. Taking x = a in the

original equation gives us

f (f (y)) = y, ∀y ∈ R, (2)

which shows that f is one to one since f (x) = f (y) implies f (f (x)) = f (f (y)), or
by (2), x = y.
Now replace x by f (x) in the original equation to obtain

f (x2 + f (y)) = f (x)f (f (x)) + y = xf (x) + y = f (f (x)2 + f (y))

Usnig the fact that f is one to one, we have

x2 + f (y) = f (x)2 + f (y).

Therefore x2 = f (x)2 and f (x) = ±x.

To eliminate the possibility that f (x) = x for some x and f (y) = −y for some
y 6= x, suppose that xy 6= 0 and f (x) = x, f (y) = −y. The original equation gives
us f (x2 − y) = x2 + y, but we know that f (x) = ±x and so ±(x2 − y) = x2 + y
implies either x = 0 or y = 0. Now it is clear that f (x) = x for all x ∈ R
and f (x) = −x for all x ∈ R satisfy the original equation and are the only such
functions.

OC158. Prove that a finite simple planar graph has an orientation so that
every vertex has out-degree at most 3.
Originally question 4 from day 1 of the Romania TST.
We received no solutions to this problem.

OC159. Let p be an odd prime number. Prove that there exists a natural
number x such that x and 4x are both primitive roots modulo p.
Originally question 3 from the 2012 Iran National Math Olympiad Third Round.
We received one correct solution and one incorrect submission. We present the
solution by Oliver Geupel.

Crux Mathematicorum, Vol. 41(2), February 2015

 THE OLYMPIAD CORNER /59

The existence of a primitive root modulo p is a well-known fact. Suppose that a

is a primitive root modulo p. Then there is a positive integer r such that 2 ≡ ar
(mod p) and therefore 4 ≡ a2r (mod p). Let p1 , p2 , . . . , p` denote the distinct prime
divisors of p − 1. For 1 6 k 6 `, let sk be an integer such that sk 6≡ 0 (mod pk ) and
sk 6≡ −2r (mod pk ). Find via the Chinese Remainder Theorem, a natural number
m such that
m ≡ sk (mod pk ) , 1 6 k 6 `.
Then, neither m nor m + 2r is divisible by any pk . Hence, each of the numbers m
and m + 2r is coprime with p − 1.
We obtain
p − 1 - m, 2m, 3m, . . . , (p − 2)m,
p − 1 - m + 2r, 2(m + 2r), 3(m + 2r), . . . , (p − 2)(m + 2r).
Thus,
am , a2m , a3m , . . . , a(p−2)m 6≡ 1 (mod p) ,
am+2r , a2(m+2r) , a3(m+2r) , . . . , a(p−2)(m+2r) 6≡ 1 (mod p) ,
since a is a primitive root modulo the prime p. We have obtained that am and
am+2r ≡ 4am (mod p) are primitive roots modulo p. Therefore, x = am has the
required property.

OC160. The incircle of triangle ABC, is tangent to sides BC, CA and AB

at D, E respectively F . Let T and S be the reflection of F with respect to B
respectively the reflection of E with respect to C. Prove that the incenter of
triangle AST is inside or on the incircle of triangle ABC.
Originally question 3 from day 2 of the Iran National Math Olympiad Second
Round.
No solutions were received.

Copyright c Canadian Mathematical Society, 2015

60/ FROM THE ARCHIVES : EXTENDING A TETRAHEDRON

Extending a tetrahedron
I. Sharygin
One of the most beautiful tools that can be used when solving geometrical problems
consists of replacing the geometric figure in question with another one, a more
convenient one in some sense. For example, if the problem involves a triangle
with a median, often it is helpful to use this triangle to construct a parallelogram
therefore extending the median to turn it into the parallelogram’s diagonal. In
this article, we will consider several problems involving a triangular pyramid, the
so-called tetrahedron, that can be solved by extending the tetrahedron to another
solid, often a parallelepiped.
The first way to extend the tetrahedron is presented in Figure 1. Here, AA1 BD
is the given tetrahedron. The plane of each of the faces DCC1 D1 , CBB1 C1 and
A1 B1 C1 D1 of the parallelepiped passes through one vertex of the tetrahedron and
is parallel to the edge of it opposite of that vertex. This way, one of the corners of
the tetrahedron becomes one of the corners of the parallelepiped.

Figure 1: Extending the tetrahedron, method 1.

Problem 1 Suppose the tetrahedron AA1 BD is a right-angle triangular pyramid,

that is, the edges AA1 , AB and AD are mutually perpendicular.
a) Prove that vertex A of the tetrahedron, the point M of intersection of the
medians of the face A1 BD and the centre of the circumsphere are collinear.
(Note : this problem appeared on the entrance exam to the Mechanical Ma-
thematical Department of the Moscow State University.)
b) Find the radius of the sphere circumscribed around this pyramid.
We extend tetrahedron AA1 BD to a parallelepiped (a right angle one) as in Fi-
gure 1. Then the sphere circumscribed around the tetrahedron is also circumscribed
around the parallelepiped. The radius of this sphere is then equal
p to half the length
of the diagonal of the parallelepiped, that is it is equal to 12 AA21 + AB 2 + AD2 ,
which answers part b) of the problem.

Crux Mathematicorum, Vol. 41(2), February 2015

 I. SHARYGIN /61

To prove part a), consider rectangle AA1 C1 C. Centre O of the sphere lies on the
diagonal AC1 . Median A1 O1 of the triangle A1 BD intersects AC1 at a point M .
Since triangles A1 C1 M and O1 AM are similar, we get :

A1 M A1 C 1
= = 2,
O1 M O1 A
meaning that M is the point of intersection of the medians of the triangle A1 BD
and we are done.
Another common way to extend a tetrahedron to a parallelepiped is as follows : for
every edge of the tetrahedron, construct a plane containing this edge and parallel
to the opposite edge (Figure 2, left). This way, the edges of the tetrahedron become
the diagonals of the faces of the parallelepiped. A little practical hint : it is easier
to sketch this construction if you start with the parallelepiped first.

Figure 2: Extending the tetrahedron, method 2.

Problem 2 Find the radius of the sphere that touches all the edges of a regular
tetrahedron with edge length a.
As one can easily see from Figure 2, a parallelepiped constructed
√ in this way around
a regular tetrahedron is a cube with edge length a/ 2. The sphere touching the
edges of the tetrahedron is the sphere inscribed in the cube. The answer is therefore
a
√
2 2
.
So the first method of extending a tetrahedron is useful when you are given all
angles at one of the vertices (especially if those angles are all right angles) ; the
second method is helpful in problems involving the opposite edges of the tetrahe-
dron.
Problem 3 Two opposite edges of the tetrahedron have length a, two other oppo-
site edges have length b and the two remaining opposite edges have length c. Find
the distance between the centre of the tetrahedron’s insphere and the centre of the
sphere that touches one face of the tetrahedron and the extension of all the others.
Consider such a tetrahedron ABCD and the parallelepiped AM CKN DLB construc-
ted as in Figure 2 on the right. Since each edge of the tetrahedron equals its op-

Copyright c Canadian Mathematical Society, 2015

62/ FROM THE ARCHIVES : EXTENDING A TETRAHEDRON

posite edge, all the faces of the parallelepiped are rectangles and hence the whole
solid is rectangular.
The centre of the sphere inscribed in the tetrahedron ABCD coincides with the
intersection point of all the parallelepiped’s diagonals (prove this fact). Without
loss of generality, assume that the external sphere touches the face DCB and the
extension of the other faces. Then the centre of this sphere lies at the vertex L of
the parallelepiped. To see this, consider the pyramid B 0 LCD equal to the pyramid
BLCD, where B 0 L = LB : the points A, D, B 0 and C lie in one plane. Therefore,
L is equidistant from the face DCB and the extension of ACD. Similarly, one can
show that L is equidistant from the planes of DCB and ACB as well as DCB
and ADB.
Therefore, the distance in question is equal to half the length of the diagonal of
the parallelepiped. Let x, y, z denote the lengths of the edges of the parallelepiped.
By Pythagoras’ Theorem, we get a system of three equations :

x2 + y 2 = a2 ,
x 2 + z 2 = b2 ,
y 2 + z 2 = c2 .

Adding them all up, we find that

AL 1p 2 1 a2 + b2 + c2
= x + y2 + z2 = .
2 2 2 2

Problem 4 Let A1 B1 CD be a tetrahedron ; let Π be the (unique) plane parallel to

the lines A1 B1 and CD and equidistant from them. Let S be the area of the cross
section of A1 B1 CD cut by Π and suppose the distance between the opposite edges
is h. Find the volume of the tetrahedron.
Suppose ABCDA1 B1 C1 D1 is the parallelepiped as in Figure 3.

Figure 3: Cross section of the tetrahedron.

