Math Problem Book I

compiled by

Kin Y. Li

Department of Mathematics

Hong Kong University of Science and Technology

 c 2001 Hong Kong Mathematical Society IMO(HK) Committee.
Copyright
Printed in Hong Kong
 Preface

There are over fifty countries in the world nowadays that hold math-
ematical olympiads at the secondary school level annually. In Hungary,
Russia and Romania, mathematical competitions have a long history, dat-
ing back to the late 1800s in Hungarys case. Many professional or ama-
teur mathematicians developed their interest in math by working on these
olympiad problems in their youths and some in their adulthoods as well.

The problems in this book came from many sources. For those involved
in international math competitions, they no doubt will recognize many of
these problems. We tried to identify the sources whenever possible, but
there are still some that escape us at the moment. Hopefully, in future
editions of the book we can fill in these missing sources with the help of the
knowledgeable readers.

This book is for students who have creative minds and are interested in
mathematics. Through problem solving, they will learn a great deal more
than school curricula can offer and will sharpen their analytical skills. We
hope the problems collected in this book will stimulate them and seduce
them to deeper understanding of what mathematics is all about. We hope
the international math communities support our efforts for using these bril-
liant problems and solutions to attract our young students to mathematics.

Most of the problems have been used in practice sessions for students
participated in the Hong Kong IMO training program. We are especially
pleased with the efforts of these students. In fact, the original motivation
for writing the book was to reward them in some ways, especially those who
worked so hard to become reserve or team members. It is only fitting to
list their names along with their solutions. Again there are unsung heros

iii
who contributed solutions, but whose names we can only hope to identify
in future editions.
As the title of the book suggest, this is a problem book. So very little
introduction materials can be found. We do promise to write another book
presenting the materials covered in the Hong Kong IMO training program.
This, for certain, will involve the dedication of more than one person. Also,
this is the first of a series of problem books we hope. From the results of
the Hong Kong IMO preliminary contests, we can see waves of new creative
minds appear in the training program continuously and they are younger
and younger. Maybe the next problem book in the series will be written by
our students.

Finally, we would like to express deep gratitude to the Hong Kong

Quality Education Fund, which provided the support that made this book
possible.

Kin Y. Li
Hong Kong
April, 2001

iv
 Advices to the Readers

The only way to learn mathematics is to do mathematics. In this

book, you will find many math problems, ranging from simple to challenging
problems. You may not succeed in solving all the problems. Very few
people can solve them all. The purposes of the book are to expose you to
many interesting and useful mathematical ideas, to develop your skills in
analyzing problems and most important of all, to unleash your potential
of creativity. While thinking about the problems, you may discover things
you never know before and putting in your ideas, you can create something
you can be proud of.

To start thinking about a problem, very often it is helpful to look at

the initial cases, such as when n = 2, 3, 4, 5. These cases are simple enough
to let you get a feeling of the situations. Sometimes, the ideas in these
cases allow you to see a pattern, which can solve the whole problem. For
geometry problems, always draw a picture as accurate as possible first.
Have protractor, ruler and compass ready to measure angles and lengths.

Other things you can try in tackling a problem include changing the
given conditions a little or experimenting with some special cases first.
Sometimes may be you can even guess the answers from some cases, then
you can study the form of the answers and trace backward.

Finally, when you figure out the solutions, dont just stop there. You
should try to generalize the problem, see how the given facts are necessary
for solving the problem. This may help you to solve related problems later
on. Always try to write out your solution in a clear and concise manner.
Along the way, you will polish the argument and see the steps of the so-
lutions more clearly. This helps you to develop strategies for dealing with
other problems.

v
 The solutions presented in the book are by no means the only ways
to do the problems. If you have a nice elegant solution to a problem and
would like to share with others (in future editions of this book), please send
it to us by email at [email protected] . Also if you have something you cannot
understand, please feel free to contact us by email. We hope this book will
increase your interest in math.
Finally, we will offer one last advice. Dont start with problem 1. Read
the statements of the problems and start with the ones that interest you the
most. We recommend inspecting the list of miscellaneous problems first.
Have a fun time.

vi
 Table of Contents

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Advices to the Readers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Algebra Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Geometry Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Number Theory Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Combinatorics Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .24
Miscellaneous Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Solutions to Algebra Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35
Solutions to Geometry Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .69
Solutions to Number Theory Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Solutions to Combinatorics Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Solutions to Miscellaneous Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
 Contributors

Chan Kin Hang, 1998, 1999, 2000, 2001 Hong Kong team member
Chan Ming Chiu, 1997 Hong Kong team reserve member
Chao Khek Lun, 2001 Hong Kong team member
Cheng Kei Tsi, 2001 Hong Kong team member
Cheung Pok Man, 1997, 1998 Hong Kong team member
Fan Wai Tong, 2000 Hong Kong team member
Fung Ho Yin, 1997 Hong Kong team reserve member
Ho Wing Yip, 1994, 1995, 1996 Hong Kong team member
Kee Wing Tao, 1997 Hong Kong team reserve member
Lam Po Leung, 1999 Hong Kong team reserve member
Lam Pei Fung, 1992 Hong Kong team member
Lau Lap Ming, 1997, 1998 Hong Kong team member
Law Ka Ho, 1998, 1999, 2000 Hong Kong team member
Law Siu Lung, 1996 Hong Kong team member
Lee Tak Wing, 1993 Hong Kong team reserve member
Leung Wai Ying, 2001 Hong Kong team member
Leung Wing Chung, 1997, 1998 Hong Kong team member
Mok Tze Tao, 1995, 1996, 1997 Hong Kong team member
Ng Ka Man, 1997 Hong Kong team reserve member
Ng Ka Wing, 1999, 2000 Hong Kong team member
Poon Wai Hoi, 1994, 1995, 1996 Hong Kong team member
Poon Wing Chi, 1997 Hong Kong team reserve member
Tam Siu Lung, 1999 Hong Kong team reserve member
To Kar Keung, 1991, 1992 Hong Kong team member
Wong Chun Wai, 1999, 2000 Hong Kong team member
Wong Him Ting, 1994, 1995 Hong Kong team member
Yu Ka Chun, 1997 Hong Kong team member
Yung Fai, 1993 Hong Kong team member

ix
Problems
 Algebra Problems

Polynomials

1. (Crux Mathematicorum, Problem 7) Find (without calculus) a fifth

degree polynomial p(x) such that p(x) + 1 is divisible by (x 1)3 and
p(x) 1 is divisible by (x + 1)3 .

2. A polynomial P (x) of the n-th degree satisfies P (k) = 2k for k =

0, 1, 2, . . . , n. Find the value of P (n + 1).

3. (1999 Putnam Exam) Let P (x) be a polynomial with real coefficients

such that P (x) 0 for every real x. Prove that

P (x) = f1 (x)2 + f2 (x)2 + + fn (x)2

for some polynomials f1 (x), f2 (x), . . . , fn (x) with real coefficients.

4. (1995 Russian Math Olympiad) Is it possible to find three quadratic

polynomials f (x), g(x), h(x) such that the equation f (g(h(x))) = 0 has
the eight roots 1, 2, 3, 4, 5, 6, 7, 8?

5. (1968 Putnam Exam) Determine all polynomials whose coefficients are

all 1 that have only real roots.

6. (1990 Putnam Exam) Is there an infinite sequence a0 , a1 , a2 , . . . of

nonzero real numbers such that for n = 1, 2, 3, . . . , the polynomial
Pn (x) = a0 + a1 x + a2 x2 + + an xn has exactly n distinct real roots?

7. (1991 Austrian-Polish Math Competition) Let P (x) be a polynomial

with real coefficients such that P (x) 0 for 0 x 1. Show that
there are polynomials A(x), B(x), C(x) with real coefficients such that
(a) A(x) 0, B(x) 0, C(x) 0 for all real x and
(b) P (x) = A(x) + xB(x) + (1 x)C(x) for all real x.
(For example, if P (x) = x(1x), then P (x) = 0+x(1x)2 +(1x)x2 .)

1
 8. (1993 IMO) Let f (x) = xn + 5xn1 + 3, where n > 1 is an integer.
Prove that f (x) cannot be expressed as a product of two polynomials,
each has integer coefficients and degree at least 1.

9. Prove that if the integer a is not divisible by 5, then f (x) = x5 x + a

cannot be factored as the product of two nonconstant polynomials with
integer coefficients.

10. (1991 Soviet Math Olympiad) Given 2n distinct numbers a1 , a2 , . . . , an,

b1 , b2 , . . . , bn , an n n table is filled as follows: into the cell in the i-th
row and j-th column is written the number ai + bj . Prove that if the
product of each column is the same, then also the product of each row
is the same.

11. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two distinct collections of n pos-

itive integers, where each collection may contain repetitions. If the two
collections of integers ai +aj (1 i < j n) and bi +bj (1 i < j n)
are the same, then show that n is a power of 2.

Recurrence Relations

12. The sequence xn is defined by

2 + xn
x1 = 2, xn+1 = , n = 1, 2, 3, . . . .
1 2xn
1
Prove that xn 6= 2 or 0 for all n and the terms of the sequence are all
distinct.

13. (1988 Nanchang City Math Competition) Define a1 = 1, a2 = 7 and

a2n+1 1
an+2 = for positive integer n. Prove that 9an an+1 + 1 is a
an
perfect square for every positive integer n.

14. (Proposed by Bulgaria for 1988 IMO) Define a0 = 0, a1 = 1 and an =

2an1 + an2 for n > 1. Show that for positive integer k, an is divisible
by 2k if and only if n is divisible by 2k .

2
15. (American Mathematical Monthly, Problem E2998) Let x and y be
xn y n
distinct complex numbers such that is an integer for some
xy
xn y n
four consecutive positive integers n. Show that is an integer
xy
for all positive integers n.

Inequalities

16. For real numbers a1 , a2 , a3 , . . . , if an1 + an+1 2an for n = 2, 3, . . . ,

then prove that

An1 + An+1 2An for n = 2, 3, . . . ,

where An is the average of a1 , a2 , . . . , an .

17. Let a, b, c > 0 and abc 1. Prove that

a b c
+ + a + b + c.
c a b

18. (1982 Moscow Math Olympiad) Use the identity 13 + 23 + + n3 =

n2 (n + 1)2
to prove that for distinct positive integers a1 , a2 , . . . , an ,
4

(a71 + a72 + + a7n ) + (a51 + a52 + + a5n ) 2(a31 + a32 + + a3n )2 .

Can equality occur?

19. (1997 IMO shortlisted problem) Let a1 an an+1 = 0 be a

sequence of real numbers. Prove that
v
u n n
uX X
t ak k( ak ak+1 ).
k=1 k=1

3
20. (1994 Chinese Team Selection Test) For 0 a b c d e and
a + b + c + d + e = 1, show that
1
ad + dc + cb + be + ea .
5

21. (1985 Wuhu City Math Competition) Let x, y, z be real numbers such
that x + y + z = 0. Show that

6(x3 + y 3 + z 3 )2 (x2 + y 2 + z 2 )3 .

22. (1999 IMO) Let n be a fixed integer, with n 2.

(a) Determine the least constant C such that the inequality
X  X 4
xi xj (x2i + x2j ) C xi
1i<jn 1in

holds for all nonnegative real numbers x1 , x2 , . . . , xn .

(b) For this constant C, determine when equality holds.

23. (1995 Bulgarian Math Competition) Let n 2 and 0 xi 1 for

i = 1, 2, . . . , n. Prove that
hni
(x1 + x2 + + xn ) (x1 x2 + x2 x3 + + xn1 xn + xn x1 ) ,
2
where [x] is the greatest integer less than or equal to x.

24. For every triplet of functions f, g, h : [0, 1] R, prove that there are
numbers x, y, z in [0, 1] such that

1
|f (x) + g(y) + h(z) xyz| .
3

25. (Proposed by Great Britain for 1987 IMO) If x, y, z are real numbers
such that x2 + y 2 + z 2 = 2, then show that x + y + z xyz + 2.

4
26. (Proposed by USA for 1993 IMO) Prove that for positive real numbers
a, b, c, d,
a b c d 2
+ + + .
b + 2c + 3d c + 2d + 3a d + 2a + 3b a + 2b + 3c 3

27. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be 2n positive real numbers such

that
(a) a1 a2 an and
(b) b1 b2 bk a1 a2 ak for all k, 1 k n.
Show that b1 + b2 + + bn a1 + a2 + + an .

28. (Proposed by Greece for 1987 IMO) Let a, b, c > 0 and m be a positive
integer, prove that
 m1
am bm cm 3 a+b+c
+ + .
b+c c+a a+b 2 3

29. Let a1 , a2 , . . . , an be distinct positive integers, show that

a1 a2 an 1
+ + + n 1 n.
2 8 n2 2

30. (1982 West German Math Olympiad) If a1 , a2 , . . . , an > 0 and a =

a1 + a2 + + an , then show that
n
X ai n
.
2a ai 2n 1
i=1

a3 b3 c3 a2 + b2 + c2
31. Prove that if a, b, c > 0, then + + .
b+c c+a a+b 2
32. Let a, b, c, d > 0 and
1 1 1 1
+ + + = 1.
1 + a4 1 + b4 1 + c4 1 + d4

5
 Prove that abcd 3.

33. (Due to Paul Erdos) Each of the positive integers a1 , . . . , an is less than
1951. The least common multiple of any two of these is greater than
1951. Show that
1 1 n
++ <1+ .
a1 an 1951

34. A sequence (Pn ) of polynomials is defined recursively as follows:

x Pn (x)2
P0 (x) = 0 and for n 0, Pn+1 (x) = Pn (x) + .
2
Prove that
2
0 x Pn (x)
n+1
for every nonnegative integer n and all x in [0, 1].

35. (1996 IMO shortlisted problem) Let P (x) be the real polynomial func-
tion, P (x) = ax3 + bx2 + cx + d. Prove that if |P (x)| 1 for all x such
that |x| 1, then
|a| + |b| + |c| + |d| 7.

36. (American Mathematical Monthly, Problem 4426) Let P (z) = az 3 +

bz 2 + cz + d, where a, b, c, d arecomplex numbers with |a| = |b| = |c| =
|d| = 1. Show that |P (z)| 6 for at least one complex number z
satisfying |z| = 1.

37. (1997 Hungarian-Israeli Math Competition) Find all real numbers

with the following property:
  for any positive integer n, there exists an
 m 1
integer m such that   < ?
n 3n
38. (1979 British Math Olympiad) If n is a positive integer, denote by p(n)
the number of ways of expressing n as the sum of one or more positive
integers. Thus p(4) = 5, as there are five different ways of expressing
4 in terms of positive integers; namely

1 + 1 + 1 + 1, 1 + 1 + 2, 1 + 3, 2 + 2, and 4.

6
 Prove that p(n + 1) 2p(n) + p(n 1) 0 for each n > 1.

Functional Equations

39. Find all polynomials f satisfying f (x2 ) + f (x)f (x + 1) = 0.

40. (1997 Greek Math Olympiad) Let f : (0, ) R be a function such

that
(a) f is strictly increasing,
(b) f (x) > x1 for all x > 0 and
(c) f (x)f (f (x) + x1 ) = 1 for all x > 0.
Find f (1).

41. (1979 Eotvos-Kurschak Math Competition) The function f is defined

for all real numbers and satisfies f (x) x and f (x + y) f (x) + f (y)
for all real x, y. Prove that f (x) = x for every real number x.

42. (Proposed by Ireland for 1989 IMO) Suppose f : R R satisfies

f (1) = 1, f (a + b) = f (a) + f (b) for all a, b R and f (x)f ( 1x ) = 1 for
x 6= 0. Show that f (x) = x for all x.

43. (1992 Polish Math Olympiad) Let Q+ be the positive rational numbers.
Determine all functions f : Q+ Q+ such that f (x + 1) = f (x) + 1
and f (x3 ) = f (x)3 for every x Q+ .

44. (1996 IMO shortlisted problem) Let R denote the real numbers and
f : R [1, 1] satisfy
     
13 1 1
f x+ + f (x) = f x + +f x+
42 6 7

for every x R. Show that f is a periodic function, i.e. there is a

nonzero real number T such that f (x + T ) = f (x) for every x R.

45. Let N denote the positive integers. Suppose s : N N is an increasing

function such that s(s(n)) = 3n for all n N. Find all possible values
of s(1997).