The volume of the tetrahedron A1 BC1 D is equal to the volume of the parallele-
piped minus the volumes of the four tetrahedra, one on each face of the initial

Crux Mathematicorum, Vol. 41(2), February 2015

 I. SHARYGIN /63

tetrahedron and each with the volume equal to one sixth of the volume of the
parallelepiped (why ?). Therefore, Vtetrahedron = Vparalellepiped /3.
Let A1 C1 and BD be the two opposite edges and let KLM N be the plane in-
tersecting the parallelepiped in midpoints of the vertical edges AA1 , BB1 , CC1
and DD1 . Then the vertices of the cross section defined in the problem are the
midpoints of the edges of the parallelogram KLM N . Therefore,

Area(KLM N ) = 2S = Area(ABCD).

Then
1 2
Vtetrahedron = Vparalellepiped = Sh.
3 3

(Using problem 4, one can easily prove Simpson’s formula used to compute volumes
of certain solids.)
In conclusion, let us show one example where it is more convenient to extend the
tetrahedron to the triangular prism.
Problem 5 Suppose that in a tetrahedron, areas of two faces are equal to S1 and
S2 and the angle between them is α. Suppose further that the areas of the two other
faces are equal to Q1 and Q2 and the angle between them is β. Show that

S12 + S22 − 2S1 S2 cos α = Q21 + Q22 − 2Q1 Q2 cos β.

Let us first prove that if the area of one lateral face of a prism is S and the areas of
two other lateral faces are equal to S1 and S2 with the angle between them equal
to α, then
S12 + S22 − 2S1 S2 cos α = S 2 .

Figure 4: Extending the tetrahedron to a prism.

Indeed, let the plane ABC be perpendicular to the lateral faces of the prism and
let ∠BAC = α (Figure 4, left). By Law of Cosines applied to triangle ABC, we
have
BC 2 = AB 2 + AC 2 − 2 · AB · AC · cos α.

Copyright c Canadian Mathematical Society, 2015

64/ FROM THE ARCHIVES : EXTENDING A TETRAHEDRON

All that is left to do now is to multiply each side by l2 , where l is the length of
the lateral edge of the prism.
Now, let us come back to the main statement of Problem 5. Given the tetrahedron
ABCD, let S4ABD = S1 , S4ADC = S2 , S4ABC = Q1 , S4DBC = Q2 , angle at
the edge AD be equal to α and angle at the edge BC be equal to β. Consider
the triangular prism with base ABC with AD as one of the lateral edges (Figure
4, right). Let S be the area of the parallelogram BB1 C1 C, then by the formula
proved above we have :

4S12 + 4S22 − 8S1 S2 cos α = S 2 .

Note that S = AD · BC · sin γ, where γ is the angle between edges BC and AD.
Similarly, considering the other triangular prism with base ACD and lateral edge
BC, we get :
4Q21 + 4Q22 − 8Q1 Q2 cos β = S 2 .
Combining the above, we are done.

Exercises.
1. Prove that the sum of squares of the edge lengths of a tetrahedron is equal
to four times the sum of squares of distances between the midpoints of its
opposite edges.
2. Given a tetrahedron ABCD, show that the directions of opposite edges AD
and BC are perpendicular if and only if AB 2 + DC 2 = AC 2 + DB 2 .
3. The lengths of two opposite edges of a tetrahedron are equal to a, the other
two b and the last two c. Find a) the volume of this tetrahedron ; b) the
radius of the circumsphere.

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

This article originally appeared in Russian in Kvant, 1976 (1), p. 60–65. It has
been translated and adapted with permission.

Crux Mathematicorum, Vol. 41(2), February 2015

 DAVID A. PIKE /65

Graphs and Edge Colouring

David A. Pike
Let’s begin with a familiar scenario : suppose that we need to schedule the matches
of a round-robin tournament involving n teams, so that once the tournament has
concluded each team will
have played against each other team exactly once. In
total there would be n2 games. Potentially they could be scheduled sequentially,
but that could make for a very long and drawn out tournament. In the interest
of completing the event as quickly as possible, we instead want to have several
teams competing simultaneously. The question that now arises is this : given that
no team can play more than one game at a time, how few time slots are needed
in order to schedule the whole tournament ?
This particular question was answered long ago by modelling it with graph theory.
In the 1890s Édouard Lucas [5] published a solution, for which he gave credit as
follows :
Parmi les diverses méthodes qui nous ont été indiquées, nous expose-
rons, de préférence, les solutions simples et ingénieuses de M. Walecki,
professeur de Mathématiques spéciales au lycée Condorcet.
At this stage it would be good to know exactly what a graph is. Formally, a graph
G consists of a set V of elements called vertices, accompanied by a set E of edges,
which themselves are pairs of vertices. Any graph can easily be represented in
the form of a drawing. For instance, in Figure 1 are two drawings of the graph
having V = {a, b, c, d, e} and E = {a, b}, {a, c}, {a, d}, {b, d}, {c, d}, {d, e} . Each
vertex is depicted as a circular node, and each edge is illustrated by drawing a line
between its two vertices.

Figure 1: Two different drawings of a graph

Note that there is no specified location for the vertices or edges of a graph. Mo-
reover, the edges do not need to be drawn as straight lines, but are free to have
bends and curves. In practice the vertices may represent real entities (such as
sports teams which do have a geographical placement) and the edges might also
represent connections with physical form (such as railway lines between cities),
but what is most important here is that the graph captures the existence of a
relationship between pairs of vertices (such as the need for their corresponding

Copyright c Canadian Mathematical Society, 2015

66/ GRAPHS AND EDGE COLOURING

teams to play against each other).

For our scenario of a round-robin tournament, we will want to consider a graph
with n vertices such that each pair of vertices is joined by an edge. Such a graph
is called a complete graph and is denoted by Kn . Figure 2 illustrates the graph
K7 , with the vertices named 0 to 6. As can be seen in this example, we also allow
edges to be drawn so that their lines intersect.

Figure 2: The complete graph K7

If we have seven teams that must each play against each other during a tournament,
then each edge of K7 corresponds to an individual game that has to be played.
But we still need to find a way to schedule the games, preferably into as few time
slots as possible. Since each edge of K7 represents a distinct game that must be
played, then the games within a single time slot correspond to a set of edges, no
two of which share a vertex. So to find a schedule, we need only find a way to
partition the edges of K7 into sets of this form.
Exercise 1 Find a schedule for these 21 games.
Having just found a schedule for the games that the seven teams must play, we
now have to consider whether there might be a better schedule.
Question 1 Is there a schedule that uses fewer time slots ?
If not, then how can you be sure that yours is indeed the best ? To determine
just how few time slots there are in an optimal schedule, we need to do some
mathematical thinking.
Observe that each of the seven teams has to play six games, so right away we know
that the number of time slots that are in any valid schedule has to be at least six.
Were you able to find a schedule that only used six time slots ?
Question 2 Can you find a schedule with six time slots ? If not, then can you
prove that there is no schedule with only six time slots ?

Crux Mathematicorum, Vol. 41(2), February 2015

 DAVID A. PIKE /67

As it happens, there is no way to schedule the tournament with seven teams so

that all of the games fit into only six time slots. To convince yourself that this is
the case, note that the teams that are playing games within a single time slot are
playing in pairs. So with seven teams, at most six of them can actually be playing
at any given time, which in turn means that at most three games can take place
at a time. In total, there are 21 games that must be played, and with at most
three that can be scheduled per time slot, we need at least seven time slots for the
tournament. With this argument in mind, we can now conclude that any schedule
that uses only seven time slots must in fact be an optimal solution.
Let’s move away from the example of n = 7 now and consider what might happen
when n is even. Is it still the case that n − 1 time slots is impossible ?
Exercise 2 Use K4 and K6 to find optimal schedules for n = 4 and n = 6.
You should find that for these two small examples it is actually possible to find
schedules with as few as 3 and 5 time slots, respectively. To try to see a general
pattern we will now consider n = 8. With teams named 0 to 6 and ∞, Figure 3
illustrates how to form four pairs of teams for the first time slot of the tournament,
and then how to form four pairs for the second time slot. Looking at this figure, a
general approach ought to become apparent : rotate the edges clockwise for each
subsequent time slot. To be a bit more technical, for each edge {u, v} of one time
slot, for the next time slot use the edge {u + 1, v + 1} where we treat ∞ + 1 as ∞
and (n − 2) + 1 as 0.

Figure 3: Team pairings for two different time slots in K8

So for even n, an optimal solution is to use this technique with the edges {0, ∞} and
{1, n − 2}, {2, n − 3}, . . . , { n−2 n
2 , 2 } for the initial time slot. The resulting schedule
will have a total of (n − 1) time slots for the tournament.
For odd n, however, recall that each time slot has to have a team that sits out. We
can find a schedule that uses n time slots by introducing a phantom team called
∞, building an optimal schedule for (n + 1) teams (note that n + 1 is even), and
then assigning byes to teams whenever they are paired with the phantom team.