7
46. Let N be the positive integers. Is there a function f : N N such that
f (1996) (n) = 2n for all n N, where f (1) (x) = f (x) and f (k+1) (x) =
f (f (k) (x))?

47. (American Mathematical Monthly, Problem E984) Let R denote the

real numbers. Find all functions f : R R such that f (f (x)) = x2 2
or show no such function can exist.

48. Let R be the real numbers. Find all functions f : R R such that for
all real numbers x and y,


f xf (y) + x = xy + f (x).

49. (1999 IMO) Determine all functions f : R R such that

f (x f (y)) = f (f (y)) + xf (y) + f (x) 1

for all x, y in R.

50. (1995 Byelorussian Math Olympiad) Let R be the real numbers. Find
all functions f : R R such that

f (f (x + y)) = f (x + y) + f (x)f (y) xy

for all x, y R.

51. (1993 Czechoslovak Math Olympiad) Let Z be the integers. Find all
functions f : Z Z such that

f (1) = f (1) and f (x) + f (y) = f (x + 2xy) + f (y 2xy)

for all integers x, y.

52. (1995 South Korean Math Olympiad) Let A be the set of non-negative
integers. Find all functions f : A A satisfying the following two
conditions:
(a) For any m, n A, 2f (m2 + n2 ) = (f (m))2 + (f (n))2 .

8
 (b) For any m, n A with m n, f (m2 ) f (n2 ).

53. (American Mathematical Monthly, Problem E2176) Let Q denote the

rational numbers. Find all functions f : Q Q such that
 
x+y f (x) + f (y)
f (2) = 2 and f = for x 6= y.
xy f (x) f (y)

54. (Mathematics Magazine, Problem 1552) Find all functions f : R R

such that

f (x + yf (x)) = f (x) + xf (y) for all x, y in R.

Maximum/Minimum

55. (1985 Austrian Math Olympiad) For positive integers n, define

f (n) = 1n + 2n1 + 3n2 + + (n 2)3 + (n 1)2 + n.

What is the minimum of f (n + 1)/f (n)?

56. (1996 Putnam Exam) Given that {x1 , x2 , . . . , xn } = {1, 2, . . . , n}, find
the largest possible value of x1 x2 + x2 x3 + + xn1 xn + xn x1 in terms
of n (with n 2).

9
 Geometry Problems

57. (1995 British Math Olympiad) Triangle ABC has a right angle at C.
The internal bisectors of angles BAC and ABC meet BC and CA
at P and Q respectively. The points M and N are the feet of the
perpendiculars from P and Q to AB. Find angle M CN.

58. (1988 Leningrad Math Olympiad) Squares ABDE and BCF G are
drawn outside of triangle ABC. Prove that triangle ABC is isosceles if
DG is parallel to AC.

59. AB is a chord of a circle, which is not a diameter. Chords A1 B1 and

A2 B2 intersect at the midpoint P of AB. Let the tangents to the circle
at A1 and B1 intersect at C1 . Similarly, let the tangents to the circle
at A2 and B2 intersect at C2 . Prove that C1 C2 is parallel to AB.

60. (1991 Hunan Province Math Competition) Two circles with centers O1
and O2 intersect at points A and B. A line through A intersects the
circles with centers O1 and O2 at points Y, Z, respectively. Let the
tangents at Y and Z intersect at X and lines Y O1 and ZO2 intersect
at P. Let the circumcircle of 4O1 O2 B have center at O and intersect
line XB at B and Q. Prove that P Q is a diameter of the circumcircle
of 4O1 O2 B.

61. (1981 Beijing City Math Competition) In a disk with center O, there
are four points such that the distance between every pair of them is
greater than the radius of the disk. Prove that there is a pair of per-
pendicular diameters such that exactly one of the four points lies inside
each of the four quarter disks formed by the diameters.

62. The lengths of the sides of a quadrilateral are positive integers. The
length of each side divides the sum of the other three lengths. Prove
that two of the sides have the same length.

63. (1988 Sichuan Province Math Competition) Suppose the lengths of the
three sides of 4ABC are integers and the inradius of the triangle is 1.
Prove that the triangle is a right triangle.

10
 Geometric Equations

64. (1985 IMO) A circle has center on the side AB of the cyclic quadri-
lateral ABCD. The other three sides are tangent to the circle. Prove
that AD + BC = AB.

65. (1995 Russian Math Olympiad) Circles S1 and S2 with centers O1 , O2

respectively intersect each other at points A and B. Ray O1 B intersects
S2 at point F and ray O2 B intersects S1 at point E. The line parallel
to EF and passing through B intersects S1 and S2 at points M and
N, respectively. Prove that (B is the incenter of 4EAF and) M N =
AE + AF.

66. Point C lies on the minor arc AB of the circle centered at O. Suppose
the tangent line at C cuts the perpendiculars to chord AB through A
at E and through B at F. Let D be the intersection of chord AB and
radius OC. Prove that CE CF = AD BD and CD 2 = AE BF.

67. Quadrilaterals ABCP and A0 B 0 C 0 P 0 are inscribed in two concentric

circles. If triangles ABC and A0 B 0 C 0 are equilateral, prove that

P 0 A2 + P 0 B 2 + P 0 C 2 = P A02 + P B 02 + P C 02 .

68. Let the inscribed circle of triangle ABC touchs side BC at D, side CA
at E and side AB at F. Let G be the foot of perpendicular from D to
FG BF
EF. Show that = .
EG CE

69. (1998 IMO shortlisted problem) Let ABCDEF be a convex hexagon

such that

AB CD EF
6 B + 6 D + 6 F = 360 and = 1.
BC DE F A

Prove that
BC AE F D
= 1.
CA EF DB
11
 Similar Triangles

70. (1984 British Math Olympiad) P, Q, and R are arbitrary points on the
sides BC, CA, and AB respectively of triangle ABC. Prove that the
three circumcentres of triangles AQR, BRP, and CP Q form a triangle
similar to triangle ABC.

71. Hexagon ABCDEF is inscribed in a circle so that AB = CD = EF.

Let P, Q, R be the points of intersection of AC and BD, CE and DF,
EA and F B respectively. Prove that triangles P QR and BDF are
similar.

72. (1998 IMO shortlisted problem) Let ABCD be a cyclic quadrilateral.

Let E and F be variable points on the sides AB and CD, respectively,
such that AE : EB = CF : F D. Let P be the point on the segment
EF such that P E : P F = AB : CD. Prove that the ratio between the
areas of triangles AP D and BP C does not depend on the choice of E
and F.

Tangent Lines

73. Two circles intersect at points A and B. An arbitrary line through B

intersects the first circle again at C and the second circle again at D.
The tangents to the first circle at C and to the second circle at D
intersect at M. The parallel to CM which passes through the point
of intersection of AM and CD intersects AC at K. Prove that BK is
tangent to the second circle.

74. (1999 IMO) Two circles 1 and 2 are contained inside the circle ,
and are tangent to at the distinct points M and N, respectively.
1 passes through the center of 2 . The line passing through the two
points of intersection of 1 and 2 meets at A and B, respectively.
The lines M A and M B meets 1 at C and D, respectively. Prove that
CD is tangent to 2 .

75. (Proposed by India for 1992 IMO) Circles G1 and G2 touch each other
externally at a point W and are inscribed in a circle G. A, B, C are

12
 points on G such that A, G1 and G2 are on the same side of chord BC,
which is also tangent to G1 and G2 . Suppose AW is also tangent to
G1 and G2 . Prove that W is the incenter of triangle ABC.

Locus

76. Perpendiculars from a point P on the circumcircle of 4ABC are drawn

to lines AB, BC with feet at D, E, respectively. Find the locus of the
circumcenter of 4P DE as P moves around the circle.

77. Suppose A is a point inside a given circle and is different from the
center. Consider all chords (excluding the diameter) passing through
A. What is the locus of the intersection of the tangent lines at the
endpoints of these chords?

78. Given 4ABC. Let line EF bisects 6 BAC and AE AF = AB AC.

Find the locus of the intersection P of lines BE and CF.

79. (1996 Putnam Exam) Let C1 and C2 be circles whose centers are 10
units apart, and whose radii are 1 and 3. Find the locus of all points
M for which there exists points X on C1 and Y on C2 such that M is
the midpoint of the line segment XY.

Collinear or Concyclic Points

80. (1982 IMO) Diagonals AC and CE of the regular hexagon ABCDEF

are divided by the inner points M and N, respectively, so that

AM CN
= = r.
AC CE
Determine r if B, M and N are collinear.

81. (1965 Putnam Exam) If A, B, C, D are four distinct points such that
every circle through A and B intersects or coincides with every circle
through C and D, prove that the four points are either collinear or
concyclic.

13
82. (1957 Putnam Exam) Given an infinite number of points in a plane,
prove that if all the distances between every pair are integers, then the
points are collinear.

83. (1995 IMO shortlisted problem) The incircle of triangle ABC touches
BC, CA and AB at D, E and F respectively. X is a point inside
triangle ABC such that the incircle of triangle XBC touches BC at
D also, and touches CX and XB at Y and Z respectively. Prove that
EF ZY is a cyclic quadrilateral.

84. (1998 IMO) In the convex quadrilateral ABCD, the diagonals AC and
BD are perpendicular and the opposite sides AB and DC are not
parallel. Suppose the point P, where the perpendicular bisectors of
AB and DC meet, is inside ABCD. Prove that ABCD is a cyclic
quadrilateral if and only if the triangles ABP and CDP have equal
areas.

85. (1970 Putnam Exam) Show that if a convex quadrilateral with side-
lengths a, b, c, d and area abcd has an inscribed circle, then it is a
cyclic quadrilateral.

Concurrent Lines

86. In 4ABC, suppose AB > AC. Let P and Q be the feet of the per-
pendiculars from B and C to the angle bisector of 6 BAC, respectively.
Let D be on line BC such that DA AP. Prove that lines BQ, P C
and AD are concurrent.

87. (1990 Chinese National Math Competition) Diagonals AC and BD

of a cyclic quadrilateral ABCD meets at P. Let the circumcenters of
ABCD, ABP, BCP, CDP and DAP be O, O1, O2 , O3 and O4 , respec-
tively. Prove that OP, O1 O3 , O2 O4 are concurrent.

88. (1995 IMO) Let A, B, C and D be four distinct points on a line, in that
order. The circles with diameters AC and BD intersect at the points
X and Y. The line XY meets BC at the point Z. Let P be a point on
the line XY different from Z. The line CP intersects the circle with

14
 diameter AC at the points C and M, and the line BP intersects the
circle with diameter BD at the points B and N. Prove that the lines
AM, DN and XY are concurrent.

89. AD, BE, CF are the altitudes of 4ABC. If P, Q, R are the midpoints
of DE, EF, F D, respectively, then show that the perpendicular from
P, Q, R to AB, BC, CA, respectively, are concurrent.

90. (1988 Chinese Math Olympiad Training Test) ABCDEF is a hexagon

inscribed in a circle. Show that the diagonals AD, BE, CF are concur-
rent if and only if AB CD EF = BC DE F A.

91. A circle intersects a triangle ABC at six points A1 , A2 , B1 , B2 , C1 , C2 ,

where the order of appearance along the triangle is A, C1 , C2 , B, A1 , A2 ,
C, B1 , B2 , A. Suppose B1 C1 , B2 C2 meets at X, C1 A1 , C2 A2 meets at
Y and A1 B1 , A2 B2 meets at Z. Show that AX, BY, CZ are concurrent.

92. (1995 IMO shortlisted problem) A circle passing through vertices B

and C of triangle ABC intersects sides AB and AC at C 0 and B 0 ,
respectively. Prove that BB 0 , CC 0 and HH 0 are concurrent, where H
and H 0 are the orthocenters of triangles ABC and AB 0 C 0 , respectively.

Perpendicular Lines

93. (1998 APMO) Let ABC be a triangle and D the foot of the altitude
from A. Let E and F be on a line passing through D such that AE
is perpendicular to BE, AF is perpendicular to CF, and E and F are
different from D. Let M and N be the midpoints of the line segments
BC and EF, respectively. Prove that AN is perpendicular to N M.

94. (2000 APMO) Let ABC be a triangle. Let M and N be the points
in which the median and the angle bisector, respectively, at A meet
the side BC. Let Q and P be the points in which the perpendicular at
N to N A meets M A and BA, respectively, and O the point in which
the perpendicular at P to BA meets AN produced. Prove that QO is
perpendicular to BC.

15
 95. Let BB 0 and CC 0 be altitudes of triangle ABC. Assume that AB 6=
AC. Let M be the midpoint of BC, H the orthocenter of ABC and D
the intersection of B 0 C 0 and BC. Prove that DH AM.

96. (1996 Chinese Team Selection Test) The semicircle with side BC of
4ABC as diameter intersects sides AB, AC at points D, E, respec-
tively. Let F, G be the feet of the perpendiculars from D, E to side
BC respectively. Let M be the intersection of DG and EF. Prove that
AM BC.

97. (1985 IMO) A circle with center O passes through the vertices A and
C of triangle ABC and intersects the segments AB and AC again at
distinct points K and N, respectively. The circumcircles of triangles
ABC and KBN intersect at exactly two distinct points B and M.
Prove that OM M B.

98. (1997 Chinese Senoir High Math Competition) A circle with center O
is internally tangent to two circles inside it at points S and T. Suppose
the two circles inside intersect at M and N with N closer to ST. Show
that OM M N if and only if S, N, T are collinear.

99. AD, BE, CF are the altitudes of 4ABC. Lines EF, F D, DE meet lines
BC, CA, AB in points L, M, N, respectively. Show that L, M, N are
collinear and the line through them is perpendicular to the line joining
the orthocenter H and circumcenter O of 4ABC.

Geometric Inequalities, Maximum/Minimum

100. (1973 IMO) Let P1 , P2 , . . . , P2n+1 be distinct points on some half of

the unit circle centered at the origin O. Show that

|OP1 + OP2 + + OP2n+1 | 1.

101. Let the angle bisectors of 6 A, 6 B, 6 C of triangle ABC intersect its

circumcircle at P, Q, R, respectively. Prove that

AP + BQ + CR > BC + CA + AB.

16
102. (1997 APMO) Let ABC be a triangle inscribed in a circle and let la =
ma /Ma , lb = mb /Mb , lc = mc /Mc , where ma , mb , mc are the lengths
of the angle bisectors (internal to the triangle) and Ma , Mb , Mc are
the lengths of the angle bisectors extended until they meet the circle.
Prove that
la lb lc
+ + 3,
sin2 A sin2 B sin2 C
and that equality holds iff ABC is equilateral.

103. (Mathematics Magazine, Problem 1506) Let I and O be the incen-

ter and circumcenter of 4ABC, respectively. Assume 4ABC is not
equilateral (so I 6= O). Prove that

6 AIO 90 if and only if 2BC AB + CA.

104. Squares ABDE and ACF G are drawn outside 4ABC. Let P, Q be
points on EG such that BP and CQ are perpendicular to BC. Prove
that BP + CQ BC + EG. When does equality hold?

105. Point P is inside 4ABC. Determine points D on side AB and E on

side AC such that BD = CE and P D + P E is minimum.

Solid or Space Geometry

106. (Proposed by Italy for 1967 IMO) Which regular polygons can be ob-
tained (and how) by cutting a cube with a plane?

107. (1995 Israeli Math Olympiad) Four points are given in space, in general
position (i.e., they are not coplanar and any three are not collinear).
A plane is called an equalizing plane if all four points have the same
distance from . Find the number of equalizing planes.

17
 Number Theory Problems

Digits

108. (1956 Putnam Exam) Prove that every positive integer has a multiple
whose decimal representation involves all ten digits.

109. Does there exist a positive integer a such that the sum of the digits
(in base 10) of a is 1999 and the sum of the digits (in base 10) of a2 is
19992 ?

110. (Proposed by USSR for 1991 IMO) Let an be the last nonzero digit
in the decimal representation of the number n!. Does the sequence
a1 , a2 , . . . , an , . . . become periodic after a finite number of terms?

Modulo Arithmetic

111. (1956 Putnam Exam) Prove that the number of odd binomial coeffi-
cients in any row of the Pascal triangle is a power of 2.