Copyright c Canadian Mathematical Society, 2015

68/ GRAPHS AND EDGE COLOURING

So when n is odd, we know that a schedule with n time slots can be achieved,
although for n 6= 7 we have not yet proved that n − 1 time slots are insufficient.

At this point hopefully you are beginning to wonder what any of this has to do
with colouring, although perhaps you’ve already discovered that each time slot can
be associated with a distinct colour. If we colour the edges of a graph so that two
edges that meet at a common vertex are not allowed to share the same colour, then
it is possible to use the colouring to form a schedule of pairings. Alternatively, if we
do not actually have colours to work with (such as this black-and-white article),
we can emulate the idea of colours with dashed lines, etc., similar to what we have
done for the graph shown in Figure 4. The solid black edges {a, c} and {b, d} could
be used to indicate two games to be played during the first day of a competition
(with team e having a bye), the short-dashed edge {c, d} provides for just one
game on the second day, the dotted edges {a, b} and {d, e} tell us which games are
to take place on the third day, and finally the long-dashed edge {a, d} corresponds
to the sole game on the fourth day of the competition. Note that in this example
the competition is not a round-robin tournament (since not every pair of teams
will play against each other).

Figure 4: Example of an edge colouring

An edge colouring for which edges of the same colour never meet at a common
vertex is called a proper edge colouring. An easy way to obtain a proper edge
colouring is to give each edge a distinct colour, but this would result in no games
taking place at the same time. As before, our goal is to determine how few time
slots are needed. With our new terminology, given a graph G we want to know
the smallest number of colours for which a proper edge colouring exists ; this value
is called the chromatic index of the graph and is denoted by χ0 (G). An obvious
lower bound on the chromatic index is that χ0 (G) > ∆(G), where ∆(G) denotes
the number of edges at any vertex with the most edges (for the graph in Figure 4,
∆(G) is 4 thanks to vertex d belonging to four edges).

For complete graphs, we have already seen that χ0 (K2` ) = 2` − 1 and χ0 (K2`+1 ) 6
2` + 1. So already we have examples of graphs, some of which have χ0 (G) = ∆(G)
and some for which χ0 (G) might be as high as ∆(G) + 1. As it turns out, provided
that each pair of vertices is joined by either one edge or none, then these are the

Crux Mathematicorum, Vol. 41(2), February 2015

 DAVID A. PIKE /69

only two possible values for χ0 (G), as was proved by Vadim Vizing in the 1960s
(this result is proved in most graph theory textbooks, such as [5]).
Vizing’s Theorem If for each pair of vertices of a graph G there is at most one
edge between them, then χ0 (G) ∈ {∆(G), ∆(G) + 1}.
Given that there are only two possible values for the chromatic index, it has become
common practice to say that graphs for which χ0 (G) = ∆(G) are Class 1 and that
graphs for which χ0 (G) = ∆(G) + 1 are Class 2. Examples of Class 1 graphs
include complete graphs with an even number of vertices, as well as all bipartite
graphs ; a graph is called bipartite if its vertex set V can be partitioned into two
subsets A and B so that every edge of the graph has one of its two vertices in A
and the other in B. Bipartite graphs have numerous applications ; indeed, whole
books have been written just on bipartite graphs (see [1] for one of them). Class 2
 include odd-length cycles (e.g., the graph having V = {1, 2, . . . , 2` + 1} and
graphs
E = {1, 2}, {2, 3}, . . . , {2`, 2` + 1}, {1, 2` + 1} .
Questions regarding how to identify which class a particular graph might be are
natural to ask. As an example, in Figure 5 is the famous Petersen graph.
Exercise 3 Determine whether the Petersen graph is Class 1 or Class 2.

Figure 5: The Petersen graph

For some graphs, there is an easy way to determine their class. To give a definition,
we will say that a graph G is overfull if |E| strictly exceeds ∆(G)b |V2 | c, where the
notation |S| denotes the cardinality of the set S and bxc denotes the greatest
integer not exceeding the real number x (so for example bπc = 3, b7c = 7 and
b−πc = −4).
Amanda Chetwynd and Anthony Hilton pioneered some of the research on overfull
graphs [2]. It is easy to prove that any graph that is overfull must be Class 2. By
way of contradiction, suppose that there exists an overfull graph G that happens
to be Class 1. Since it is Class 1, χ0 (G) = ∆(G). Moreover, each colour can be used

Copyright c Canadian Mathematical Society, 2015

70/ GRAPHS AND EDGE COLOURING

on at most b |V2 | c edges, for if a colour occurred on any more edges then at least two
edges of that colour would have to meet at a common vertex. With ∆(G) colours,
each occurring on at most b |V2 | c edges, it follows that |E| 6 ∆(G)b |V2 | c, in violation
of the graph being overfull. Thus we have obtained the desired contradiction, from
which we conclude that the graph cannot be Class 1.
Having previously determined that χ0 (Kn ) 6 ∆(G) + 1 when n is odd was not
itself a proof that complete graphs with an odd number of vertices are Class 2.
However, by verifying that any complete graph with an odd number of vertices
is overfull, we can now confirm that K2`+1 is Class 2. The Petersen graph is also
Class 2.
It is not too hard to deduce that if a graph is overfull then it necessarily must
have an odd number of vertices. This condition is not sufficient though, for there
do exist Class 1 graphs having an odd number of vertices (simply refer to Figure 4
to see an example).
The examples that we have seen so far have not been very difficult. However,
determining whether a given graph is Class 1 versus Class 2 is generally not an
easy problem. Indeed, Ian Holyer proved in 1981 that it is so hard that it is NP-
complete [3]. Nevertheless, motivated both by scientific curiosity as well as the
applications that exist for edge colourings, this continues to be an active area of
research whereby people try to find faster colouring algorithms and also try to
establish that certain types of graphs are Class 1 versus Class 2.

References

[1] A.S. Asratian, T.M.J. Denley and R. Häggkvist. Bipartite graphs and their
applications, Cambridge University Press, Cambridge (1998).
[2] A.G. Chetwynd and A.J.W. Hilton. Star multigraphs with three vertices of
maximum degree. Math. Proc. Camb. Phil. Soc. 100 (1986) 303–317.
[3] I. Holyer. The NP-completeness of edge-coloring. SIAM Journal on Computing
10 (1981) 718–720.
[4] É. Lucas. Récréations Mathématiques, Vol. II, Gauthier-Villars et Fils, Paris
(1892) 162.
[5] D.B. West. Introduction to Graph Theory, Prentice-Hall, Englewood Cliffs, NJ,
1996.

Crux Mathematicorum, Vol. 41(2), February 2015

 PROBLEMS /71

PROBLEMS
Nous invitons les lecteurs à présenter leurs solutions, commentaires et généralisations
pour n’importe quel problème présenté dans cette section. De plus, nous les encourageons
à soumettre des propositions de problèmes. S’il vous plaı̂t vous référer aux règles de
soumission à l’endos de la couverture ou en ligne.
Pour faciliter l’examen des solutions, nous demandons aux lecteurs de les faire parvenir
au rédacteur au plus tard le 1 avril 2016 ; toutefois, les solutions reçues après cette date
seront aussi examinées jusqu’au moment de la publication.
La rédaction souhaite remercier Rolland Gaudet, professeur titulaire à la retraite à l’Uni-
versité de Saint-Boniface, d’avoir traduit les problèmes.

3960. Proposé par George Apostolopoulos. Correction.

Soient a, b, c des nombres réels non négatifs tels que a + b + c = 4. Démontrer que

a2 b b2 c c2 a 1
+ 2 + 2 6 .
3a2 + b + 4ac 3b + c + 4ab 3c + a2 + 4bc
2 2 2

4011. Proposé par Abdilkadir Altinas.

Dans un triangle non équilatéral ABC, soient H l’orthocentre de ABC et J l’or-
thocentre du triangle orthique DEF de ABC (c’est-à-dire le triangle formé par
les altitudes de ABC). Si ∠BAC = 60◦ , montrer que AJ ⊥ HJ.

4012. Proposé par Leonard Giugiuc.

Soit n un entier tel que n > 3. Considérer des nombres réels ak , 1 6 k 6 n tels
que
Xn
a1 > a2 > . . . > an−1 > 1 > an > 0 et ak = n.
k=1

Démontrer que

(n − 2)(n + 1) X n(n − 1)
6 ai aj 6 .
2 2
16i<j6n

4013. Proposé par Mehmet Şahin.

Soient a, b et c les côtés du triangle ABC, D le pied de l’altitude émanant de A et
E le mi point de BC. Poser θ = ∠DAE et supposer que ∠ACB = 2θ. Démontrer
que les côtés du triangle vérifient

(a − b)2 = 2c2 − b2 .

Copyright c Canadian Mathematical Society, 2015

72/ PROBLEMS

4014. Proposé par Mihaela Berinedanu.

Soit n un nombre naturel et soient x, y et z des nombres réels positifs tels que
x + y + z + nxyz = n + 3. Démontrer que
y z x
(1 + + nyz)(1 + + nzx)(1 + + nxy) > (n + 2)3 .
x y z

4015. Proposé par Michel Bataille.

Déterminer tous les nombres réels a tels que

x sin x
a cos x + (1 − a) cos >
3 x

pour tout x non nul dans l’intervalle (− 3π 3π

2 , 2 ).