112. Let a1 , a2 , a3 , . . . , a11 and b1 , b2 , b3 , . . . , b11 be two permutations of the

natural numbers 1, 2, 3, . . . , 11. Show that if each of the numbers a1 b1 ,
a2 b2 , a3 b3 , . . . , a11 b11 is divided by 11, then at least two of them will
have the same remainder.

113. (1995 Czech-Slovak Match) Let a1 , a2 , . . . be a sequence satisfying a1 =

2, a2 = 5 and
an+2 = (2 n2 )an+1 + (2 + n2 )an
for all n 1. Do there exist indices p, q and r such that ap aq = ar ?

Prime Factorization

114. (American Mathematical Monthly, Problem E2684) Let An be the set

of positive integers which are less than n and are relatively prime to n.
For which n > 1, do the integers in An form an arithmetic progression?

18
115. (1971 IMO) Prove that the set of integers of the form 2k 3 (k =
2, 3, . . .) contains an infinite subset in which every two members are
relatively prime.

116. (1988 Chinese Math Olympiad Training Test) Determine the smallest
value of the natural number n > 3 with the property that whenever
the set Sn = {3, 4, . . . , n} is partitioned into the union of two sub-
sets, at least one of the subsets contains three numbers a, b and c (not
necessarily distinct) such that ab = c.

Base n Representations

117. (1983 IMO) Can you choose 1983 pairwise distinct nonnegative integers
less than 105 such that no three are in arithmetic progression?

118. (American Mathematical Monthly, Problem 2486) Let p be an odd

prime number and r be a positive integer not divisible by p. For any
positive integer k, show that there exists a positive integer m such that
the rightmost k digits of mr , when expressed in the base p, are all 1s.

119. (Proposed by Romania for 1985 IMO) Show that the sequence {an }
defined by an = [n 2] for n = 1, 2, 3, . . . (where the brackets denote
the greatest integer function) contains an infinite number of integral
powers of 2.

Representations

120. Find all (even) natural numbers n which can be written as a sum of
two odd composite numbers.

121. Find all positive integers which cannot be written as the sum of two
or more consecutive positive integers.

122. (Proposed by Australia for 1990 IMO) Observe that 9 = 4+5 = 2+3+4.
Is there an integer N which can be written as a sum of 1990 consecutive
positive integers and which can be written as a sum of (more than one)
consecutive integers in exactly 1990 ways?

19
123. Show that if p > 3 is prime, then pn cannot be the sum of two positive
cubes for any n 1. What about p = 2 or 3?

124. (Due to Paul Erdos and M. Suranyi) Prove that every integer k can be
represented in infinitely many ways in the form k = 12 22 m2
for some positive integer m and some choice of signs + or .

125. (1996 IMO shortlisted problem) A finite sequence of integers a0 , a1 , . . . ,

an is called quadratic if for each i {1, 2, . . . , n}, |ai ai1 | = i2 .
(a) Prove that for any two integers b and c, there exists a natural
number n and a quadratic sequence with a0 = b and an = c.
(b) Find the least natural number n for which there exists a quadratic
sequence with a0 = 0 and an = 1996.

126. Prove that every integer greater than 17 can be represented as a sum of
three integers > 1 which are pairwise relatively prime, and show that
17 does not have this property.

Chinese Remainder Theorem

127. (1988 Chinese Team Selection Test) Define xn = 3xn1 + 2 for all
positive integers n. Prove that an integer value can be chosen for x0 so
that x100 is divisible by 1988.

128. (Proposed by North Korea for 1992 IMO) Does there exist a set M
with the following properties:
(a) The set M consists of 1992 natural numbers.
(b) Every element in M and the sum of any number of elements in M
have the form mk , where m, k are positive integers and k 2?

Divisibility

129. Find all positive integers a, b such that b > 2 and 2a + 1 is divisible by
2b 1.

20
130. Show that there are infinitely many composite n such that 3n1 2n1
is divisible by n.

131. Prove that there are infinitely many positive integers n such that 2n +1
is divisible by n. Find all such ns that are prime numbers.

132. (1998 Romanian Math Olympiad) Find all positive integers (x, n) such
that xn + 2n + 1 is a divisor of xn+1 + 2n+1 + 1.

133. (1995 Bulgarian Math Competition) Find all pairs of positive integers
x2 + y 2
(x, y) for which is an integer and divides 1995.
xy

134. (1995 Russian Math Olympiad) Is there a sequence of natural numbers

in which every natural number occurs just once and moreover, for any
k = 1, 2, 3, . . . the sum of the first k terms is divisible by k?

135. (1998 Putnam Exam) Let A1 = 0 and A2 = 1. For n > 2, the number
An is defined by concatenating the decimal expansions of An1 and
An2 from left to right. For example, A3 = A2 A1 = 10, A4 = A3 A2 =
101, A5 = A4 A3 = 10110, and so forth. Determine all n such that An
is divisible by 11.

136. (1995 Bulgarian Math Competition) If k > 1, show that k does not
divide 2k1 + 1. Use this to find all prime numbers p and q such that
2p + 2q is divisible by pq.

137. Show that for any positive integer n, there is a number whose decimal
representation contains n digits, each of which is 1 or 2, and which is
divisible by 2n .

138. For a positive integer n, let f (n) be the largest integer k such that 2k
divides n and g(n) be the sum of the digits in the binary representation
of n. Prove that for any positive integer n,
(a) f (n!) = n
g(n);

2n (2n)!
(b) 4 divides = if and only if n is not a power of 2.
n n!n!

21
139. (Proposed by Australia for 1992 IMO) Prove that for any positive in-
teger m, there exist an infinite number of pairs of integers (x, y) such
that
(a) x and y are relatively prime;
(b) y divides x2 + m;
(c) x divides y 2 + m.

140. Find all integers n > 1 such that 1n + 2n + + (n 1)n is divisible

by n.

141. (1972 Putnam Exam) Show that if n is an integer greater than 1, then
n does not divide 2n 1.

142. (Proposed by Romania for 1985 IMO) For k 2, let n1 , n2 , . . . , nk be

positive integers such that
   
n2  (2n1 1), n3  (2n2 1), . . . , nk  (2nk1 1), n1  (2nk 1).

Prove that n1 = n2 = = nk = 1.

143. (1998 APMO) Determine the largest of all integer n with theproperty
that n is divisible by all positive integers that are less than 3 n.

144. (1997 Ukrainian Math Olympiad) Find the smallest integer n such that
among any n integers (with possible repetitions), there exist 18 integers
whose sum is divisible by 18.

Perfect Squares, Perfect Cubes

1 1 1
145. Let a, b, c be positive integers such that + = . If the greatest
a b c
common divisor of a, b, c is 1, then prove that a + b must be a perfect
square.

146. (1969 Eotvos-Kurschak

Math Competition) Let n be a positive integer.
Show that if 2 + 2 28n2 + 1 is an integer, then it is a square.

22
147. (1998 Putnam Exam) Prove that, for any integers a, b, c, there exists a
positive integer n such that n3 + an2 + bn + c is not an integer.

148. (1995 IMO shortlisted problem) Let k be a positive integer. Prove that
there are infinitely many perfect squares of the form n2k 7, where n
is a positive integer.
a b c
149. Let a, b, c be integers such that + + = 3. Prove that abc is the
b c a
cube of an integer.

Diophantine Equations

150. Find all sets of positive integers x, y and z such that x y z and
xy + y z = z x .

151. (Due to W. Sierpinski in 1955) Find all positive integral solutions of

3x + 4y = 5z .

152. (Due to Euler, also 1985 Moscow Math Olympiad) If n 3, then prove
that 2n can be represented in the form 2n = 7x2 + y 2 with x, y odd
positive integers.

153. (1995 IMO shortlisted problem) Find all positive integers x and y such
that x + y 2 + z 3 = xyz, where z is the greatest common divisor of x
and y.

154. Find all positive integral solutions to the equation xy + yz + zx =

xyz + 2.

155. Show that if the equation x2 + y 2 + 1 = xyz has positive integral

solutions x, y, z, then z = 3.

156. (1995 Czech-Slovak Match) Find all pairs of nonnegative integers x and
y which solve the equation px y p = 1, where p is a given odd prime.

157. Find all integer solutions of the system of equations

x+y+z = 3 and x3 + y 3 + z 3 = 3.

23
 Combinatorics Problems

Counting Methods

158. (1996 Italian Mathematical Olympiad) Given an alphabet with three

letters a, b, c, find the number of words of n letters which contain an
even number of as.

159. Find the number of n-words from the alphabet A = {0, 1, 2}, if any
two neighbors can differ by at most 1.

160. (1995 Romanian Math Olympiad) Let A1 , A2 , . . . , An be points on a

circle. Find the number of possible colorings of these points with p
colors, p 2, such that any two neighboring points have distinct colors.

Pigeonhole Principle

161. (1987 Austrian-Polish Math Competition) Does the set {1, 2, . . . , 3000}
contain a subset A consisting of 2000 numbers such that x A implies
2x 6 A?

162. (1989 Polish Math Olympiad) Suppose a triangle can be placed inside
a square of unit area in such a way that the center of the square is not
inside the triangle. Show that one side of the triangle has length less
than 1.

163. The cells of a 7 7 square are colored with two colors. Prove that
there exist at least 21 rectangles with vertices of the same color and
with sides parallel to the sides of the square.

164. For n > 1, let 2n chess pieces be placed at the centers of 2n squares of
an n n chessboard. Show that there are four pieces among them that
formed the vertices of a parallelogram. If 2n is replaced by 2n 1, is
the statement still true in general?

165. The set {1, 2, . . . , 49} is partitioned into three subsets. Show that at
least one of the subsets contains three different numbers a, b, c such
that a + b = c.

24
 Inclusion-Exclusion Principle

166. Let m n > 0. Find the number of surjective functions from Bm =

{1, 2, . . . , m} to Bn = {1, 2, . . . , n}.

167. Let A be a set with 8 elements. Find the maximal number of 3-element
subsets of A, such that the intersection of any two of them is not a 2-
element set.

168. (a) (1999 China Hong Kong Math Olympiad) Students have taken a
test paper in each of n (n 3) subjects. It is known that for any
subject exactly three students get the best score in the subject, and
for any two subjects excatly one student gets the best score in every
one of these two subjects. Determine the smallest n so that the above
conditions imply that exactly one student gets the best score in every
one of the n subjects.
(b) (1978 Austrian-Polish Math Competition) There are 1978 clubs.
Each has 40 members. If every two clubs have exactly one common
member, then prove that all 1978 clubs have a common member.

Combinatorial Designs

169. (1995 Byelorussian Math Olympiad) In the begining, 65 beetles are

placed at different squares of a 9 9 square board. In each move, every
beetle creeps to a horizontal or vertical adjacent square. If no beetle
makes either two horizontal moves or two vertical moves in succession,
show that after some moves, there will be at least two beetles in the
same square.

170. (1995 Greek Math Olympiad) Lines l1 , l2 , . . . , lk are on a plane such

that no two are parallel and no three are concurrent. Show that we
can label the C2k intersection points of these lines by the numbers
1, 2, . . . , k 1 so that in each of the lines l1 , l2 , . . . , lk the numbers
1, 2, . . . , k 1 appear exactly once if and only if k is even.

171. (1996 Tournaments of the Towns) In a lottery game, a person must

select six distinct numbers from 1, 2, 3, . . . , 36 to put on a ticket. The

25
 lottery commitee will then draw six distinct numbers randomly from
1, 2, 3, . . . , 36. Any ticket with numbers not containing any of these six
numbers is a winning ticket. Show that there is a scheme of buying
9 tickets guaranteeing at least a winning ticket, but 8 tickets is not
enough to guarantee a winning ticket in general.

172. (1995 Byelorussian Math Olympiad) By dividing each side of an equi-

lateral triangle into 6 equal parts, the triangle can be divided into 36
smaller equilateral triangles. A beetle is placed on each vertex of these
triangles at the same time. Then the beetles move along different edges
with the same speed. When they get to a vertex, they must make a
60 or 120 turn. Prove that at some moment two beetles must meet
at some vertex. Is the statement true if 6 is replaced by 5?

Covering, Convex Hull

173. (1991 Australian Math Olympiad) There are n points given on a plane
such that the area of the triangle formed by every 3 of them is at most
1. Show that the n points lie on or inside some triangle of area at most
4.

174. (1969 Putnam Exam) Show that any continuous curve of unit length
can be covered by a closed rectangles of area 1/4.

175. (1998 Putnam Exam) Let F be a finite collection of open discs in the
plane whose union covers a set E. Show that there is a pairwise disjoint
subcollection D1 , . . . , Dn in F such that the union of 3D1 , . . . , 3Dn
covers E, where 3D is the disc with the same center as D but having
three times the radius.

176. (1995 IMO) Determine all integers n > 3 for which there exist n points
A1 , A2 , . . . , An in the plane, and real numbers r1 , r2 , . . . , rn satisfying
the following two conditions:
(a) no three of the points A1 , A2 , . . . , An lie on a line;
(b) for each triple i, j, k (1 i < j < k n) the triangle Ai Aj Ak has
area equal to ri + rj + rk .

26
177. (1999 IMO) Determine all finite sets S of at least three points in the
plane which satisfy the following condition: for any two distinct points
A and B in S, the perpendicular bisector of the line segment AB is an
axis of symmetry of S.

27
 Miscellaneous Problems

178. (1995 Russian Math Olympiad) There are n seats at a merry-go-around.

A boy takes n rides. Between each ride, he moves clockwise a certain
number (less than n) of places to a new horse. Each time he moves a
different number of places. Find all n for which the boy ends up riding
each horse.

179. (1995 Israeli Math Olympiad) Two players play a game on an infinite
board that consists of 1 1 squares. Player I chooses a square and
marks it with an O. Then, player II chooses another square and marks
it with X. They play until one of the players marks a row or a column
of 5 consecutive squares, and this player wins the game. If no player
can achieve this, the game is a tie. Show that player II can prevent
player I from winning.

180. (1995 USAMO) A calculator is broken so that the only keys that still
work are the sin, cos, tan, sin1 , cos1 , and tan1 buttons. The dis-
play initially shows 0. Given any positive rational number q, show that
pressing some finite sequence of buttons will yield q. Assume that the
calculator does real number calculations with infinite precision. All
functions are in terms of radians.

181. (1977 Eotvos-Kurschak Math Competition) Each of three schools is

attended by exactly n students. Each student has exactly n + 1 ac-
quaintances in the other two schools. Prove that one can pick three
students, one from each school, who know one another. It is assumed
that acquaintance is mutual.

182. Is there a way to pack 250 1 1 4 bricks into a 10 10 10 box?

183. Is it possible to write a positive integer into each square of the first
quadrant such that each column and each row contains every positive
integer exactly once?

184. There are n identical cars on a circular track. Among all of them, they
have just enough gas for one car to complete a lap. Show that there is

28
 a car which can complete a lap by collecting gas from the other cars
on its way around the track in the clockwise direction.

185. (1996 Russian Math Olympiad) At the vertices of a cube are written
eight pairwise distinct natural numbers, and on each of its edges is
written the greatest common divisor of the numbers at the endpoints
of the edge. Can the sum of the numbers written at the vertices be the
same as the sum of the numbers written at the edges?

186. Can the positive integers be partitioned into infinitely many subsets
such that each subset is obtained from any other subset by adding the
same integer to each element of the other subset?

187. (1995 Russian Math Olympiad) Is it possible to fill in the cells of a

9 9 table with positive integers ranging from 1 to 81 in such a way
that the sum of the elements of every 3 3 square is the same?

188. (1991 German Mathematical Olympiad) Show that for every positive
integer n 2, there exists a permutation p1 , p2 , . . . , pn of 1, 2, . . . , n
such that pk+1 divides p1 + p2 + + pk for k = 1, 2, . . . , n 1.

189. Each lattice point of the plane is labeled by a positive integer. Each
of these numbers is the arithmetic mean of its four neighbors (above,
below, left, right). Show that all the numbers are equal.

190. (1984 Tournament of the Towns) In a party, n boys and n girls are
paired. It is observed that in each pair, the difference in height is less
than 10 cm. Show that the difference in height of the k-th tallest boy
and the k-th tallest girl is also less than 10 cm for k = 1, 2, . . . , n.

191. (1991 Leningrad Math Olympiad) One may perform the following two
operations on a positive integer:
(a) multiply it by any positive integer and
(b) delete zeros in its decimal representation.
Prove that for every positive integer X, one can perform a sequence of
these operations that will transform X to a one-digit number.

29
192. (1996 IMO shortlisted problem) Four integers are marked on a circle.
On each step we simultaneously replace each number by the difference
between this number and next number on the circle in a given direction
(that is, the numbers a, b, c, d are replaced by a b, b c, c d, d a).
Is it possible after 1996 such steps to have numbers a, b, c, d such that
the numbers |bc ad|, |ac bd|, |ab cd| are primes?

193. (1989 Nanchang City Math Competition) There are 1989 coins on a
table. Some are placed with the head sides up and some the tail sides
up. A group of 1989 persons will perform the following operations:
the first person is allowed turn over any one coin, the second person is
allowed turn over any two coins, . . . , the k-th person is allowed turn
over any k coins, . . . , the 1989th person is allowed to turn over every
coin. Prove that
(1) no matter which sides of the coins are up initially, the 1989 persons
can come up with a procedure turning all coins the same sides up
at the end of the operations,
(2) in the above procedure, whether the head or the tail sides turned
up at the end will depend on the initial placement of the coins.

194. (Proposed by India for 1992 IMO) Show that there exists a convex
polygon of 1992 sides satisfying the following conditions:
(a) its sides are 1, 2, 3, . . . , 1992 in some order;
(b) the polygon is circumscribable about a circle.

195. There are 13 white, 15 black, 17 red chips on a table. In one step, you
may choose 2 chips of different colors and replace each one by a chip of
the third color. Can all chips become the same color after some steps?

196. The following operations are permitted with the quadratic polynomial
ax2 + bx + c:
(a) switch a and c,
(b) replace x by x + t, where t is a real number.
By repeating these operations, can you transform x2 x 2 into x2
x 1?

30
197. Five numbers 1, 2, 3, 4, 5 are written on a blackboard. A student may
erase any two of the numbers a and b on the board and write the
numbers a + b and ab replacing them. If this operation is performed re-
peatedly, can the numbers 21, 27, 64, 180, 540 ever appear on the board?

198. Nine 1 1 cells of a 10 10 square are infected. In one unit time, the
cells with at least 2 infected neighbors (having a common side) become
infected. Can the infection spread to the whole square? What if nine
is replaced by ten?

199. (1997 Colombian Math Olympiad) We play the following game with
an equilateral triangle of n(n + 1)/2 dollar coins (with n coins on each
side). Initially, all of the coins are turned heads up. On each turn, we
may turn over three coins which are mutually adjacent; the goal is to
make all of the coins turned tails up. For which values of n can this be
done?

200. (1990 Chinese Team Selection Test) Every integer is colored with one
of 100 colors and all 100 colors are used. For intervals [a, b], [c, d] having
integers endpoints and same lengths, if a, c have the same color and
b, d have the same color, then the intervals are colored the same way,
which means a + x and c + x have the same color for x = 0, 1, . . . , b a.
Prove that 1990 and 1990 have different colors.

31
Solutions
 Solutions to Algebra Problems

Polynomials

1. (Crux Mathematicorum, Problem 7) Find (without calculus) a fifth

degree polynomial p(x) such that p(x) + 1 is divisible by (x 1)3 and
p(x) 1 is divisible by (x + 1)3 .

Solution. (Due to Law Ka Ho, Ng Ka Wing, Tam Siu Lung) Note

(x 1)3 divides p(x) + 1 and p(x) 1, so (x 1)3 divides their sum
p(x) + p(x). Also (x + 1)3 divides p(x) 1 and p(x) + 1, so (x + 1)3
divides p(x)+p(x). Then (x1)3 (x+1)3 divides p(x)+p(x), which is
of degree at most 5. So p(x)+p(x) = 0 for all x. Then the even degree
term coefficients of p(x) are zero. Now p(x)+1 = (x1)3 (Ax2 +Bx1).
Comparing the degree 2 and 4 coefficients, we get B 3A = 0 and
3 + 3B A = 0, which implies A = 3/8 and B = 9/8. This yields
p(x) = 3x5 /8 + 5x3 /4 15x/8.

2. A polynomial P (x) of the n-th degree satisfies P (k) = 2k for k =

0, 1, 2, . . . , n. Find the value of P (n + 1).
 
x x(x 1) (x r + 1)
Solution. For 0 r n, the polynomial =
r r!
is of degree r. Consider the degree n polynomial
     
x x x
Q(x) = + + + .
0 1 n

By the binomial theorem, Q(k) = (1 + 1)k = 2k for k = 0, 1, 2, . . . , n.

So P (x) = Q(x) for all x. Then
     
n+1 n+1 n+1
P (n +1) = Q(n +1) = + + + = 2n+1 1.
0 1 n

3. (1999 Putnam Exam) Let P (x) be a polynomial with real coefficients

such that P (x) 0 for every real x. Prove that

P (x) = f1 (x)2 + f2 (x)2 + + fn (x)2

35
 for some polynomials f1 (x), f2 (x), . . . , fn (x) with real coefficients.

Solution. (Due to Cheung Pok Man) Write P (x) = aR(x)C(x), where

a is the coefficient of the highest degree term, R(x) is the product of all
real root factors (xr) repeated according to multiplicities and C(x) is
the product of all conjugate pairs of nonreal root factors (xzk )(xzk ).
Then a 0. Since P (x) 0 for every real x and a factor (x r)2n+1
would change sign near a real root r of odd multiplicity, each real root
of P must have even multiplicity. So R(x) = f (x)2 for some polynomial
f (x) with real coefficients.
Next pick one factor from each conjugate pair of nonreal factors
and let the product of these factors (x zk ) be equal to U (x) + iV (x),
where U (x), V (x) are polynomials with real coefficients. We have

P (x) = af (x)2 (U (x) + iV (x))(U (x) iV (x))

= ( af (x)U (x))2 + ( af (x)V (x))2 .

4. (1995 Russian Math Olympiad) Is it possible to find three quadratic

polynomials f (x), g(x), h(x) such that the equation f (g(h(x))) = 0 has
the eight roots 1, 2, 3, 4, 5, 6, 7, 8?

Solution. Suppose there are such f, g, h. Then h(1), h(2), . . . , h(8) will
be the roots of the 4-th degree polynomial f (g(x)). Since h(a) =
h(b), a 6= b if and only if a, b are symmetric with respect to the axis
of the parabola, it follows that h(1) = h(8), h(2) = h(7), h(3) =
h(6), h(4) = h(5) and the parabola y = h(x) is symmetric with re-
spect to x = 9/2. Also, we have either h(1) < h(2) < h(3) < h(4) or
h(1) > h(2) > h(3) > h(4).
Now g(h(1)), g(h(2)), g(h(3)), g(h(4)) are the roots of the quadratic
polynomial f (x), so g(h(1)) = g(h(4)) and g(h(2)) = g(h(3)), which
implies h(1) + h(4) = h(2) + h(3). For h(x) = Ax2 + Bx + C, this would
force A = 0, a contradiction.

5. (1968 Putnam Exam) Determine all polynomials whose coefficients are

all 1 that have only real roots.

36
 Solution. If a polynomial a0 xn +a1 xn1 + +an is such a polynomial,
then so is its negative. Hence we may assume a0 = 1. Let r1 , . . . , rn be
the roots. Then r12 + + rn2 = a21 2a2 and r12 rn2 = a2n . If the roots
2/n
are all real, then by the AM-GM inequality, we get (a21 2a2 )/n an .
Since a1 , a2 = 1, we must have a2 = 1 and n 3. By simple
checking, we get the list

(x 1), (x + 1), (x2 + x 1), (x2 x 1),

(x3 + x2 x 1), (x3 x2 x + 1).

6. (1990 Putnam Exam) Is there an infinite sequence a0 , a1 , a2 , . . . of

nonzero real numbers such that for n = 1, 2, 3, . . . , the polynomial
Pn (x) = a0 + a1 x + a2 x2 + + an xn has exactly n distinct real roots?

Solution. Yes. Take a0 = 1, a1 = 1 and proceed by induction.

Suppose a0 , . . . , an have been chosen so that Pn (x) has n distinct real
roots and Pn (x) or as x depending upon whether n
is even or odd. Suppose the roots of Pn (x) is in the interval (T, T ).
Let an+1 = (1)n+1 /M , where M is chosen to be very large so that
T n+1 /M is very small. Then Pn+1 (x) = Pn (x) + (x)n+1 /M is very
close to Pn (x) on [T, T ] because |Pn+1 (x) Pn (x)| T n+1 /M for
every x on [T, T ]. So, Pn+1 (x) has a sign change very close to every
root of Pn (x) and has the same sign as Pn (x) at T. Since Pn (x) and
Pn+1 (x) take on different sign when x , there must be another
sign change beyond T . So Pn+1 (x) must have n + 1 real roots.

7. (1991 Austrian-Polish Math Competition) Let P (x) be a polynomial

with real coefficients such that P (x) 0 for 0 x 1. Show that
there are polynomials A(x), B(x), C(x) with real coefficients such that
(a) A(x) 0, B(x) 0, C(x) 0 for all real x and
(b) P (x) = A(x) + xB(x) + (1 x)C(x) for all real x.
(For example, if P (x) = x(1x), then P (x) = 0+x(1x)2 +(1x)x2 .)

Solution. (Below all polynomials have real coefficients.) We induct

on the degree of P (x). If P (x) is a constant polynomial c, then c 0

37
 and we can take A(x) = c, B(x) = C(x) = 0. Next suppose the degree
n case is true. For the case P (x) is of degree n + 1. If P (x) 0 for
all real x, then simply let A(x) = P (x), B(x) = C(x) = 0. Otherwise,
P (x) has a root x0 in (, 0] or [1, +).
Case x0 in (, 0]. Then P (x) = (x x0 )Q(x) and Q(x) is of degree n
with Q(x) 0 for all x n [0, 1]. So Q(x) = A0 (x)+xB0 (x)+(1x)C0 (x),
where A0 (x), B0 (x), C0 (x) 0 for all x in [0, 1]. Using x(1 x) =
x(1 x)2 + (1 x)x2 , we have

P (x) = (x x0 )(A0 (x) + xB0 (x) + (1 x)C0 (x))

= (x0 A0 (x)+ x2 B0 (x)) +x (A0 (x) x0 B0 (x) +(1 x)2 C0 (x))
| {z } | {z }
A(x) B(x)

+ (1 x) (x0 C0 (x) + x2 B0 (x)),

| {z }
C(x)

where the polynomials A(x), B(x), C(x) 0 for all x in [0, 1].
Case x0 in [1, +). Consider Q(x) = P (1 x). This reduces to the
previous case. We have Q(x) = A1 (x) + xB1 (x) + (1 x)C1 (x), where
the polynomials A1 (x), B1 (x), C1 (x) 0 for all x in [0, 1]. Then

P (x) = Q(1 x) = A1 (1 x) +x C1 (1 x) +(1 x) B1 (1 x),

| {z } | {z } | {z }
A(x) B(x) C(x)

where the polynomials A(x), B(x), C(x) 0 for all x in [0, 1].

8. (1993 IMO) Let f (x) = xn + 5xn1 + 3, where n > 1 is an integer.

Prove that f (x) cannot be expressed as a product of two polynomials,
each has integer coefficients and degree at least 1.

Solution. Suppose f (x) = b(x)c(x) for nonconstant polynomials b(x)

and c(x) with integer coefficients. Since f (0) = 3, we may assume
b(0) = 1 and b(x) = xr + 1. Since f (1) 6= 0, r > 1. Let
z1 , . . . , zr be the roots of b(x). Then |z1 zr | = |b(0)| = 1 and
r
Y
|b(5)| = |(5 z1 ) (5 zr )| = |zin1 (zi + 5)| = 3r 9.
i=1

38
 However, b(5) also divides f (5) = 3, a contradiction.

9. Prove that if the integer a is not divisible by 5, then f (x) = x5 x + a

cannot be factored as the product of two nonconstant polynomials with
integer coefficients.

Solution. Suppose f can be factored, then f (x) = (x b)g(x) or

f (x) = (x2 bx + c)g(x). In the fomer case, b5 b + a = f (b) = 0. Now
b5 b (mod 5) by Fermats little theorem or simply checking the cases
b 0, 1, 2, 3, 4 (mod 5). Then 5 divides b b5 = a, a contradiction. In
the latter case, divding f (x) = x5 x + a by x2 bx + c, we get the
remainder (b4 + 3b2 c + c2 1)x + (b3 c + 2bc2 + a). Since x2 bx + c is
a factor of f (x), both coefficients equal 0. Finally,

0 = b(b4 + 3b2 c + c2 1) 3(b3 c + 2bc2 + a) = b5 b 5bc2 3a

implies 3a = b5 b 5bc2 is divisible by 5. Then a would be divisible

by 5, a contradiction.

10. (1991 Soviet Math Olympiad) Given 2n distinct numbers a1 , a2 , . . . , an,

b1 , b2 , . . . , bn , an n n table is filled as follows: into the cell in the i-th
row and j-th column is written the number ai + bj . Prove that if the
product of each column is the same, then also the product of each row
is the same.

Solution. Let

P (x) = (x + a1 )(x + a2 ) (x + an ) (x b1 )(x b2 ) (x bn ),

then deg P < n. Now P (bj ) = (bj + a1 )(bj + a2 ) (bj + an ) = c, some

constant, for j = 1, 2, . . . , n. So P (x)c has distinct roots b1 , b2 , , bn.
Therefore, P (x) = c for all x and so

c = P (ai ) = (1)n+1 (ai + b1 )(ai + b2 ) (ai + bn )

for i = 1, 2, . . . , n. Then the product of each row is (1)n+1 c.

11. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two distinct collections of n pos-

itive integers, where each collection may contain repetitions. If the two

39
 collections of integers ai +aj (1 i < j n) and bi +bj (1 i < j n)
are the same, then show that n is a power of 2.

Solution. (Due to Law Siu Lung) Consider the functions f (x) =

Xn Xn
ai
x and g(x) = xbi . Since the ai s and bi s are distinct, f and g
i+1 i=1
are distinct polynomials. Now
X
n X X
2
f (x) = x2ai + 2 xai +aj = f (x2 ) + 2 xai +aj .
i=1 1i<jn 1i<jn

Since the ai + aj s and the bi + bj s are the same, so f (x)2 f (x2 ) =

g(x)2 g(x2 ). Since f (1) g(1) = n n = 0, so f (x) g(x) = (x
1)k Q(x) for some k 1 and polynomial Q such that Q(1) 6= 0. Then

f (x2 ) g(x2 ) (x2 1)k Q(x2 ) 2

k Q(x )
f (x) + g(x) = = = (x + 1) .
f (x) g(x) (x 1)k Q(x) Q(x)

Setting x = 1, we have n = 2k1 .

Recurrence Relations

12. The sequence xn is defined by

2 + xn
x1 = 2, xn+1 = , n = 1, 2, 3, . . . .
1 2xn
1
Prove that xn 6= 2
or 0 for all n and the terms of the sequence are all
distinct.

Solution. (Due to Wong Chun Wai) The terms xn s are clearly rational
by induction. Write xn = pn /qn , where pn , qn are relatively prime inte-
gers and qn > 0. Then q1 = 1 and pn+1 /qn+1 = (2qn + pn )/(qn 2pn ).
So qn+1 divides qn 2pn , which implies every qn is odd by induction.
Hence, every xn 6= 12 .
Next, to show every xn 6= 0, let = arctan 2, then xn = tan n
by induction. Suppose xn = 0 and n is the least such index. If n =

40
 2m is even, then 0 = x2m = tan 2m = 2xm /(1 x2m ) would imply
xm = 0, a contradiction to n being least. If n = 2m + 1 is odd, then
0 = x2m+1 = tan( + 2m) = (2 + x2m )/(1 2x2m ) would imply
2
x2m = 2. Then 2 = 2xm /(1 xm ) would imply xm = (1 5)/2 is
irrational, a contradiction. Finally, if xm = xn for some m > n, then
xmn = tan(m n) = (xm xn )/(1 + xm xn ) = 0, a contradiction.
Therefore the terms are nonzero and distinct.

13. (1988 Nanchang City Math Competition) Define a1 = 1, a2 = 7 and

a2n+1 1
an+2 = for positive integer n. Prove that 9an an+1 + 1 is a
an
perfect square for every positive integer n.