4016. Proposé par George Apostolopoulos.

Soient x, y et z des nombres réels positifs. Déterminer la valeur maximale de
l’expression
x + 2y y + 2z z + 2x
+ + .
2x + 3y + z 2y + 3z + x 2z + 3x + y

4017. Proposé par Michel Bataille.

Soit P un point sur le cercle inscrit γ du triangle ABC. Les perpendiculaires vers
BC, CA et AB, passant par P , rencontrent γ aux points U , V et W respectivement.
Démontrer qu’un des nombres P U · BC, P V · CA, P W · AB est égal à la somme
des deux autres.

4018. Proposé par Ovidiu Furdui.

Soit Z 1 Z 1
In = ··· ln(x1 x2 · · · xn ) ln(1 − x1 x2 · · · xn )dx1 dx2 · · · dxn ,
0 0

où n > 1 est entier. Démontrer que l’intégrale converge et déterminer sa valeur.

4019. Proposé par George Apostolopoulos.

Un triangle avec longueurs de côtés a, b et c possède un périmètre de longueur 3.
Démontrer que

a3 + b3 + c3 + a4 + b4 + c4 > 2(a2 b2 + b2 c2 + c2 a2 ).

Crux Mathematicorum, Vol. 41(2), February 2015

 PROBLEMS /73

4020. Proposé par Leonard Giugiuc et Daniel Sitaru.

Soit ABC un triangle. Supposer que les bissectrices internes de A, B et C inter-
sectent les côtés BC, CA et AB en D, E et F respectivement. Le cercle inscrit de
∆ABC touche les côtés BC, CA et AB en M, N et P respectivement. Démontrer
que [M N P ] 6 [DEF ], où [·] dénote la surface du triangle en question.

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

3960. Proposed by George Apostolopoulos. Correction.

Let a, b, c be nonnegative real numbers such that a + b + c = 4. Prove that

a2 b b2 c c2 a 1
+ 2 + 2 6 .
3a2 2 2 2
+ b + 4ac 3b + c + 4ab 3c + a + 4bc 2

4011. Proposed by Abdilkadir Altinas.

In non-equilateral triangle ABC, let H be the orthocenter of ABC and J be the
orthocenter of the orthic triangle DEF of ABC (that is the triangle formed by
the feet of the altitudes of ABC). If ∠BAC = 60◦ , show that AJ ⊥ HJ.

4012. Proposed by Leonard Giugiuc.

Let n be an integer with n > 3. Consider real numbers ak , 1 6 k 6 n such that
n
X
a1 > a2 > . . . > an−1 > 1 > an > 0 and ak = n.
k=1

Prove that
(n − 2)(n + 1) X n(n − 1)
6 ai aj 6 .
2 2
16i<j6n

4013. Proposed by Mehmet Şahin.

Let a, b, c be the sides of triangle ABC, D be the foot of the altitude from A and
E be the midpoint of BC. Define θ = ∠DAE and suppose that ∠ACB = 2θ.
Prove that the sides of the triangle satisfy

(a − b)2 = 2c2 − b2 .

4014. Proposed by Mihaela Berinedanu.

Let n be a natural number and let x, y and z be positive real numbers such that
x + y + z + nxyz = n + 3. Prove that
y z x
(1 + + nyz)(1 + + nzx)(1 + + nxy) > (n + 2)3
x y z
and determine when equality holds.

Copyright c Canadian Mathematical Society, 2015

74/ PROBLEMS

4015. Proposed by Michel Bataille.

Find all real numbers a such that
x sin x
a cos x + (1 − a) cos >
3 x
for every nonzero x of the interval (− 3π 3π
2 , 2 ).

4016. Proposed by George Apostolopoulos.

Let x, y, z be positive real numbers. Find the maximal value of the expression
x + 2y y + 2z z + 2x
+ + .
2x + 3y + z 2y + 3z + x 2z + 3x + y

4017. Proposed by Michel Bataille.

Let P be a point of the incircle γ of a triangle ABC. The perpendiculars to
BC, CA and AB through P meet γ again at U, V and W , respectively. Prove that
one of the numbers P U · BC, P V · CA, P W · AB is the sum of the other two.

4018. Proposed by Ovidiu Furdui.

Let Z 1 Z 1
In = ··· ln(x1 x2 · · · xn ) ln(1 − x1 x2 · · · xn )dx1 dx2 · · · dxn ,
0 0
where n > 1 is an integer. Prove that this integral converges and find its value.

4019. Proposed by George Apostolopoulos.

A triangle with side lengths a, b, c has perimeter 3. Prove that

a3 + b3 + c3 + a4 + b4 + c4 > 2(a2 b2 + b2 c2 + c2 a2 ).

4020. Proposed by Leonard Giugiuc and Daniel Sitaru.

Let ABC be a triangle and let the internal bisectors from A, B and C intersect
the sides BC, CA and AB in D, E and F , respectively. The incircle of ∆ABC
touches the sides BC, CA and AB in M, N , and P , respectively. Prove that
[M N P ] 6 [DEF ], where [·] denotes the area of the specified triangle.

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /75

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 2014: 40(2), p. 73–76,
unless otherwise specified.
An asterisk (?) after a number indicates that a problem was proposed without a solution.

2832?. Proposed by Walther Janous. [2003: 176; 2004: 184]

Let n be a positive integer, and let
3n ÇÇ å Ç åå
X 6n + 2 − j 6n + 1 − j
a(n) = (−2)j + .
j=0
j+1 j

Prove that
(a) a(n) = 3 if and only if n = 1, and
(b) the sequence {a(n)}∞n=1 is strictly increasing.

We give the solution to part (b) by C. R. Pranesachar.

We shall prove that the statement is false. In fact, there exist infinitely many
positive integers n such that a(n) > a(n + 1).
Let ÇÇ å Ç åå
X
jn+1−j n−j
b(n) = (−2) + , n > 0.
j+1 j
j>0

Also
P letn−j
c(n)

= b(6n + 1). Then a(n) = |c(n)|, as the full range of j is used. Since
n
j>0 j is the coefficient of x in the series

1
1 + x(1 + x) + x2 (1 + x)2 + · · · + xn (1 + x)n + · · · =
1 − x − x2
j n−j
is the coefficient of xn in the series
P

we see that j>0 (−2) j

1
1 + x(1 − 2x) + x2 (1 − 2x)2 + · · · + xn (1 − 2x)n + · · · = .
1 − x + 2x2
Now,
Ç å Ç å Ç å
X n+1−j X n+2−j 1 1X n+2−j
(−2)j = (−2)j−1 = − (−2)j ,
j+1 j 2 2 j
j>0 j>1 j>0

so this sum is the coefficient of xn+2 in

1 1
− .
2(1 − x) 2(1 − x + 2x2 )

Copyright c Canadian Mathematical Society, 2015

76/ SOLUTIONS

Thus b(n) is the coefficient of xn+2 in

x2 1 1 1 2x2 − 1
2
+ − 2
= +
1 − x + 2x 2(1 − x) 2(1 − x + 2x ) 2(1 − x) 2(1 − x + 2x2 )
1 x−2
=1+ + .
2(1 − x) 2(1 − x + 2x2 )

Hence b(n) = 12 + A0 αn + B0 β n , where α and β are the roots of λ2 − λ + 2 = 0,

and A0 , B0 are two fixed numbers. We have
√ √
1+i 7 1−i 7
α= , β= .
2 2
Using the initial values b(0) = 2 and b(1) = 3, we get easily that
√ √
3−i 7 3+i 7
A0 = , B0 = .
4 4
Since a(n) = |c(n)|, we compute c(n). In fact
Ç √ å Ç √ å
1 3−i 7 6n+1 3+i 7
c(n) = b(6n + 1) = + α + β 6n+1
2 4 4
Ç √ å Ç √ å
1 5+i 7 6n 5−i 7
= + α + β 6n .
2 4 4

Since Ç √ å6 √ √
6 1+i 7 9+i 7 6 9−i 7
α = = and β = ,
2 2 2
we have
Ç √ åÇ √ ån Ç √ åÇ √ ån
1 5+i 7 9+i 7 5−i 7 9−i 7
c(n) = + + , n > 1.
2 4 2 4 2

If we set √ √
5+i 7 5−i 7
A= , B= ,
4 4
then √ √
9 + 5i 7 9 − 5i 7
A2 = , B2 = .
8 8
Hence
1 1
b(6n + 1) =+ A2 (4A2 )n + B(4B 2 )n = + 4n (A2n+1 + B 2n+1 ).
2 2
√ iθ √ −iθ Ä ä
5
Now we may take A = 2e , B = 2e , where θ = arccos 4√ 2
is acute. So

1 √
c(n) = + 8n (2 2) cos(2n + 1)θ.
2

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /77

Hence
1 √
a(n) = |c(n)| = + 8n (2 2) cos(2n + 1)θ , n > 1.
2
Now we exploit the properties of θ. Surprisingly, θ has some nice ‘solution-friendly’
properties that are precisely needed. Firstly,
Å ã
2 2 25 9
cos 2θ = 2 cos θ = 2 cos θ − 1 = 2 −1= .
32 16

From this we infer that θ cannot be a rational multiple of π (the proof is left as
an exercise for the reader). Therefore, the set {cos(2n + 1)θ : n ∈ N} is dense
5
in [−1, 1]. The next property of θ that we use is as follows: since 4√ 2
> 32 , we
have cos θ > cos 30◦ , and so θ < 30◦ . Hence, if φ ∈ (θ, 30◦ ), an interval of positive
length, we may write φ = θ + τ for some suitable τ ∈ (0, 30◦ − θ).
Further
cos φ − 8 cos(φ + 2θ) = cos(θ + τ ) − 8 cos(3θ + τ )
= cos τ (cos θ − 8 sin 3θ) + sin τ (8 sin 3θ − sin θ)
= sin τ (8 sin 3θ − sin θ) > 0.