Solution. (Due to Chan Kin Hang) (Since an+2 depends on an+1 and
an , it is plausible that the sequence satisfies a linear recurrence relation
an+2 = can+1 + c0 an . If this is so, then using the first 4 terms, we find
c = 7, c0 = 1.) Define b1 = a1 , b2 = a2 , bn+2 = 7bn+1 bn for n 1.
Then b3 = 48 = a3 . Suppose ak = bk for k n + 1, then

b2n+1 1 (7bn bn1 )2 1

an+2 = =
bn bn
= 49bn 14bn1 + bn2
= 7bn+1 bn = bn+2 .

So ak = bk for all k.
Next, writing out the first few terms of 9an an+1 + 1 will suggest
that 9an an+1 + 1 = (an + an+1 )2 . The case n = 1 is true as 9 7 + 1 =
(1 + 7)2 . Suppose this is true for n = k. Using the recurrence relations
and (*) 2a2k+1 2 = 2ak ak+2 = 14ak ak+1 2a2k , we get the case n = k+1
as follow:
9ak+1 ak+2 + 1 = 9ak+1 (7ak+1 ak ) + 1
= 63a2k+1 (ak + ak+1 )2 + 2
= 62a2k+1 2ak ak+1 a2k + 2
= 64a2k+1 16ak+1 ak + a2k by (*)
= (8ak+1 ak )2 = (ak+1 + ak+2 )2 .

41
14. (Proposed by Bulgaria for 1988 IMO) Define a0 = 0, a1 = 1 and an =
2an1 + an2 for n > 1. Show that for positive integer k, an is divisible
by 2k if and only if n is divisible by 2k .
n
Solution.
n By the binomial
theorem, if (1 + 2) = An + B n 2, then
(1 2) = An Bn 2. Multiplying these 2 equations, we get A2n
2Bn2 = (1)n . This implies An is always odd. Using characteristic
equation method to solve the given recurrence relations on an , we find
that an = Bn . Now write n = 2k m, where m is odd. We have k = 0
(i.e. n is odd) if and only if 2Bn2 = A2n + 1 2 2n
(mod 4), (i.e. Bn is
odd). Nextsuppose case k is true. Since (1+ 2) = (An +Bn 2)2 =
A2n + B2n 2, so B2n = 2An Bn . Then it follows case k implies case
k + 1.

15. (American Mathematical Monthly, Problem E2998) Let x and y be

xn y n
distinct complex numbers such that is an integer for some
xy
xn y n
four consecutive positive integers n. Show that is an integer
xy
for all positive integers n.

Solution. For nonnegative integer n, let tn = (xn y n )/(x y). So

t0 = 0, t1 = 1 and we have a recurrence relation

tn+2 + btn+1 + ctn = 0, where b = (x + y), c = xy.

Suppose tn is an integer for m, m + 1, m + 2, m + 3. Since cn = (xy)n =

t2n+2 tn tn+2 is an integer for n = m, m + 1, so c is rational. Since
cm+1 is integer, c must, in fact, be an integer. Next
tm tm+3 tm+1 tm+2
b= .
cm
So b is rational. From the recurrence relation, it follows by induction
that tn = fn1 (b) for some polynomial fn1 of degree n1 with integer
coefficients. Note the coefficient of xn1 in fn1 is 1, i.e. fn1 is monic.
Since b is a root of the integer coefficient polynomial fm (z) tm+1 = 0,
b must be an integer. So the recurrence relation implies all tn s are
integers.

42
 Inequalities

16. For real numbers a1 , a2 , a3 , . . . , if an1 + an+1 2an for n = 2, 3, . . . ,

then prove that

An1 + An+1 2An for n = 2, 3, . . . ,

where An is the average of a1 , a2 , . . . , an .

Solution. Expressing in ak , the required inequality is equivalent to

n2 + n 2 n(n 1)
a1 + + an1 an + an+1 0.
2 2

(From the cases n = 2, 3, we easily see the pattern.) We have

n2 + n 2 n(n 1)
a1 + + an1 an + an+1
2 2
X
n
k(k 1)
= (ak1 2ak + ak+2 ) 0.
2
k=2

17. Let a, b, c > 0 and abc 1. Prove that

a b c
+ + a + b + c.
c a b

Solution. (Due to Leung Wai Ying) Since abc 1, we get 1/(bc) a,

1/(ac) b and 1/(ab) c. By the AM-GM inequality,
r
2a c a a c 2
3 a
+ = + + 3 3a.
c b c c b bc

Similarly, 2b/a + a/c 3b and 2c/b + b/a 3c. Adding these and
dividing by 3, we get the desired inequality.

43
 p p p
Alternatively, let x = 9 a4 b/c2 , y = 9 c4 a/b

2 and z = 9 b4 c/a2 .

We have a = x2 y, b = z 2 x, c = y 2 z and xyz = abc 1. Using this

3

and the rearrangement inequality, we get

a b c x2 z2 y2
+ + = + +
c a b yz xy zx
x2 z2 y2

xyz + + = x3 + y 3 + z 3
yz xy zx
x2 y + y 2 z + z 2 x = a + b + c.

18. (1982 Moscow Math Olympiad) Use the identity 13 + 23 + + n3 =

n2 (n + 1)2
to prove that for distinct positive integers a1 , a2 , . . . , an ,
4
(a71 + a72 + + a7n ) + (a51 + a52 + + a5n ) 2(a31 + a32 + + a3n )2 .

Can equality occur?

Solution. For n = 1, a71 + a51 2(a31 )2 = a51 (a1 1)2 0 and so case
n = 1 is true. Suppose the case n = k is true. For the case n = k + 1,
without loss of generality, we may assume a1 < a2 < . . . < ak+1 . Now

2(a31 + + a3k+1 )2 2(a31 + + a3k )2

=2a6k+1 + 4a3k+1 (a31 + + a3k )
2a6k+1 + 4a3k+1 (13 + 23 + + (ak+1 1)3 )
(ak+1 1)2 a2k
=2a6k+1 + 4a3k+1 = a7k+1 + a5k+1 .
4
So (a71 + + a7k+1 ) + (a51 + + a5k+1 ) 2(a31 + + a3k+1 )2 follows.
Equality occurs if and only if a1 , a2 , . . . , an are 1, 2, . . . , n.

19. (1997 IMO shortlisted problem) Let a1 an an+1 = 0 be a

sequence of real numbers. Prove that
v
u n n
uX X
t ak k( ak ak+1 ).
k=1 k=1

44

Solution. (Due to Lee Tak Wing) Let xk = ak ak+1 . Then ak =
(xk + xk+1 + + xn )2 . So,
Xn Xn n
X X
2
ak = (xk + xk+1 + + xn ) = kx2k + 2 ixi xj
k=1 k=1 k=1 1i<jn
Xn X n
X
p
2
kx2k +2 ijxi xj = kxk .
k=1 1i<jn k=1

Taking square root of both sides, we get the desired inequality.

20. (1994 Chinese Team Selection Test) For 0 a b c d e and

a + b + c + d + e = 1, show that
1
ad + dc + cb + be + ea .
5

Solution. (Due to Lau Lap Ming) Since a b c d e, so

d + e c + e b + d a + c a + b. By Chebysevs inequality,
ad + dc + cb + be + ea
a(d + e) + b(c + e) + c(b + d) + d(a + c) + e(a + b)
=
2
(a + b + c + d + e)((d + e) + (c + e) + (b + d) + (a + c) + (a + b))

10
1
= .
5

21. (1985 Wuhu City Math Competition) Let x, y, z be real numbers such
that x + y + z = 0. Show that
6(x3 + y 3 + z 3 )2 (x2 + y 2 + z 2 )3 .

Solution. (Due to Ng Ka Wing) We have z = (x + y) and so

(x2 + y 2 + z 2 )3 = (x2 + y 2 + (x + y)2 )3
3
3 27
(x + y)2 = (x + y)4 (x + y)2
2 8
27 4
2
(2 xy) (x + y)2 = 6 3xy(x + y)
8

2
= 6 x3 + y 3 (x + y)3 = 6(x3 + y 3 + z 3 )2 .

45
 Comments. Let f (w) = (w x)(w y)(w z) = w 3 + bw + c. Then
x2 +y 2 +z 2 = (x+y +z)2 2(xy +yz +zx) = 2b and 0 = f (x)+f (y)+
f (z) = (x3 + y 3 + z 3 ) + b(x + y + z) + 3c implies x3 + y 3 + z 3 = 3c.
So the inequality is the same as 2(4b3 27c2 ) 0. For the cubic
polynomial f (w) = w 3 + bw + c, it is well-known that the discriminant
4 = (x y)2 (y z)2 (z x)2 equals 4b3 27c2 . The inequality follows
easily from this.

22. (1999 IMO) Let n be a fixed integer, with n 2.

(a) Determine the least constant C such that the inequality

X  X 4
xi xj (x2i + x2j ) C xi
1i<jn 1in

holds for all nonnegative real numbers x1 , x2 , . . . , xn .

(b) For this constant C, determine when equality holds.

Solution. (Due to Law Ka Ho and Ng Ka Wing) We will show the

least C is 1/8. By the AM-GM inequality,
 X 4  X X 2
xi = x2i +2 xi xj
1in 1in 1i<jn
 s X X 2
2 2 xi xj x2i
1i<jn 1in
X
=8 xi xj (x21 + + x2n )
1i<jn
X
8 xi xj (x2i + x2j ).
1i<jn

(Equality holds in the second inequality if and only if at least n 2

of the xi s are zeros. Then equality holds in the first inequality if and
only if the remaining pair of xi s are equal.) Overall, equality holds if
and only if two of the xi s are equal and the others are zeros.

46
23. (1995 Bulgarian Math Competition) Let n 2 and 0 xi 1 for
i = 1, 2, . . . , n. Prove that
hni
(x1 + x2 + + xn ) (x1 x2 + x2 x3 + + xn1 xn + xn x1 ) ,
2

where [x] is the greatest integer less than or equal to x.

Solution. When x2 , . . . , xn are fixed, the left side is a degree one poly-
nomial in x1 , so the maximum value is attained when x1 = 0 or 1.
The situation is similar for the other xi s. So when the left side is
maximum, every xi s is 0 or 1 and the value is an integer. Now


2 (x1 + + xn ) (x1 x2 + x2 x3 + + xn1 xn + xn x1 )
=n (1 x1 )(1 x2 ) (1 x2 )(1 x3 ) . . . (1 xn )(1 x1 )
x1 x2 x2 x3 . . . xn x1 .

Since 0 xi 1, the expression above is at most n. So

hni
max (x1 + + xn ) (x1 x2 + x2 x3 + + xn1 xn + xn x1 ) .
2

24. For every triplet of functions f, g, h : [0, 1] R, prove that there are
numbers x, y, z in [0, 1] such that

1
|f (x) + g(y) + h(z) xyz| .
3

Solution. Suppose for all x, y, z in [0, 1], |f (x) + g(y) + h(z) xyz| <
1/3. Then

1 1
|f (0) + g(0) + h(0)| < , |f (0) + g(y) + h(z)| < ,
3 3

1 1
|f (x) + g(0) + h(z)| < , |f (x) + g(y) + h(0)| < .
3 3
47
 Since
f (0) + g(y) + h(z) f (x) + g(0) + h(z)
f (x) + g(y) + h(z) = +
2 2
f (x) + g(y) + h(0) f (0) g(0) h(0)
+ + ,
2 2
by the triangle inequality, |f (x) + g(y) + h(z)| < 2/3. In particular,
|f (1) + g(1) + h(1)| < 2/3. However, |1 f (1) g(1) h(1)| < 1/3.
Adding these two inequality and applying the triangle inequality to the
left side, we get 1 < 1, a contradiction.
There is a simpler proof. By the triangle inequality, the sum of
| (f (0) + g(0) + h(0))|, |f (0) + g(1) + h(1)|, |f (1) + g(0) + h(1)|,
|f (1)+g(1)+h(0)|, |(f (1)+g(1)+h(1)1)|, |(f (1)+g(1)+h(1)1)|
is at least 2. So, one of them is at least 2/6.

25. (Proposed by Great Britain for 1987 IMO) If x, y, z are real numbers
such that x2 + y 2 + z 2 = 2, then show that x + y + z xyz + 2.

Solution. (Due to Chan Ming Chiu) If one of x, y, z is nonpositive,

say z, then

2 + xyz x y z = (2 x y) z(1 xy) 0

p
because x + y 2(x2 + y 2 ) 2 and xy (x2 + y 2 )/2 1. So we
may assume x, y, z are positive, say 0 < x y z. If z 1, then

2 + xyz x y z = (1 x)(1 y) + (1 z)(1 xy) 0.

If z > 1, then
p p
(x + y) + z 2((x + y)2 + z 2 ) = 2 xy + 1 xy + 2 xyz + 2.

26. (Proposed by USA for 1993 IMO) Prove that for positive real numbers
a, b, c, d,

a b c d 2
+ + + .
b + 2c + 3d c + 2d + 3a d + 2a + 3b a + 2b + 3c 3

48
 Solution. Let
r
a p
x1 = , y1 = a(b + 2c + 3d),
b + 2c + 3d
r
b p
x2 = , y2 = b(c + 2d + 3a),
c + 2d + 3a
r
c p
x3 = , y3 = c(d + 2a + 3b),
d + 2a + 3b
r
d p
x4 = , y4 = d(a + 2b + 3c).
a + 2b + 3c
The inequality to be proved is x21 + x22 + x23 + x24 2/3. By the Cauchy-
Schwarz inequality, (x21 + + x24 )(y12 + + y42 ) (a + b + c + d)2 . To
finish, it suffices to show (a + b + c + d)2 /(y12 + y22 + y32 + y42 ) 2/3.
This follows from
3(a + b + c + d)2 2(y12 + y22 + y32 + y42 )
=3(a + b + c + d)2 8(ab + ac + ad + bc + bd + cd)
=(a b)2 + (a c)2 + (a d)2 + (b c)2 + (b d)2 + (c d)2 0.

27. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be 2n positive real numbers such

that
(a) a1 a2 an and
(b) b1 b2 bk a1 a2 ak for all k, 1 k n.
Show that b1 + b2 + + bn a1 + a2 + + an .

Solution. Let ck = bk /ak and dk = (c1 1)+(c2 1)+ +(ck 1) for

1 k n. By the AM-GM inequality and (b), (c1 + c2 + + ck )/k

k
c1 c2 ck 1, which implies dk 0. Finally,

(b1 + b2 + + bn ) (a1 + a2 + + an )
=(c1 1)a1 + (c2 1)a2 + + (cn 1)an
=d1 a1 + (d2 d1 )a2 + + (dn dn1 )an
=d1 (a1 a2 ) + d2 (a2 a3 ) + + dn an 0.