(In fact, one has sin 3θ > sin θ since 0 < θ < 3θ < 90◦ .)
Thus cos φ > 8 cos(φ + 3θ). Now the set {cos(2n + 1)θ : n ∈ N} being dense in the
interval (cos 30◦ , cos θ), we have that for some m ∈ N, we get (2m + 1)θ = 2kπ + φ,
where k ∈ N and φ ∈ (θ, 30◦ ). Hence

cos(2m + 1)θ = cos φ > 8 cos(φ + 2θ) = 8 cos(2m + 3)θ > 0.

This is sufficient to infer that c(m) > c(m + 1) > 0, and hence a(m) > a(m + 1).
Thus a(n) > a(n + 1), for infinitely many n as (2n + 1)θ visits (θ, 30◦ ) (mod 2π)
infinitely often. Also one has θ = 27◦ 550 800 (approximately) and so we can as
well extend the interval (θ, 30◦ ) to (θ, 34◦ ) safely, as 34◦ + 2θ is still less than 90◦ .
Note that we have proved the result only for positive values of c(n). It may happen
that |c(n)| > |c(n + 1)| for some negative values of c(n) also. Values of n less than
100 for which this happens are given below. These values can be obtained by
giving the above numerical value for θ and relevant values of n.

c(13) = 1305410163123, c(14) = 286249224103;

38
c(42) ≈ −2.078035580 · 10 , c(43) ≈ −1.327909464 · 1038 ;
c(55) ≈ −1.079614797 · 1050 , c(56) ≈ 1.974401435 · 1049 ;
c(84) ≈ 1.723273540 · 1076 , c(85) ≈ 4.481080748 · 1075 ;
c(97) ≈ 8.905486681 · 1087 , c(98) ≈ −5.427940879 · 1087 .

Although (b) is false, (a) may be still true and it is believed by this solver that
(a) is in fact true.

Copyright c Canadian Mathematical Society, 2015

78/ SOLUTIONS

Editor’s comments. The approach above might provide a solution to part (a).
Starting with the expression for c(n) derived above, express everything as a rational
expression in α: writing x = α6n+3 , manipulate c(n) = ±3 into monic quadratic
expressions and get a value of x as a complex number. A simple recurrence for
the real part of powers of α, in reduced form, should then yield the desired result.
The reader should work out the details to see if there is an unforeseen pitfall.

3911. Proposed by Paul Bracken.

Let x0 ∈ (0, 1 − 1/a], where a > 1, and define the sequence xn = xn−1 − x2n−1 for
n ∈ N. Prove that xn satisfies the inequalities
x0 x0
< xn < , n ∈ N.
anx0 + 1 nx0 + 1

We have received eight correct solutions. We present the solution by Arkady Alt
slightly modified by the editor.
Note first that since x0 ∈ (0, 1) and x1 − x0 = −x20 < 0, we have x1 < x0 .
Furthermore, since x0 > x20 , we have that x1 = x0 − x20 > 0. Hence, 0 < x1 < 1.
By similar argument and induction, it is easily shown that the sequence (xn ) is
strictly decreasing and xn ∈ (0, 1) for all n ∈ N.
Since
1 1 1 1 1
= = = + ,
xk xk−1 − x2k−1 xk (1 − xk−1 ) xk−1 1 − xk−1
we have
1 1 1
− =
xk xk−1 1 − xk−1
for all k ∈ N. Hence for all n ∈ N we have
n Å ã X n
1 1 X 1 1 1
− = − =
xn x0 xk xk−1 1 − xk−1
k=1 k=1
n
X 1 n n
6 = < = an,
k=1
1 − x0 1 − x0 1 − (1 − a1 )

from which we get

1 1 anx0 + 1
< + an = ,
xn x0 x0
so
x0
< xn . (1)
anx0 + 1
Using (1), we get
n n n
1 1 X 1 X 1 X 1
− = > >
xn x0 1 − xk−1 1 − xn 1 − anxx00+1
k=1 k=1 k=1
n n(anx0 + 1)
= x0 = > n,
1− anx0 +1 anx0 + 1 − x0

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /79

so
1 1 nx0 + 1
> +n= .
xn x0 x0
Hence,
x0
xn < . (2)
nx0 + 1
From (1) and (2), the proof is complete.

3912. Proposed by Michel Bataille.

Let ABC be a scalene triangle with no right angle and H as its orthocenter. If
A1 , B1 and C1 are the midpoints of BC, CA and AB respectively, prove that the
orthocenters of HAA1 , HBB1 and HCC1 are collinear.
We received seven submissions, five of which were correct and two incomplete. We
present a composite of the solutions by Šefket Arslanagić and by the proposer.

Define the point A0 to be the reflection of H in A1 . Then HBA0 C is a parallelogram

(because its diagonals bisect one another), which implies that A0 C ⊥ AC (because
BH is parallel to A0 C and perpendicular to AC). Similarly, A0 B||CH so that
A0 B ⊥ AB; consequently, AA0 is a diameter of the circumcircle Γ of ∆ABC.
Let K be the orthogonal projection of A onto the line A1 H (note that K 6= A
because ∠BAC 6= 90◦ ). This point K is on the circle γ1 with diameter AA1 and,
from the preceding remark (which implies that A0 , A1 , K, H are collinear), is also
on Γ. It follows that AK is the radical axis of the circles Γ and γ1 .
Let γ denote the Euler (or nine-point) circle of ∆ABC (which passes through the
feet of the altitudes and the midpoints of the sides). Since ∠BAC 6= 90◦ , we
have γ 6= γ1 . In addition, both γ1 and γ pass through A1 and the orthogonal
projection D of A onto BC (which are distinct points because AB 6= AC). Thus,
the line BC is the radical axis of γ1 and γ. Now, the orthocentre P of ∆AHH1 ,
which is the point of intersection of AK and BC, is the radical center of the

Copyright c Canadian Mathematical Society, 2015

80/ SOLUTIONS

circles Γ, γ, γ1 . Thus, P is on the radical axis of the circles Γ and γ. The same is
true of the orthocentres of triangles HBB1 and HCC1 . The desired result follows
immediately.

But we can deduce yet more: these three orthocentres lie on the orthic axis of
∆ABC (which, consequently, must coincide with the radical axis of Γ, γ), as we
now show. If E and F are the feet of the altitudes from the vertices B and C of
∆ABC, then they lie on the circle whose diameter is AH, and that circle, call it
γ2 , also contains K (because AK ⊥ HK). Thus the radical axis of γ2 and γ1 must
be AK, while the radical axis of γ2 and the Euler circle γ is EF . Putting these
lines together with the radical axis BC of circles γ and γ1 , we see that the radical
centre of these three circles must be the common point of AK, BC, and EF , which
we know to be P (the intersection of BC and AK). By analogous arguments, the
orthocentres of triangles HBB1 and HCC1 must likewise be the intersections of
the sides DE and DF of the orthic triangle DEF with the corresponding sides
AB and AC of the initial triangle. Of course, a triangle and its orthic triangle are
perspective from the orthocentre, whence they must be perspective from a line,
namely the orthic axis. We have just seen that the corresponding sides of the two
triangles intersect in the orthocentres of HAA1 , HBB1 , HCC1 , which completes
a second proof that these three points are collinear.

Editor’s comments. Both incomplete submissions provided neat arguments to

show that the three orthocentres satisfy one of the conditions required for the
converse of Menelaus’s theorem, but (as was pointed out in the editorial comments
following problem 3885 [2014 : 399]) a second condition must be satisfied: zero or
two of the orthocentres must lie on the sides of ∆ABC (while one or three lie on
the extensions of those sides).