49
28. (Proposed by Greece for 1987 IMO) Let a, b, c > 0 and m be a positive
integer, prove that
 m1
am bm cm 3 a+b+c
+ + .
b+c c+a a+b 2 3

Solution. Without loss of generality, assume a b c. So a + b

1 1 1
c + a b + c, which implies . By the AM-HM
b+c c+a a+b
inequality,
(b + c) + (c + a) + (a + b) 3
.
3 1 1 1
+ +
b+c c+a a+b
1 1 1 9
This yields + + . By the Chebysev in-
b+c c+a a+b 2(a + b + c)
equality and the power mean inequality respectively, we have
am bm cm 1 1 1 1

+ + (am + bm + cm ) + +
b+c c+a a+b 3 b+c c+a a+b
a + b + c
m 9

3 2(a + b + c)
3 a + b + c
m1
= .
2 3

29. Let a1 , a2 , . . . , an be distinct positive integers, show that

a1 a2 an 1
+ + + n 1 n.
2 8 n2 2

Solution. (Due to Chan Kin Hang) Arrange a1 , a2 , . . . , an into in-

creasing order as b1 , b2 , . . . , bn . Then bn n because they are distinct
1 1 1
positive integers. Since , , . . . , , by the rearrangement inequal-
2 8 n2n
ity,
a1 a2 an b1 b2 bn
+ + + n + + + n
2 8 n2 2 8 n2
1 2 n 1
+ + + n = 1 n.
2 8 n2 2
50
30. (1982 West German Math Olympiad) If a1 , a2 , . . . , an > 0 and a =
a1 + a2 + + an , then show that
n
X ai n
.
2a ai 2n 1
i=1

Solution. By symmetry, we may assume a1 a2 . . . an . Then

1 1
... For convenience, let ai = aj if i j (mod n).
2a an 2a a1 .
For m = 0, 1, . . . , n 1, by the rearrangement inequality, we get
n
X n
X
am+i ai
.
2a ai 2a ai
i=1 i=1

n
X n
X nai
a
Adding these n inequalities, we get . Since
i=1
2a ai i=1
2a a i
a 1 1 ai
= + , we get
2a ai 2 2 2a ai
n n
n 1 X ai X ai
+ n .
2 2 i=1 2a ai i=1
2a a i

Solving for the sum, we get the desired inequality.

a3 b3 c3 a2 + b2 + c2
31. Prove that if a, b, c > 0, then + + .
b+c c+a a+b 2

Solution. (Due to Ho Wing Yip) By symmetry, we may assume a

1 1 1
b c. Then a + b c + a b + c. So . By the
b+c c+a a+b
rearrangement inequality, we have

a3 b3 c3 a3 b3 c3
+ + + + ,
a+b b+c c+a b+c c+a a+b

a3 b3 c3 a3 b3 c3
+ + + + .
c+a a+b b+c b+c c+a a+b

51
 Adding these, then dividing by 2, we get

1 a3 + b3 b3 + c3 c3 + a3
a3 b3 c3
+ + + + .
2 a+b b+c c+a b+c c+a a+b

Finally, since (x3 + y 3 )/(x + y) = x2 xy + y 2 (x2 + y 2 )/2, we have

a2 + b2 + c2 1 a2 + b2 b2 + c2 c2 + a2

= + +
2 2 2 2 2
1 a +b3 3 3
b +c 3
c + a3

3
+ +
2 a+b b+c c+a
3 3 3
a b c
+ + .
b+c c+a a+b

32. Let a, b, c, d > 0 and

1 1 1 1
+ + + = 1.
1 + a4 1 + b4 1 + c4 1 + d4

Prove that abcd 3.

Solution. Let a2 = tan , b2 = tan , c2 = tan , d2 = tan . Then

cos2 + cos2 + cos2 + cos2 = 1. By the AM-GM inequality,

sin2 = cos2 + cos2 + cos2 3(cos cos cos )2/3 .

Multiplying this and three other similar inequalities, we have

sin2 sin2 sin2 sin2 81 cos2 cos2 cos2 cos2 .

Then abcd = tan tan tan tan 3.

33. (Due to Paul Erdos) Each of the positive integers a1 , . . . , an is less than
1951. The least common multiple of any two of these is greater than
1951. Show that
1 1 n
++ <1+ .
a1 an 1951

52
 Solution. Observe that none of the numbers 1, 2, . . . , 1951 is a common
multiple of more than one ai s. The number of multiples of ai among
1, 2, . . . , 1951 is [1951/ai]. So we have [1951/a1 ]+ +[1951/an ] 1951.
Since x 1 < [x], so
1951
1951

1 + + 1 < 1951.
a1 an
Dividing by 1951 and moving the negative terms to the right, we get
the desired inequality.

34. A sequence (Pn ) of polynomials is defined recursively as follows:

x Pn (x)2
P0 (x) = 0 and for n 0, Pn+1 (x) = Pn (x) + .
2
Prove that
2
0 x Pn (x)
n+1
for every nonnegative integer n and all x in [0, 1].

Solution. (Due to Wong Chun Wai) For x in [0, 1],

x + Pn (x)

x Pn+1 (x) = ( x Pn (x)) 1 .
2

By induction, we can show that 0 Pn (x) x 1 for all x in [0, 1].
Then
n1 n1
x Pn (x) Y x Pk+1 (x) Y x + Pk (x)
x
n
= = 1 1 .
x x Pk (x) 2 2
k=0 k=0

Multiplying both sides by x and applying the AM-GM inequality, we
have
x
n
0 x Pn (x) x 1

2
 n x x n+1
2 2 x + (1 2 ) + + (1 2 )

n n+1
2 n
n+1 2
= .
n n+1 n+1

53
35. (1996 IMO shortlisted problem) Let P (x) be the real polynomial func-
tion, P (x) = ax3 + bx2 + cx + d. Prove that if |P (x)| 1 for all x such
that |x| 1, then
|a| + |b| + |c| + |d| 7.

Solution. Note the four polynomials P (x) satisfy the same condi-
tions as P (x). One of these have a, b 0. The problem stays the same
if P (x) is replaced by this polynomial. So we may assume a, b 0.
Case c in [0, +). If d 0, then |a| + |b| + |c| + |d| = a + b + c + d =
P (1) 1. If d < 0, then

|a| + |b| + |c| + |d| = a + b + c + d + 2(d) = P (1) 2P (0) 3.

Case c in (, 0). If d 0, then

|a| + |b| + |c| + |d| = a + b c + d

4 1 8 1 8 1
= P (1) P (1) P ( ) + P ( )
3 3 3 2 3 2
4 1 8 8
+ + + = 7.
3 3 3 3
If d < 0, then

|a| + |b| + |c| + |d| = a + b c d

5 1 4 1
= P (1) 4P ( ) + P ( )
3 2 3 2
5 4
+ 4 + = 7.
3 3
Comments. Tracing the equality cases, we see that the maximum 7 is
obtained by P (x) = (4x3 3x) only.

36. (American Mathematical Monthly, Problem 4426) Let P (z) = az 3 +

bz 2 + cz + d, where a, b, c, d arecomplex numbers with |a| = |b| = |c| =
|d| = 1. Show that |P (z)| 6 for at least one complex number z
satisfying |z| = 1.

54
 Solution. (Due to Yung Fai) We have aa = |a|2 = 1 and similarly for
b, c, d. Using w + w = 2Re w, we get

|P (z)|2 = (az 3 + bz 2 + cz + d)(az 3 + bz 2 + cz + d)

= 4 + 2Re adz 3 + (ac + bd)z 2 + (ab + bc + cd)z .

Let Q(z) = adz 3 + (ac + bd)z 2 + (ab + bc + cd)z, then |P (z)|2 = 4 +

2Re Q(z). Now we use the roots of unity trick! Let be a cube root
of unity not equal to 1. Since 1 + + 2 = 0 and 1 + 2 + 4 = 0, so

Q(z) + Q(z) + Q( 2 z)
=3adz 3 + (ac + bd)(1 + + 2 ) + (ad + bc + cd)(1 + 2 + 4 )z
=3adz 3 .

If we now choose z to be a cube root of ad, then |z| = 1 and Re Q(z) +

Re Q(z) + Re Q( 2 z) = 3. So |P (z)|2 + |P (z)|
2
+ |P ( 2 z)|2 = 18.
Then one of |P (z)|, |P (z)|, |P ( 2 z)| is at least 6.

37. (1997 Hungarian-Israeli Math Competition) Find all real numbers

with the following property:
  for any positive integer n, there exists an
 m  1
integer m such that   < ?
n 3n
Solution. The condition holds if and only if x is an integer. If x
is an integer, then for any n, take m = nx. Conversely, suppose the
condition holds for x. Let mk be the integer corresponding to n =
2k , k = 0, 1, 2, . . . . By the triangle inequality,
m mk+1   mk   mk+1  1 1 1
 k  
 k k+1   k x + x k+1  < + = .
2 2 2 2 3 2k 3 2k+1 2k+1

Since the leftmost expression is |2mk mk+1 |/2k+1 , the inequalities

imply it is 0, that is mk /2k = mk+1 /2k+1 for every k. Then |x m0 | =
|x (mk /2k )| 1/(3 2k ) for every k. Therefore, x = m0 is an integer.

38. (1979 British Math Olympiad) If n is a positive integer, denote by p(n)

the number of ways of expressing n as the sum of one or more positive

55
 integers. Thus p(4) = 5, as there are five different ways of expressing
4 in terms of positive integers; namely

1 + 1 + 1 + 1, 1 + 1 + 2, 1 + 3, 2 + 2, and 4.

Prove that p(n + 1) 2p(n) + p(n 1) 0 for each n > 1.

Solution. The required inequality can be written as p(n + 1) p(n)

p(n) p(n 1). Note that adding a 1 to each p(n 1) sums of n 1
will yield p(n 1) sums of n. Conversely, for each sum of n whose
least summand is 1, removing that 1 will results in a sum of n 1. So
p(n) p(n 1) is the number of sums of n whose least summands are
at least 2. For every one of these p(n) p(n 1) sums of n, increasing
the largest summand by 1 will give a sum of n + 1 with least summand
at least 2. So p(n + 1) p(n) p(n) p(n 1).

Functional Equations

39. Find all polynomials f satisfying f (x2 ) + f (x)f (x + 1) = 0.

Solution. If f is constant, then f is 0 or 1. If f is not constant,

then let z be a root of f. Setting x = z and x = z 1, respectively,
we see that z 2 and (z 1)2 are also roots, respectively. Since f has
n
finitely many roots and z 2 are all roots, so we must have |z| = 0 or
1. Since z is a root implies (z 1)2 is a root, |z 1| also equals 0
or 1. It follows that z = 0 or 1. Then f (x) = cxm (x 1)n for some
real c and nonnegative integers m, n. If c 6= 0, then after simplifying
the functional equation, we will see that n = m and c = 1. Therefore,
f (x) = 0 or xn (1 x)n for nonnegative integer n.

40. (1997 Greek Math Olympiad) Let f : (0, ) R be a function such

that
(a) f is strictly increasing,
(b) f (x) > x1 for all x > 0 and
(c) f (x)f (f (x) + x1 ) = 1 for all x > 0.
Find f (1).

56
 Solution. Let t = f (1). Setting x = 1 in (c), we get tf (t + 1) = 1. So
t 6= 0 and f (t +1) = 1/t. Setting x = t +1 in (c), we get f (t +1)f (f (t +
1
1)+ t+1 ) = 1. Then f ( 1t + t+1
1
) = t = f (1). Since f is strictly increasing,
1 1

t
+ t+1
= 1. Solving, we get t = (1 5)/2. If t = (1 + 5)/2 > 0,
1
then 1 < t = f (1) < f (1 + t) = t < 1, a contradiction. Therefore,

f (1) = t = (1 5)/2. (Note f (x) = (1 5)/(2x) is such a function.)

41. (1979 Eotvos-Kurschak Math Competition) The function f is defined

for all real numbers and satisfies f (x) x and f (x + y) f (x) + f (y)
for all real x, y. Prove that f (x) = x for every real number x.

Solution. (Due to Ng Ka Wing) Since f (0 + 0) f (0) + f (0), so

0 f (0). Since f (0) 0 also, we get f (0) = 0. For all real x,

0 = f (x + (x)) f (x) + f (x) x + (x) = 0.

So f (x) + f (x) = 0, hence f (x) = f (x) for all real x. Since

f (x) x, so x f (x) = f (x) x. Therefore, f (x) = x for
all real x.

42. (Proposed by Ireland for 1989 IMO) Suppose f : R R satisfies

f (1) = 1, f (a + b) = f (a) + f (b) for all a, b R and f (x)f ( 1x ) = 1 for
x 6= 0. Show that f (x) = x for all x.

Solution. (Due to Yung Fai) From f (0 + 0) = f (0) + f (0), we get

f (0) = 0. From 0 = f (x + (x)) = f (x) + f (x), we get f (x) =
f (x). By induction, f (nx) = nf (x) for positive integer n. For x = n1 ,
1 = f (1) = f (n n1 ) = nf ( n1 ). Then f ( n1 ) = n1 and f ( m
n
) = f (m n1 ) =
mf ( n1 ) = m
n . So f (x) = x for rational x. (The argument up to this point
is well-known. The so-called Cauchys equation f (a + b) = f (a) + f (b)
implies f (x) = f (1)x for rational x.)
1
Next we will show f is continuous at 0. For 0 < |x| < 2n , we have
1 1 1 1
| nx | > 2. So there is w such that w + w = nx . We have |f ( nx )| =
q 1 1
|f (w) + f ( w1 )| 2 f (w)f ( w1 ) = 2. So |f (x)| = 1 . Then
n|f ( nx )| 2n
lim f (x) = 0 = f (0).
x0

57
 Now for every real x, let rn be a rational number agreeing with
x to n places after the decimal point. Then lim (x rn ) = 0. By
n
continuity at 0, f (x) = lim (f (x rn ) + f (rn )) = lim rn = x. There-
n n
fore, f (x) = x for all x. (This first and third paragraphs show the
Cauchy equation with continuity at a point has the unique solution
f (x) = f (1)x.)

43. (1992 Polish Math Olympiad) Let Q+ be the positive rational numbers.
Determine all functions f : Q+ Q+ such that f (x + 1) = f (x) + 1
and f (x3 ) = f (x)3 for every x Q+ .

Solution. From f (x + 1) = f (x) + 1, we get f (x + n) = f (x) + n for

all positive integer n. For pq Q+ , let t = f ( pq ). On one hand,

p p3
f (( + q ) ) = f ( 3 + 3p2 + 3pq 3 + q 6 ) = t3 + 3p2 + 3pq 3 + q 6
2 3
q q

and on the other hand,

p p
f (( + q 2 )3 ) = (f ( ) + q 2 )3 = t3 + 3t2 q 2 + 3tq 4 + q 6 .
q q

Equating the right sides and simplifying the equation to a quadratic in

t, we get the only positive root t = pq . So f (x) = x for all x Q+ .

44. (1996 IMO shortlisted problem) Let R denote the real numbers and
f : R [1, 1] satisfy
     
13 1 1
f x+ + f (x) = f x+ +f x+
42 6 7

for every x R. Show that f is a periodic function, i.e. there is a

nonzero real number T such that f (x + T ) = f (x) for every x R.

Solution. Setting x = w + k6 for k = 0, 1, . . . , 5, we get 6 equations.

Adding these and cancelling terms, we will get f (w + 87 ) + f (w) =
f (w + 1) + f (w + 17 ) for all w. Setting w = z + k7 for k = 0, 1, . . . , 6

58
 in this new equation, we get 7 equations. Adding these and cancelling
terms, we will get f (z +2)+f (z) = 2f (z +1) for all z. Rewriting this as
f (z +2)f (z +1) = f (z +1)f (z), we see that f (z +n)f (z +(n1))
is a constant, say c. If c 6= 0, then
k
X
f (z + k) = (f (z + n) f (z + (n 1))) + f (z)
n=1
= kc + f (z) 6 [1, 1]
for large k, a contradiction. So c = 0 and f (z + 1) = f (z) for all z.

45. Let N denote the positive integers. Suppose s : N N is an increasing

function such that s(s(n)) = 3n for all n N. Find all possible values
of s(1997).

Solution. (Due to Chan Kin Hang) Note that if s(m) = s(n), then
3m = s(s(m)) = s(s(n)) = 3n implies m = n. From this, we see
that s is strictly increasing. Next we have n < s(n) for all n (otherwise
s(n) n for some n, which yields the contradiction that 3n = s(s(n))
s(n) n.) Then s(n) < s(s(n)) = 3n. In particular, 1 < s(1) < 3
implies s(1) = 2 and s(2) = s(s(1)) = 3. With the help of s(3n) =
s(s(s(n))) = 3s(n), we get s(3k ) = 23k and s(23k ) = s(s(3k )) = 3k+1 .
Now there are 3k 1 integers in each of the open intervals (3k , 23k )
and (23k , 3k+1 ). Since f is strictly increasing, we must have s(3k +j) =
2 3k + j for j = 1, 2, . . . , 3k 1. Then s(2 3k + j) = s(s(3k + j)) =
3(3k +j). Since 1997 = 236 +539 < 37 , so s(1997) = 3(36 +539) = 3804.

46. Let N be the positive integers. Is there a function f : N N such that

f (1996) (n) = 2n for all n N, where f (1) (x) = f (x) and f (k+1) (x) =
f (f (k) (x))?