3913. Proposed by Ovidiu Furdui.

Calculate
Z ∞ Z ∞
dxdy
.
0 0 (ex + ey )2

We received ten correct solutions and two incorrect submissions. We present the
solution by Madhav Modak.

Denote the repeated integral by I. Then by change of variables we have

∞ ∞
e−2y e−2x
Z Z
I= dxdy
0 0 (e−x + e−y )2
∞ e−y
−e−2y
Z Z
= (t − e−y )dtdy
0 1+e−y t2

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /81

where e−x + e−y = t, and −e−x dx = dt. Proceeding with the integral, we have
Z ∞ Z 1+e−y  1 e−y 
I= e−2y − 2 dtdy
0 e−y t t
Z ∞ −y i1+e−y
h e
= e−2y log t + dy
0 t e−y
Z ∞ h 1 + e−y  1 1 i
= e−2y log + e −y
− dy
0 e−y 1 + e−y e−y
Z ∞ Z ∞  −3y
h i e 
= e−2y log(1 + e−y ) + y dy + − e −2y
dy
0 0 1 + e−y
h e−2y i∞ Z ∞ e−2y −e−y
Z ∞
= log(1 + e−y ) − · dy + ye−2y dy
−2 0 0 −2 1 + e−y 0
Z ∞  −3y
e −2y

+ − e dy
0 1 + e−y
1 ∞ e−3y h e−2y i∞ Z ∞ e−2y
Z
1
= log 2 − dy + y − dy
2 2 0 1 + e−y −2 0 0 −2
Z ∞ −3y Z ∞
e
+ dy − e−2y dy
0 1 + e−y 0
1 ∞ e−3y 1 ∞ −2y
Z Z
1
= log 2 + dy − e dy.
2 2 0 1 + e−y 2 0

Letting 1 + e−y = w, and −e−y dy = dw, we have

1 1 −(w − 1)2 1 h e−2y i∞
Z
1
= log 2 + dw −
2 2 2 w 2 −2 0
1 2
Z
1 1 1
= log 2 + (w − 2 + )dw −
2 2 1 w 4
1 13  1
= log 2 + − 2 + log 2 − ,
2 2 2 4
so that
1
I = log 2 − .
2

3914. Proposed by George Apostolopoulos; generalized by the Editorial Board.

Let ABC be a triangle with circumradius R, inradius r and semiperimeter s, such
that s = kr. Prove that
2k R k2 − 3
√ < < .
3 3 r 12
We received 13 correct solutions. We present a hybrid of several solutions that
efficiently applied formulae (implicitly and explicitly) from O. Bottema et al., Ge-
ometric Inequalities, Wolters-Noordhoff, Groningen, 1969.

Copyright c Canadian Mathematical Society, 2015

82/ SOLUTIONS

As noted in most of the solutions, we assume that the triangle is not equilateral,
as both of the inequalities become equations if the given triangle is equilateral.
√
Using k = s/r, the left-hand inequality is equivalent to 2s < 3R 3, which is
inequality 5.3 from Bottema et al.
The right-hand inequality is equivalent to

3r(4R + r) < s2 ,

which is given both in 5.5 and 5.6 of Bottema et al.

3915. Proposed by Marcel Chiriţă.

Let M and N be points on the sides AB and AC, respectively, of triangle ABC,
and define O = BN ∩CM . Show that there are infinitely many examples (that are
not affinely equivalent) for which the areas of the four regions M BO, BCO, CN O
and AM ON are all integers.
We received four correct solutions to this problem, each utilizing a different con-
struction. We feature three of them.
Solution 1, based on the construction by Digby Smith.
For arbitrary p, q ∈ N with p > q, the numbers p2 − q 2 , 2pq, and p2 + q 2 (and any
multiples thereof) form a Pythagorean Triple. The configuration AM BCN from
the problem is defined by

BC = 4pq(p2 − q 2 )(p2 + q 2 ),
BM = CN = 4pq(p2 − q 2 )(2pq),
BN = CM = 4pq(p2 − q 2 )(p2 − q 2 ).

Then ABC is isosceles and BM C and CN B are congruent and right angled. By
their definitions, the areas of both BM C and CN B are equal to BM · CM/2 and
thus integers. If we define D to be the midpoint of BC, then BDO ∼ BN C and

N C · BD 8p2 q 2 (p2 − q 2 ) · 2pq(p2 − q 2 )(p2 + q 2 )

DO = = = 4p2 q 2 (p2 + q 2 ).
BN 4pq(p2 − q 2 )2

The area of BCO is equal to BD · DO/2 and therefore integer. It follows that the
areas of BM O and CN O are also integers. Finally ADC ∼ BN C and thus

BN · DC 4pq(p2 − q 2 )2 · 2pq(p2 − q 2 )(p2 + q 2 )

AD = = = (p2 − q 2 )2 (p2 + q 2 ).
NC 8p2 q 2 (p2 − q 2 )

The area of ABC is equal to BC · AD/2 and therefore integer. By subtracting

the areas of BM O, BCO, and CN O, we obtain finally that the area of AM ON
is also an integer. As p, q can be arbitrarily chosen, we obtain an infinite number
of configurations that are not affinely equivalent (e.g. choose all pairs p, q with
gcd (p, q) = 1).

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /83

Solution 2, abridged version of the solution by the Missouri State University Prob-
lem Solving Group.
Let A = (0, 0), B = (1, 0), C = (0, 1), M = (a, 0), and N = (0, b), where a and
b are rational and 0 < a < b < 1. The equations of the lines BN and CM have
rational coefficients, so the coordinates of O are rational. The area of a triangle
with vertices (x1 , y1 ), (x2 , y2 ), and (x3 , y3 ) is
Ñ é
1 x1 y1
1
det 1 x2 y2 .
2
1 x3 y3
Therefore, the areas of M BO, BCO, CN O, and AM ON are all rational. By
stretching the triangle ABC, the corresponding areas can be made to be integers.
Since stretching does not alter the ratios AM/M B and AN/N C, the configurations
are not affinely equivalent for distinct choices of a and b.

Solution 3, by Titu Zvonaru.

Let a, b, m, n be positive integers and let ABC be a triangle with BC = 2a and
hA = b(m + 1)(n + 1)(m + n + 1). Choose the points M and N on AB and AC
such that
BM 1 CN 1
= , = .
BA m+1 CA n+1
Denote by [XY . . . Z] the area of the polygon XY . . . Z. Then
[ABC] [ABC]
[BM C] = , [CN B] = .
m+1 n+1
Suppose that AO intersects BC at A0 . By Van Aubel’s Theorem for Cevian
triangles we obtain
AO AM AN
0
= + =m+n
OA MB NC
and therefore OA0 = AA0 /(m + n + 1). It follows that
[ABC]
[BOC] = .
m+n+1
Thus the areas [ABC], [BM C], [CN B], and [BOC] are all integers and by taking
differences of these areas so are [M BO], [CN O], and [AM ON ].

3916. Proposed by Nathan Soedjak.

Let a, b, c be positive real numbers. Prove that
Å ã2 Å ã2  2 Å ã
ab bc ca ab + bc + ca 2
+ + ≥3 .
c a b a+b+c

There were 23 correct solutions, with two solutions from one solver, as well as a
Maple verification. We present a sampling of the different approaches.

Copyright c Canadian Mathematical Society, 2015

84/ SOLUTIONS

Solution 1, by Mohammed Aassila.

Note that x2 + y 2 + z 2 > xy + yz + zx and (x + y + z)2 > 3(xy + yz + zx) for real
x, y, z. The left side of the inequality is not less than b2 + c2 + a2 . However

(a2 + b2 + c2 )(a + b + c)2 > (ab + bc + ca)[3(ab + bc + ca)] = 3(ab + bc + ca)2 ,

and the desired result follows.

Solution 2, by Michel Bataille.

By homogeneity, we may suppose that a + b + c = 1. The inequality is then
equivalent to

(ab)4 + (bc)4 + (ca)4 > 3(a2 b2 c2 )(ab + bc + ca)2 .

Observe that
1 4
x4 + y 4 + z 4 = [(x + x4 + y 4 + z 4 ) + (x4 + y 4 + y 4 + z 4 ) + (x4 + y 4 + z 4 + z 4 )]
4
> x2 yz + xy 2 z + xyz 2 = xyz(x + y + z),

and (x + y + z)2 > 3(xy + yz + zx). Applying these inequalities leads to

(ab)4 + (bc)4 + (ca)4 = [(ab)4 + (bc)4 + (ca)4 ][(a + b + c)2 ]

> [a2 b2 c2 (ab + bc + ca)][3(ab + bc + ca)]
= 3(a2 b2 c2 )(ab + bc + ca)2 ,

as desired.

Solution 3, by Dionne Bailey, Elsie Campbell, and Charles Dimminnie; Angel

Plaza; Cao Minh Quang; and Edmund Swylan, independently.
Since x2 + y 2 + z 2 > xy + yz + zx,

ab bc ca (ab)2 + (bc)2 + (ca)2 abc(a + b + c)

+ + = > = a + b + c.
c a b abc abc
Using either the convexity of the function x2 or the inequality of the root-mean-
square and arithmetic mean, we find that
Å ã2 Å ã2 Å ã
ab bc  ca 2 1 ab bc ca 2
+ + > + +
c a b 3 c a b
1 1 (a + b + c)4
> (a + b + c)2 =
3 3 (a + b + c)2
Å ã
[3(ab + bc + ca)]2 ab + bc + ca 2
> =3 .
3(a + b + c)2 a+b+c

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /85

Solution 4, by Paolo Perfetti.

Since Å ã
1 x2 y 2 x2 y 2 y2 z2 z 2 x2
+ + + > xy
4 z2 z2 x2 y2
by the arithmetic-geometric means inequality, we can follow the strategy of Solu-
tion 2 to obtain
Å ã2 Å ã2  2
ab bc ca 3(ab + bc + ca)2 3(ab + bc + ca)2
+ + > ab + bc + ca = >
c a b 3(ab + bc + ca) (a + b + c)2

as desired.

Solution 5 by Henry Ricardo.

We have
Å ã2 Å ã2  2 ï Å ã Å 2 ã Å 2 ãò
ab bc ca 1 2 b2 c2 2 a c2 2 b a2
+ + = a + + b + + c +
c a b 2 c2 b2 c2 a2 a2 b2
> a2 + b2 + c2
(a + b + c)2
>
3
Å ã
ab + bc + ca 2
>3 .
a+b+c

3917. Proposed by Peter Y. Woo.

Given a circle Z, its center O, and a point A on Z, with only a long unmarked
ruler, and no compass, can you draw:
i) points B,C and D on Z so that ABCD is a square?
ii) the square AOBA0 ?
iii) the points B, W 00 , W and W 0 on Z such that angles AOB, AOW 00 , AOW
and AOW 0 are 90◦ , 60◦ , 45◦ and 30◦ ?
There were five correct solutions to this problem. We feature the one by the Mis-
souri State University Problem Solving Group.
We need the following basic construction: Given three collinear points A, B, C such
←→
that AB = BC and a point P not on AC, we want to construct a line through
←→ −→
P parallel to AC. To do this, we choose a point Q on the ray AP such that P is
←→ ←→
between A and Q. Denote the intersection of BQ and CP by R and denote the
←→ ←→ ←
→
intersection of AR and QC by S. We claim that P S is the line we seek. By Ceva’s
theorem,
AB CS QP
· · = 1,
BC SQ P A

Copyright c Canadian Mathematical Society, 2015

86/ SOLUTIONS

QP QS ←→ ←→
but since AB = BC this yields = and therefore P S and AC are parallel.
PA SC
←→
i) The intersection of OA and Z gives C. Choose any point X on Z other than A
or C and use the basic construction above to obtain a line through X parallel
to line AC. If this line only meets Z in a single point let B = X and D be
the intersection of line OB and Z. If the line meets Z in two distinct points,
←→ ←→
S and T , let U be the intersection of AS and CT . Then B and D are the
←→ ←→
intersections of U O with Z. Note that by symmetry, U O is perpendicular to
←→
AC, which makes ABCD a square.
←→
ii) Using the basic construction, we draw a line through A parallel to BD and a
←→
line through B parallel to AC. Their intersection is the point A0 .
iii) We constructed B in part i). The point where OA0 meets Z gives W . Let
E denote the intersection of OA0 and AB. Using the basic construction, if
←→
we draw a line ` through E parallel to AC, the point of intersection of ` and
Z that lies between A and B gives W 0 (note that ` bisects OB, which gives
←→ 00
sin(^AOW 0 ) = 1/2). Similarly, a line through E parallel to BD gives W .

3918. Proposed by George Apostolopoulos.

Let a, b and c be positive real numbers such that a2 + b2 + c2 = 1. Prove that
» √
2/3 2/3 2/3
2+ 3
(ab) + (bc) + (ac) < .
3

We received 22 correct solutions and one incorrect solution. We present the solu-
tion by Cristinel Mortici, slightly modified by the editor.
Recall the Power Mean Inequality: for x, y, z > 0 and m > n
Å ã1/m Å ã1/n
xm + y m + z m xn + y n + z n
> .
3 3

The Power Mean Inequality with m = 1 and n = 2/3 gives

Ç å3/2
(ab)2/3 + (bc)2/3 + (ca)2/3 ab + bc + ca a2 + b2 + c2 1
6 6 = .
3 3 3 3

It follows that Ä ä1/2

(ab)2/3 + (bc)2/3 + (ca)2/3 6 31/6 .
Finally, the Geometric Mean-Arithmetic Mean Inequality gives us
√ √
1/6
√ 1/3 1+1+ 3 2+ 3
3 = (1 · 1 · 3) < = ,
3 3
completing the proof.

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /87

3919. Proposed by Michel Bataille.

Let I be the incentre of triangle ABC. The line segment AI meets the incircle
at M and the perpendicular to AM at M intersects BI at N . If P is a point of
the line AI, prove that P C is perpendicular to AI if and only if P N is parallel to
BM .
We received seven correct solutions. We present a composite of the similar solu-
tions by Šefket Arslanagić and by Peter Woo.
On the one hand,
IM IP
P N ||BM ⇐⇒ ∆BIM ∼ ∆N IP ⇐⇒ = .
IB IN
On the other hand,
IP
P C ⊥ AI ⇐⇒ ∆CP I has a right angle at P ⇐⇒ cos ∠P IC = .
IC
Let D be the foot of the perpendicular from I to BC. Then ID = IM = r (the
inradius), and in right triangle BDI we have IB = sinr B , whence
2

IM r B
=  = sin . (1)
IB r 2
B
sin 2

In right triangle IDC, we have

r
IC = . (2)
sin C2
Because ∠N IM is exterior to ∆BIA, ∠N IM = A
2 + B
2 = 90◦ − C
2; consequently,
in right triangle N M I we have
C IM r
cos ∠N IM = sin = =
2 IN IN
and, with the help of equation (2),
r
IN = = IC. (3)
sin C2
Because ∠P IC is an exterior angle of ∆AIC, ∠P IC = A
2 + C2 = 90◦ − B2 , whence
B
cos ∠P IC = sin . (4)
2
Putting the pieces together, we deduce
IM IP
P N ||BM ⇐⇒ =
IB IN
B IM IP
⇐⇒ sin = = (from (1) and (3))
2 IB IC
IP
⇐⇒ = cos ∠P IC (from (4))
IC
⇐⇒ P C ⊥ AI.

Copyright c Canadian Mathematical Society, 2015

88/ SOLUTIONS

3920. Proposed by Alina Sı̂ntǎmǎrian.

Evaluate
∞
X 16n2 + 20n + 7
.
n=0
(4n + 2)!

There were 15 submitted solutions for this problem, 14 of which were correct. We
present three solutions, representative of the two main solution methods utilized
together with one variant.
Solution 1, by the AN-anduud Problem Solving Group.
Consider the following two power series,
∞ ∞
X x2n+1 X xn
sin x = (−1)n · , and ex = , x ∈ R.
n=0
(2n + 1)! n=0
n!

Hence, we have
∞ ∞ Å ã
X 1 X 1 1
sin 1 = (−1)n · = − ,
n=0
(2n + 1)! n=0 (4n + 1)! (4n + 3)!

and
∞
X 1
e= .
n=1
n!

Using the above considerations, we get

∞ ∞
X 16n2 + 20n + 7 X (4n + 2)(4n + 1) + 2(4n + 2) + 1
=
n=0
(4n + 2)! n=0
(4n + 2)!
∞ Å ã
X 1 2 1
= + +
n=0
(4n)! (4n + 1)! (4n + 2)!
∞
X 1 ∞ Å ã
X 1 1
= + −
n=0
n! n=0 (4n + 1)! (4n + 3)!
= e + sin 1.

Solution 2, by the group of Dionne Bailey, Elsie Campbell, and Charles Diminnie.
To begin, we note that for n > 0,

16n2 + 20n + 7 (4n + 2) (4n + 1) + 2 (4n + 2) + 1

=
(4n + 2)! (4n + 2)!
1 2 1
= + + ,
(4n)! (4n + 1)! (4n + 2)!

Crux Mathematicorum, Vol. 41(2), February 2015

 SOLUTIONS /89

and hence,
∞ ∞ ∞ ∞
X 16n2 + 20n + 7 X 1 X 1 X 1
= +2 +
n=0
(4n + 2)! n=0
(4n)! n=0
(4n + 1)! n=0 (4n + 2)!