Solution. For such a function f (2n) = f (1997) (n) = f (1996) (f (n)) =

2f (n). So if n = 2e q, where e, q are nonnegative integers and q odd,
then f (n) = 2e f (q). To define such a function, we need to define it at
odd integer q. Now define

q+2 if q 1, 3, . . . , 3989 (mod 3992)
f (q) =
2(q 3990) if q 3991 (mod 3992)

59
 and f (2n) = 2f (n) for positive integer n. If q = 3992m + (2j 1), j =
1, 2, . . . , 1995, then f (1996j) (q) = q + 2(1996 j) = 3992m + 3991,
f (1997j) (q) = 2(3992m + 1) and
f (1996) (q) = 2f (j1) (3992m + 1) = 2(3992m + 1 + (2j 1)) = 2q.
If q = 3992m + 3991, then f (q) = 2(3992m + 1) and
f (1996) (q) = 2f (1995) (3992m + 1) = 2(3992m + 1 + 2 1995) = 2q.
So f (1996) (q) = 2q for odd q. If n = 2e q, then
f (1996) (n) = 2e f (1996) (q) = 2e (2q) = 2n.

47. (American Mathematical Monthly, Problem E984) Let R denote the

real numbers. Find all functions f : R R such that f (f (x)) = x2 2
or show no such function can exist.

Solution. Let g(x) = x2 2 and suppose f (f (x)) = g(x). Put h(x) =

g(g(x)) = x4 4x2 + 2. The fixed points of g (i.e. the solutions of the
equation g(x) = x) are 1 and 2. The set of fixed points of h contains
the fixed points of g and is S = {1, 2, (1 5)/2}. Now observe that
x S implies h(f (x)) = f (h(x)) = f (x), i.e. f (x) S. Also, x, y S
and f (x) = f (y) imply x = h(x) = h(y) = y. So f is a bijection S S.
If c = 1 or 2, then g(f (c)) = f (f (f (c))) = f (g(c))
= f (c) and
consequently {f (1), f (2)} = {1, 2}. For a = (1 + 5)/2, since f
induces a bijection S S andg(a) = a2 2 6= a implies f (a) 6= a, we
must have f (a) = b = (1 5)/2. It follows that f (b) = a and we
have a contradiction a = f (b) = f (f (a)) = g(a).

48. Let R be the real numbers. Find all functions f : R R such that for
all real numbers x and y,


f xf (y) + x = xy + f (x).

Solution 1. (Due to Leung Wai Ying) Putting x = 1, y = 1 f (1)

and letting a = f (y) + 1, we get


f (a) = f f (y) + 1 = y + f (1) = 1.

60
 Putting y = a and letting b = f (0), we get


b = f xf (a) + x = ax + f (x),

so f (x) = ax + b. Putting this into the equation, we have

a2 xy abx ax + b = xy ax + b.

Equating coefficients, we get a = 1 and b = 0, so f (x) = x or f (x) =

x. We can easily check both are solutions.

Solution 2. Setting x = 1, we get

f f (y) + 1 = y + f (1).


For every real number a, let y = a f (1), then f f (y) + 1 = a and
f is surjective. In particular, there is b such that f (b) = 1. Also, if
f (c) = f (d), then

c + f (1) = f f (c) + 1


= f f (d) + 1
= d + f (1).


So c = d and f is injective. Taking x = 1, y = 0, we get f f (0) + 1 =
f (1). Since f is injective, we get f (0) = 0.
For x 6= 0, let y = f (x)/x, then


f xf (y) + x = 0 = f (0).

By injectivity, we get xf (y) + x = 0. Then

f f (x)/x = f (y) = 1 = f (b)

and so f (x)/x = b for every x 6= 0. That is, f (x) = bx. Putting

this into the given equation, we find f (x) = x or f (x) = x, which are
easily checked to be solutions.

49. (1999 IMO) Determine all functions f : R R such that

f (x f (y)) = f (f (y)) + xf (y) + f (x) 1

61
 for all x, y in R.

Solution. Let A be the range of f and c = f (0). Setting x = y = 0, we

2
get f (c) = f (c) + c 1. So c 6= 0. For x = f (y) A, f (x) = c+1
2
x2 .
Next, if we set y = 0, we get

{f (x c) f (x) : x R} = {cx + f (c) 1 : x R} = R

because c 6= 0. This means A A = {y1 y2 : y1 , y2 A} = R.

Now for an arbitrary x R, let y1 , y2 A be such that x = y1 y2 .
Then
f (x) = f (y1 y2 ) = f (y2 ) + y1 y2 + f (y1 ) 1
c + 1 y22 c + 1 y12
= + y1 y2 + 1
2 2 2 2
(y1 y2 )2 x2
= c = c .
2 2
c+1 x2 2
However, for x A, f (x) = 2
2
. So c = 1. Therefore, f (x) = 1 x2
for all x R.

50. (1995 Byelorussian Math Olympiad) Let R be the real numbers. Find
all functions f : R R such that

f (f (x + y)) = f (x + y) + f (x)f (y) xy

for all x, y R.

Solution. (Due to Yung Fai) Clearly, from the equation, f (x) is not
constant. Putting y = 0, we get f (f (x)) = (1 + f (0))f (x). Replacing
x by x + y, we get

(1 + f (0))f (x + y) = f (f (x + y)) = f (x + y) + f (x)f (y) xy,

which simplifies to (*) f (0)f (x + y) = f (x)f (y) xy. Putting y = 1 in

(*), we get f (0)f (x + 1) = f (x)f (1) x. Putting y = 1 and replacing

62
 x by x + 1 in (*), we get f (0)f (x) = f (x + 1)f (1) + x + 1. Eliminating
f (x + 1) in the last two equations, we get

(f 2 (0) f (1)f (1))f (x) = (f (0) f (1))x + f (0).

If f 2 (0) f (1)f (1) = 0, then putting x = 0 in the last equation,

we get f (0) = 0. By (*), f (x)f (y) = xy. Then f (x)f (1) = x for
all x R. So f 2 (0) f (1)f (1) = 1, resulting in a contradiction.
Therefore, f 2 (0) f (1)f (1) 6= 0 and f (x) is a degree 1 polynomial.
Finally, substituting f (x) = ax + b into the original equation, we
find a = 1 and b = 0, i.e. f (x) = x for all x R.

51. (1993 Czechoslovak Math Olympiad) Let Z be the integers. Find all
functions f : Z Z such that

f (1) = f (1) and f (x) + f (y) = f (x + 2xy) + f (y 2xy)

for all integers x, y.

Solution. We have (*) f (1) + f (n) = f (1 + 2n) + f (n) and f (n) +

f (1) = f (n)+f (1+2n). Since f (1) = f (1), this gives f (1+2n) =
f (1 + 2n) for every integer n. So f (k) has the same value for every
odd k. Then equation (*) implies f (n) = f (n) for every integer n. So
we need to find f (n) for nonnegative integers n only.
If we let x = (2k + 1), y = n, then x and x + 2xy are odd. The
functional equation gives f (n) = f (y) = f (y 2xy) = f (n(4k + 3)).
If we let x = n, y = (2k + 1), then similarly, we get f (n) = f (x) =
f (x +2xy) = f (n(4k 1)) = f (n(4k +1)). So f (n) = f (nm) for every
odd m.
For a positive integer n, we can factor n = 2e m, where e, m are
nonnegative integers and m odd. Then f (n) = f (2e ). So any such func-
tion f is determined by the values f (0), f (1), f (2), f (4), f (8), f (16), . . .
(which may be arbitrary). All other values are given by f (n) = f (2e ) as
above. Finally, we check such functions satisfy the equations. Clearly,
f (1) = f (1). If x or y = 0, then the functional equation is clearly
satisfied. If x = 2e m, y = 2d n, where m, n are odd, then

f (x) + f (y) = f (2e ) + f (2d ) = f (x(1 + 2y)) + f (y(1 2x)).

63
52. (1995 South Korean Math Olympiad) Let A be the set of non-negative
integers. Find all functions f : A A satisfying the following two
conditions:
(a) For any m, n A, 2f (m2 + n2 ) = (f (m))2 + (f (n))2 .
(b) For any m, n A with m n, f (m2 ) f (n2 ).

Solution. For m = 0, we get 2f (n2 ) = f (0)2 + f (n)2 . Let m > n, then

f (m)2 f (n)2 = 2(f (m2 ) f (n2 )) 0. So f (m) f (n). This means
f is nondecreasing. Setting m = 0 = n, we get 2f (0) = f (0)2 + f (0)2 ,
which implies f (0) = 0 or 1.
Case f (0) = 1. Then 2f (n2 ) = 1 + f (n)2 . For n = 1, we get f (1) = 1.
k
For m = 1 = n, we get f (2) = 1. Assume f (22 ) = 1. Then for
k k+1 k k+1
n = 22 , we get 2f (22 ) = 1 + f (22 )2 = 2. So f (22 ) = 1. Since
k
lim 22 = and f is nondecreasing, so f (n) = 1 for all n.
k

Case f (0) = 0. Then 2f (n2 ) = f (n)2 . So f (n) is even for all n. For m =
1 = n, we get 2f (2) = f (1)2 +f (1)2 , which implies f (2) = f (1)2 . Using
k k
2f (n2 ) = f (n)2 repeatedly (or by induction), we get 22 1 f (22 ) =
k+1
f (1)2 . Now 2f (1) = f (1)2 implies f (1) = 0 or 2. If f (1) = 0,
k
then lim 22 = and f nondecreasing imply f (n) = 0 for all n. If
k
k k+1
f (1) = 2, then f (22 ) = 22 . Now
f (m + 1)2 = 2f ((m + 1)2 ) = 2f (m2 + 2m + 1)
2f (m2 + 1) = f (m)2 + f (1)2 > f (m)2 .
As f (n) is always even, we get f (m + 1) f (m) + 2. By induction, we
k k k k
get f (n) 2n. Since f (22 ) = 22 +1 = 2 22 for all k, lim 22 =
k
and f is nondecreasing, so f (n) = 2n for all n.
It is easy to check that f (n) = 1, f (n) = 0 and f (n) = 2n are
solutions. Therefore, they are the only solutions.

53. (American Mathematical Monthly, Problem E2176) Let Q denote the

rational numbers. Find all functions f : Q Q such that
 
x+y f (x) + f (y)
f (2) = 2 and f = for x 6= y.
xy f (x) f (y)

64
Solution. We will show f (x) = x is the only solution by a series of
observations.
(1) Setting y = 0, we get f (1) = (f (x) + f (0))/(f (x) f (0)), which
yields (f (1) 1)f (x) = f (0)(1 + f (1)). (Now f is not constant
because the denominator in the equation cannot equal 0.) So,
f (1) = 1 and then f (0) = 0.
(2) Setting y = x, we get 0 = f (x) + f (x), so f (x) = f (x).
(3) Setting y = cx, c 6= 1, x 6= 0, we get
 
f (x) + f (cx) 1+c 1 + f (c)
=f = ,
f (x) f (cx) 1c 1 f (c)

which implies f (cx) = f (c)f (x). Taking c = q, x = p/q, we get

f (p/q) = f (p)/f (q).
(4) Setting y = x 2, we get f (x 1) = (f (x) + f (x 2))/(f (x)
f (x 2)). If f (n 2) = n 2 6= 0 and f (n 1) = n 1, then
this equation implies f (n) = n. Since f (1) = 1 and f (2) = 2, then
f (n) = n for all positive integers by induction and (2), (3) will
imply f (x) = x for all x Q.
Comments. The condition f (2) = 2 can also be deduced from the
functional equation as shown below in (5). If rational numbers are
replaced by real numbers, then again the only solution is still f (x) = x
as shown below in (6) and (7).
(5) We have

f (2) + 1 f (3) + f (2) f (2)2 + 1

f (3) = , f (5) = = .
f (2) 1 f (3) f (2) 1 + 2f (2) f (2)2

Also, f (2)2 = f (4) = (f (5) + f (3))/(f (5) f (3)). Substituting the

equations for f (3) and f (5) in terms of f (2) and simplifying, we
get f (2)2 = 2f (2). (Now f (2) 6= 0, otherwise f ((x + 2)/(x 2)) =
(f (x) + 0)/(f (x) 0) = 1 will force f to be constant.) Therefore,
f (2) = 2.
(6) Note f (x) 6= 0 for x > 0, otherwise f (cx) =0 for c 6= 1 will force f
to be constant. So, if x > 0, then f (x) = f ( x)2 > 0. If x > y 0,
then f (x) f (y) = (f (x) + f (y))/f ((x + y)/(x y)) > 0. This
implies f is strictly increasing for positive real numbers.

65
 (7) For x > 0, if x < f (x), then picking r Q such that x < r < f (x)
will give the contradiction that f (x) < f (r) = r < f (x). Similarly,
f (x) < x will also lead to a contradiction. So, f (x) = x for all x.

54. (Mathematics Magazine, Problem 1552) Find all functions f : R R

such that

f (x + yf (x)) = f (x) + xf (y) for all x, y in R.

Solution. It is easy to check that f (x) = 0 and f (x) = x are solutions.

Suppose f is a solution that is not the zero function. (We will show
f (x) = x for all x.)
Step 1. Setting y = 0, x = 1, we get f (0) = 0. If f (x) = 0, then
0 = xf (y) for all y, which implies x = 0 as f is not the zero function.
So f (x) = 0 if and only if x = 0.
Step 2. Setting x = 1, we get the equation (*) f (1+yf (1)) = f (1)+f (y)
for all y. If f (1) 6= 1, then setting y = 1/(1 f (1)) in (*), we get
f (y) = f (1) + f (y), resulting in f (1) = 0, contradicting step 1. So
f (1) = 1 and (*) becomes f (1 + y) = f (1) + f (y), which implies
f (n) = n for every integer n.
Step 3. For integer n, real z, setting x = n, y = z 1 in the functional
equation, we get

f (nz) = f (n + (z 1)f (n)) = n + nf (z 1) = nf (z).

Step 4. If a = b, then f (a) = f (b) = f (b) implies f (a) + f (b) =

0 = f (a + b). If a 6= b, then a + b 6= 0 and f (a + b) 6= 0 by step 1.
Setting x = (a + b)/2, y = (a b)/(2f ( a+b
2 ), we get

a+b ab a+b
a+b a+b ab

f (a) = f + f ( ) = f ( ) + f ,
2 2f ( a+b
2 )
2 2 2 2f ( a+b
2 )

a+b ba a+b
a+b a+b ba

f (b) = f + f ( ) = f ( ) + f .
2 2f ( a+b
2
) 2 2 2 2f ( a+b
2
)

66
 Adding these, we get f (a) + f (b) = 2f ( a+b
2 ) = f (a + b) by step 3.

Step 5. Applying step 4 to the functional equation, we get the equation

(**) f (yf (x)) = xf (y). Setting y = 1, we get f (f (x) = x. Then f is
bijective. Setting z = f (x) in (**), we get f (yz) = f (y)f (z) for all y, z.
Step 6. Setting z = y in the last equation, we get f (y 2 ) = f (y)2 0.
Setting z = y, we get f (y 2 ) = f (y 2 ) = f (y)2 0. So f (a) > 0
if and only if a > 0.
Step 7. Setting y = 1 in the functional equation, we get f (xf (x)) =
f (x) x. Since x f (x) and f (x) x are of opposite signs, by step 6,
we must have x f (x) = 0 for all x, i.e. f (x) = x for all x.

Maximum/Minimum

55. (1985 Austrian Math Olympiad) For positive integers n, define

f (n) = 1n + 2n1 + 3n2 + + (n 2)3 + (n 1)2 + n.

What is the minimum of f (n + 1)/f (n)?

Solution. For n = 1, 2, 3, 4, 5, 6, f (n + 1)/f (n) = 3, 8/3, 22/8, 65/22,

209/65, 732/209, respectively. The minimum of these is 8/3. For n > 6,
we will show f (n + 1)/f (n) > 3 > 8/3. This follows from

f (n + 1)
>1n+1 + 2n + 3n1 + 4n2 + 5n3 + 6n4 + + (n 1)3 + n2
>1n+1 + 2n + 3n1 + 4n2 + 5n3 + 3(6n5 + + (n 1)2 + n)
=1n+1 + + 5n3 + 3(f (n) 1n 2n1 + 3n2 + 4n3 + 5n4 )
=3f (n) + 2(5n4 1) + 2n1 (2n5 1) > 3f (n).

Therefore, 8/3 is the answer.

56. (1996 Putnam Exam) Given that {x1 , x2 , . . . , xn } = {1, 2, . . . , n}, find
the largest possible value of x1 x2 + x2 x3 + + xn1 xn + xn x1 in terms
of n (with n 2).