(since the Ratio Test easily confirms that each of the three series on the right
converges).
The remainder of this solution depends on the following known series:
∞ ∞
X (−1)k X (−1)k
sin 1 = , cos 1 = ,
(2k + 1)! (2k)!
k=0 k=0
∞ ∞
X 1 X 1
sinh 1 = , cosh 1 = .
(2k + 1)! (2k)!
k=0 k=0

Since we have
® ®
k 2 if k is even k+1 0 if k is even
(−1) + 1 = and (−1) +1= ,
0 if k is odd 2 if k is odd
we obtain:

∞ ∞ ∞
X (−1)k + 1 X 2 X 1
sin 1 + sinh 1 = = =2 ,
(2k + 1)! n=0
[2(2n) + 1]! n=0
(4n + 1)!
k=0
∞ ∞ ∞
X (−1)k + 1 X 2 X 1
cos 1 + cosh 1 = = =2 ,
(2k)! n=0
[2(2n)]! n=0
(4n)!
k=0
∞ ∞ ∞
X (−1)k+1 + 1 X 2 X 1
− cos 1 + cosh 1 = = =2 .
(2k)! n=0
[2(2n + 1)]! n=0
(4n + 2)!
k=0

Therefore, we obtain,
∞
X 16n2 + 20n + 7 cos 1 + cosh 1 − cos 1 + cosh 1
= + (sin 1 + sinh 1) +
n=0
(4n + 2)! 2 2
= sin 1 + sinh 1 + cosh 1
e − e−1 e + e−1
= sin 1 + +
2 2
= sin 1 + e.

Solution 3, by Paolo Perfetti.

First, we have:
16n2 + 20n + 7 1 2 1
= + + .
(4n + 2)! (4n)! (4n + 1)! (4n + 2)!

Copyright c Canadian Mathematical Society, 2015

90/ SOLUTIONS

Let
∞
X x4n
f (x) = ,
n=0
(4n)!
so that we obtain:
∞ ∞ ∞
X x4n−1 X x4n−2 X x4n−2
f 0 (x) = , f 00 (x) = , f 000 (x) =
n=1
(4n − 1)! n=1
(4n − 2)! n=1
(4n − 2)!
∞ ∞
X x4n−4 X x4n
f iv (x) = = = f (x).
n=1
(4n − 4)! n=0 (4n)!

Thus f (x) satisfies f iv (x) = f (x), f (0) = 1, f 0 (0) = 0, f 00 (0) = 0, f 000 (0) = 0,
whose unique solution is f (x) = 12 cosh x + 21 cos x. Evaluating, we get
∞
1 1 X 1
f (1) = cosh 1 + cos 1 = .
2 2 n=0
(4n)!

Moreover, if we define
∞
X x4n+1
g(x) = ,
n=0
(4n + 1)!
P∞
we get g(1) = 1
n=0 (4n+1)! and g 0 (x) = f (x), g(0) = 0. This implies

1 1 1 1
g(x) = sinh x + sin x, g(1) = sinh 1 + sin 1.
2 2 2 2
Finally, defining
∞
X x4n+2
h(x) = ,
n=0
(4n + 2)!
P∞
we get h(1) = 1
n=0 (4n+2)! and h0 (x) = g(x), h(0) = 0. This implies

1 1 1 1
h(x) = cosh x − cos x, h(1) = cosh 1 − cos 1.
2 2 2 2
Summing up the terms, we obtain

f (1) + 2g(1) + h(1) = e + sin 1.

Editor’s Comment. The presented solutions illustrate three techniques: rearrange

the summations wisely to get a simple expression, rearrange the summations and
then recall other atypical power series that make things work, and solve a couple of
DEs to avoid having to work too much with power series. Wagon commented that
the sum can be explicitly computed when the numerator is an arbitrary quadratic
in n.

Crux Mathematicorum, Vol. 41(2), February 2015

 AUTHORS’ INDEX /91

AUTHORS’ INDEX
Solvers and proposers appearing in this issue
(Bold font indicates featured solution.)

Proposers
Abdilkadir Altintaş, mathematics teacher, Emirdağ, Turkey: 4011
Mihaela Berindeanu, Bucharest, Romania: 4014
George Apostolopoulos, Messolonghi, Greece: 4016, 4019
Michel Bataille, Rouen, France: 4015, 4017
Ovidiu Furdui, Campia Turzii, Cluj, Romania: 4018
Leonard Giugiuc, Romania : 4012
Leonard Giugiuc and Daniel Sitaru, Romania : 4020
Mehmet Şahin, Ankara, Turkey: 4013

Solvers - individuals
Mohammed Aassila, Strasbourg, France: 3916, 3918
Arkady Alt, San Jose, CA, USA : 3911, 3914, 3916, 3918, 3920
George Apostolopoulos, Messolonghi, Greece : 3914, 3918
Šefket Arslanagić, University of Sarajevo, Sarajevo, Bosnia and Herzegovina : CC107,
CC108, CC110, 3912, 3914, 3916, 3918, 3919
Fernando Ballesta Yagüe, Murcia, Spain: CC107, CC108
Michel Bataille, Rouen, France : OC156, OC157, 3911, 3912, 3913, 3914, 3916, 3917,
3918, 3919, 3920
Paul Bracken,University of Texas, Edinburg, TX : 3911
Scott Brown, Auburn University Montgomery, Montgomery, AL, USA: CC107
Marcel Chiritiţǎ, Bucharest, Romania : 3915
Matei Coiculescu, East Lyme High School, East Lyme, CT, USA : CC107, CC108
Roberto de la Cruz, Centre de Recerca Matemàtica, Barcelona, Spain : 3920
Prithwijit De, Homi Bhabha Centre for Science Education, Mumbai, India : 3911, 3912
Ovidiu Furdui, Technical University of Cluj-Napoca, Romania : 3913
Oliver Geupel, Brühl, NRW, Germany: OC159, 3913
John G. Heuver, Grande Prairie, AB : OC156
Václav Konečný, Big Rapids, MI, USA: 3919
Kee-Wai Lau, Hong Kong, China : 3911, 3913, 3916, 3918, 3920
Carl Libis, Columbia Southern University, Orange Beach, AL, USA : 3920
Joseph M. Ling, University of Calgary, Calgary, AB : OC157
Salem Malikić, student, Simon Fraser University, Burnaby, BC : 3914, 3916, 3918
Phil McCartney, Northern Kentucky University, Highland Heights, KY, USA : 3916,
3918, 3920
Dragoljub Milošević, Gornji Milanovac, Serbia : CC107, 3914, 3916
Madhav R. Modak, formerly of Sir Parashurambhau College, Pune, India : 3913, 3916,
3918, 3919, 3920
Cristinel Mortici, Valahia University of Târgovişte, Romania: 3911, 3918
Ricard Peiró i Estruch, IES Abastos, València, Spain: 3914, 3918
Paolo Perfetti, Dipartimento di Matematica, Università degli studi di Tor Vergata Roma,
Rome, Italy : CC109, 3913, 3916 (2 solutions), 3918, 3920

Copyright c Canadian Mathematical Society, 2015

92/ AUTHORS’ INDEX

Angel Plaza, University of Las Palmas de Gran Canaria, Spain : 3916

C.R. Pranesachar, Indian Institute of Science, Bangalore, India : 2832b
Cao Minh Quang, Nguyen Binh Khiem High School, Vinh Long, Vietnam : 3916, 3918
Henry Ricardo, Tappan, NY, USA : CC108, CC110, OC157, 3916, 3918
Alina Sı̂ntămărian, Technical University of Cluj-Napoca, Cluj-Napoca, Romania : 3920
Digby Smith, Mount Royal University, Calgary, AB : CC107, CC108, CC110, 3911, 3913,
3915, 3916, 3918, 3920
Nathan Soedjak, Springfield, MO, USA : 3916
Albert Stadler, Herliberg, Switzerland: 3913, 3916, 3918, 3920
Edmund Swylan, Riga, Latvia : 3912, 3914, 3916, 3917, 3918, 3919
Daniel Văcaru, Pitesti, Romania : CC107, OC156, 3914
Stan Wagon, Macalester College, St. Paul, MN, USA : 3916, 3918
Peter Y. Woo, Biola University, La Mirada, CA, USA : 3914, 3916, 3917, 3919
Roger Zarnowski, Northern Kentucky University, Highland Heights, KY, USA : 3911
Titu Zvonaru, Cománeşti, Romania : CC107, CC108, CC110, OC156, 3912, 3914, 3915,
3916, 3917, 3918, 3919

Solvers - collaborations
AN-anduud Problem Solving Group, Ulaanbaatar, Mongolia : 3913, 3914, 3916, 3918,
3920
Dionne Bailey, Elsie Campbell, and Charles R. Diminnie, Angelo State University,
San Angelo, USA : 3918, 3916, 3920
D. M. Bătineţu-Giurgiu, Bucharest, Titu Zvonaru, Cománeşti, and Neculai Stanciu,
Buzǎu, Romania : 3918
D. M. Bătineţu-Giurgiu, Bucharest and Neculai Stanciu, Buzǎu, Romania : 3914, 3916
Missouri State University Problem Solving Group : 3913, 3915, 3917, 3920

Crux Mathematicorum, Vol. 41(2), February 2015