67
Solution. Let Mn be the largest such cyclic sum for x1 , x2 , . . . , xn .
In case n = 2, we have M2 = 4. Next suppose Mn is attained by
some permutation of 1, 2, . . . , n. Let x, y be the neighbors of n. Then
removing n from the permutation, we get a permutation of 1, 2, . . . , n
1. The difference of the cyclic sums before and after n is removed is
nx + ny xy = n2 (n x)(n y) n2 2. (Equality holds if and
only if x, y are n 1, n 2.) So Mn (n2 2) Mn1 . Then

2 2 2 2n3 + 3n2 11n + 18

Mn M2 + (3 2) + (4 2) + + (n 2) = .
6
Following the equality case above, we should consider the permutation
constructed as follows: starting with 1, 2, we put 3 between 1 and 2
to get 1, 3, 2, then put 4 between 3 and 2 to get 1, 3, 4, 2, then put 5
between 3 and 4 to get 1, 3, 5, 4, 2 and so on. If n is odd, the permu-
tation is 1, 3, . . . , n 2, n, n 1, . . . , 4, 2. If n is even, the permutation
is 1, 3, . . . , n 1, n, n 2, . . . , 4, 2. The cyclic sum for each of these two
permutations is (2n3 + 3n2 11n + 18)/6 because of the equality case
at each stage. Therefore, Mn = (2n3 + 3n2 11n + 18)/6.

68
 Solutions to Geometry Problems

57. (1995 British Math Olympiad) Triangle ABC has a right angle at C.
The internal bisectors of angles BAC and ABC meet BC and CA
at P and Q respectively. The points M and N are the feet of the
perpendiculars from P and Q to AB. Find angle M CN.

Solution. (Due to Poon Wai Hoi) Using protractor, the angle should
be 45 . To prove this, observe that since P is on the bisector of 6 BAC,
we have P C = P M. Let L be the foot of the perpendicular from C
to AB. Then P M k CL. So 6 P CM = 6 P M C = 6 M CL. Similarly,
6 QCN = 6 N CL. So 6 M CN = 1 6 P CQ = 45 .
2

58. (1988 Leningrad Math Olympiad) Squares ABDE and BCF G are
drawn outside of triangle ABC. Prove that triangle ABC is isosceles if
DG is parallel to AC.

Solution. (Due to Ng Ka Man, Ng Ka Wing, Yung Fai) From B, draw

a perpendicular to AC (and hence also perpendicular to DG.) Let it
intersect AC at X and DG at Y. Since 6 ABX = 90 6 DBY = 6 BDY
and AB = BD, the right triangles ABX and BDY are congruent and
AX = BY. Similarly, the right triangles CBX and BGY are congruent
and BY = CX. So AX = CX, which implies AB = CB.

59. AB is a chord of a circle, which is not a diameter. Chords A1 B1 and

A2 B2 intersect at the midpoint P of AB. Let the tangents to the circle
at A1 and B1 intersect at C1 . Similarly, let the tangents to the circle
at A2 and B2 intersect at C2 . Prove that C1 C2 is parallel to AB.

Solution. (Due to Poon Wai Hoi) Let OC1 intersects A1 B1 at M,

OC2 intersects A2 B2 at N. and OC1 intersect AB at K. Since OC1 is
a perpendicular bisector of A1 B1 , so OM A1 B1 . Similarly, ON
A2 B2 . Then O, N, P, M are concyclic. So 6 ON M = 6 OP M. Since
6 OKP = 90 6 KP M = 6 OP M, we have 6 OKP = 6 ON M. From
the right triangles OA1 C1 and OB2 C2 , we get OM OC1 = OA21 =
OB22 = ON OC2 . By the converse of the intersecting chord theorem,

69
 we get M, N, C1 , C2 are concyclic. So 6 OC1 C2 = 6 ON M = 6 OKP.
Then C1 C2 k KP, that is C1 C2 is parallel to AB.

60. (1991 Hunan Province Math Competition) Two circles with centers O1
and O2 intersect at points A and B. A line through A intersects the
circles with centers O1 and O2 at points Y, Z, respectively. Let the
tangents at Y and Z intersect at X and lines Y O1 and ZO2 intersect
at P. Let the circumcircle of 4O1 O2 B have center at O and intersect
line XB at B and Q. Prove that P Q is a diameter of the circumcircle
of 4O1 O2 B.

Solution. (First we need to show P, O1 , O2 , B are concyclic. Then we

will show 90 = 6 QBP = 6 XBP. Since 6 XY P, 6 P ZX are both 90 ,
it suffices to show X, Y, B, P, Z are concyclic.) Connect O1 A and O2 A.
In 4Y P Z,
6 O1 P Z = 180 (6 O1 Y Z + 6 O2 ZA)

= 180 (6 O1 AY + 6 O2 AZ)
= 6 O1 AO2 = 6 O1 BO2 .
So B, P, O1, O2 are concyclic. Connect BY and BZ. Then

6 Y BZ = 180 (6 AY B + 6 AZB)
1 1
= 180 ( 6 AO1 B + 6 AO2 B)
2 2

= 180 (6 BO1 O2 + 6 BO2 O1 )
= 6 O1 BO2 = 6 O1 P Z = 6 Y P Z.

So Y, Z, P, B are concyclic. Since 6 XY P = 6 XZP = 90 , so the points

Y, X, Z, P, B are concyclic. Then 6 QBP = 6 XBP = 180 6 XZP =
90 . Therefore, P Q is a diameter of the circumcircle of 4O1 O2 B.

61. (1981 Beijing City Math Competition) In a disk with center O, there
are four points such that the distance between every pair of them is
greater than the radius of the disk. Prove that there is a pair of per-
pendicular diameters such that exactly one of the four points lies inside
each of the four quarter disks formed by the diameters.

70
 Solution. (Due to Lee Tak Wing) By the distance condition on the
four points, none of them equals O and no pair of them are on the same
radius. Let us name the points A, B, C, D in the order a rotating radius
encountered them. Since AB > OA, OB, so 6 AOB > 6 OBA, 6 BAO.
Hence 6 AOB > 60 . Similarly, 6 BOC, 6 COD, 6 DOA > 60 . Let
6 AOB be the largest among them, then 60 < 6 AOB < 360 3
60 = 180 . Let EF be the diameter bisecting 6 AOB and with A, B, E
on the same half disk. Now EF and its perpendicular through O
divide the disk into four quarter disks. We have 90 = 30 + 60 <
6 EOB + 6 BOC.

In the case 60 < 6 AOB < 120 , we get 6 EOB + 6 BOC <
60 + 120 = 180 . In the case 120 6 AOB < 180 , we get 6 AOB +
6 BOC < 360 260 = 240 and 6 EOB + 6 BOC < 240 6 AOE
240 120 /2 = 180 . So A, B, C each is on a different quarter disk.
Similarly, 90 < 6 EOD = 6 EOA + 6 AOD < 180 . Therefore, D will
lie on the remaining quarter disk.

62. The lengths of the sides of a quadrilateral are positive integers. The
length of each side divides the sum of the other three lengths. Prove
that two of the sides have the same length.

Solution. (Due to Chao Khek Lun and Leung Wai Ying) Suppose the
sides are a, b, c, d with a < b < c < d. Since d < a + b + c < 3d and
d divides a + b + c, we have a + b + c = 2d. Now each of a, b, c dvides
a + b + c + d = 3d. Let x = 3d/a, y = 3d/b and z = 3d/c. Then
a < b < c < d implies x > y > z > 3. So z 4, y 5, x 6. Then
3d 3d 3d
2d = a + b + c + + < 2d,
6 5 4
a contradiction. Therefore, two of the sides are equal.

63. (1988 Sichuan Province Math Competition) Suppose the lengths of the
three sides of 4ABC are integers and the inradius of the triangle is 1.
Prove that the triangle is a right triangle.

Solution. (Due to Chan Kin Hang) Let a = BC, b = CA, c = AB be

the side lengths, r be the inradius and s = (a + b + c)/2. Since the area

71
 p
of the triangle is rs, we get s(s a)(s b)(s c) = 1 s = s. Then

(s a)(s b)(s c) = s = (s a) + (s b) + (s c).

Now 4(a + b + c) = 8s = (2s 2a)(2s 2b)(2s 2c) = (b + c a)(c +

a b)(a + b c). In (mod 2), each of b + c a, c + a b, a + b c are
the same. So either they are all odd or all even. Since their product is
even, they are all even. Then a + b + c is even and s is an integer.
The positive integers x = s a, y = s b, z = s c satisfy xyz =
x + y + z. Suppose x y z. Then yz 3 for otherwise xyz > 3x
x + y + z. This implies x = 3, y = 2, z = 1, s = 6, a = 3, b = 4, c = 5.
Therefore, the triangle is a right triangle.

Geometric Equations

64. (1985 IMO) A circle has center on the side AB of the cyclic quadri-
lateral ABCD. The other three sides are tangent to the circle. Prove
that AD + BC = AB.

Solution. Let M be on AB such that M B = BC. Then

180 6 ABC 6 CDA
6 CM B = = = 6 CDO.
2 2
This implies C, D, M, O are concyclic. Then
6 DCB 180 6 DAM 6 AM D + 6 ADM
6 AM D = 6 OCD = = = .
2 2 2
So 6 AM D = 6 ADM. Therefore, AM = AD and AB = AM + M B =
AD + BC.

65. (1995 Russian Math Olympiad) Circles S1 and S2 with centers O1 , O2

respectively intersect each other at points A and B. Ray O1 B intersects
S2 at point F and ray O2 B intersects S1 at point E. The line parallel
to EF and passing through B intersects S1 and S2 at points M and
N, respectively. Prove that (B is the incenter of 4EAF and) M N =
AE + AF.

72
 Solution. Since
1
6 EAB = 6 EO1 B = 90 6 O1 BE = 90 6 F BO2 = 6 BAF,
2

AB bisects 6 EAF and 6 O1 BE = 90 6 EAB = 90 12 6 EAF. Now

6 EBA+ 6 F BA = 6 EBA+(180 6 O1 BA) = 180 + 6 O1 BE = 270
16
2
EAF. Then 6 EBF = 90 + 6 EAF, which implies B is the incenter
of 4EAF (because the incenter is the unique point P on the bisector of
6 EAF such that 6 EP F = 90 1 6 EAF ). Then 6 AEB = 6 BEF =
2
6 EBM since EF k M N. So EBAM is an isosceles trapezoid. Hence
EA = M B. Similarly F A = N B. Therefore, M N = M B + N B =
AE + AF.

66. Point C lies on the minor arc AB of the circle centered at O. Suppose
the tangent line at C cuts the perpendiculars to chord AB through A
at E and through B at F. Let D be the intersection of chord AB and
radius OC. Prove that CE CF = AD BD and CD 2 = AE BF.

Solution. (Due to Wong Chun Wai) Note that 6 EAD, 6 ECD, 6 F CD,
6 F BD are right angles. So A, D, C, E are concyclic and B, D, C, F are
concyclic. Then 6 ADE = 6 ACE = 6 ABC = 6 DF C, say the measure
of these angles is . Also, 6 BDF = 6 BCF = 6 BAC = 6 DEC, say
the measure of these angle is . Then

CE CF = (DE cos )(DF cos ) = (DE cos )(DF cos ) = AD BD,

CD 2 = (DE sin )(DF sin ) = (DE sin )(DF sin ) = AE BF.

67. Quadrilaterals ABCP and A0 B 0 C 0 P 0 are inscribed in two concentric

circles. If triangles ABC and A0 B 0 C 0 are equilateral, prove that

P 0 A2 + P 0 B 2 + P 0 C 2 = P A02 + P B 02 + P C 02 .

Solution. Let O be the center of both circles and E be the midpoint

of A0 B 0 . From 4P A0 B 0 with median P E, by cosine law, we get P A02 +
P B 02 = 2(P E 2 + EB 02 ). From 4P C 0 E with cevian P O (note C 0 O =

73
 2OE), by cosine law again, we get P C 02 + 2P E 2 = 3(P O 2 + 2OE)2 .
Putting these together, we get

P A02 + P B 02 + P C 02 = 2(EB 02 + OE 2 ) + 3P O 2 + 4OE 2

= 2B 0 O 2 + 3P O 2 + C 0 O 2
= 3(P O 2 + P 0 O 2 ).

Similarly, P 0 A2 + P 0 B 2 + P C 02 = 3(P O 2 + P 0 O 2 ).
Alternatively, the problem can be solved using complex numbers.
Without loss of generality, let the center be at the origin, A0 be at
rei = r(cos + i sin ) and P be at Rei . Let = e2i/3 . We have

P A02 + P B 02 + P C 02
=|Rei rei |2 + |Rei rei |2 + |Rei rei 2 |2


=3R2 2Re Rrei() (1 + + 2 ) + 3r 2
=3R2 + 3r 2 .

Similarly, P 0 A2 + P 0 B 2 + P 0 C 2 = 3R2 + 3r 2 .

68. Let the inscribed circle of triangle ABC touchs side BC at D, side CA
at E and side AB at F. Let G be the foot of perpendicular from D to
FG BF
EF. Show that = .
EG CE
Solution. (Due to Wong Chun Wai) Let I be the incenter of 4ABC.
Then 6 BDI = 90 = 6 EGD. Also, 6 DEG = 12 6 DIF = 6 DIB.
So 4BDI, 4EGD are similar. Then BD/ID = DG/EG. Likewise,
4CDI, 4F GD are similar and CD/ID = DG/F G. Therefore,

FG DG/EG BD/ID BD BF
= = = = .
EG DG/F G CD/ID CD CE

69. (1998 IMO shortlisted problem) Let ABCDEF be a convex hexagon

such that
AB CD EF
6 B + 6 D + 6 F = 360 and = 1.
BC DE F A
74
Prove that
BC AE F D
= 1.
CA EF DB

Solution. Let P be such that 6 F EA = 6 DEP and 6 EF A = 6 EDP,

where P is on the opposite side of lines DE and CD as A. Then
4F EA, 4DEP are similar. So

FA DP EF EA
= and () = .
EF PE ED EP
Since 6 B + 6 D + 6 F = 360 , we get 6 ABC = 6 P DC. Also,

AB DE F A DP
= = .
BC CD EF CD
Then 4ABC, 4P DC are similar. Consequently, we get 6 BCA =
6 DCP and (**) CB/CD = CA/CP. Since 6 F ED = 6 AEP, by (*),
4F ED, 4AEP are similar. Also, since 6 BCD = 6 ACP, by (**),
4BCD, 4ACP are similar. So AE/EF = P A/F D and BC/CA =
DB/P A. Multiplying these and moving all factors to the left side, we
get the desired equation.
Using complex numbers, we can get an algebraic solution. Let
a, b, c, d, e, f denote the complex numbers corresponding to A, B, C, D,
E, F, respectively. (The origin may be taken anywhere on the plane.)
Since ABCDEF is convex, 6 B, 6 D and 6 F are the arguments of the
complex numbers (a b)/(c b), (c d)/(e d) and (e f )/(a f ),
respectively. Then the condition 6 B + 6 D + 6 F = 360 implies that
the product of these three complex numbers is a positive real number.
It is equal to the product of their absolute values AB/BC, CD/DE
and EF/F A. Since (AB/BC)(CD/DE)(EF/F A) = 1, we have

ab cd ef
= 1.
cb ed af

So
0 = (a b)(c d)(e f ) (c b)(e d)(a f )
= (b c)(a e)(f d) (a c)(f e)(b d).

75
 Then  
BC AE F D  b c a e f d
=   = 1.
CA EF DB ac f e bd

Similar Triangles

70. (1984 British Math Olympiad) P, Q, and R are arbitrary points on the
sides BC, CA, and AB respectively of triangle ABC. Prove that the
three circumcentres of triangles AQR, BRP, and CP Q form a triangle
similar to triangle ABC.

Solution. Let the circumcenters of triangles AQR, BRP and CP Q be

A0 , B 0 and C 0 , respectively. A good drawing suggests the circles pass
through a common point! To prove this, let circumcircles of triangles
AQR and BRP intersect at R and X. Then 6 QXR = 180 6 CAB =
6 ABC + 6 BCA and 6 RXP = 180 6 ABC = 6 CAB + 6 BCA. So
6 P XQ = 360 6 QXR 6 RXP = 180 6 BCA, which implies X
is on the circumcircle of triangle CP Q. Now

6 C 0 A0 B 0 = 6 C 0 A0 X + 6 XA0B 0
1 1
= 6 QA0 X + 6 RA0 X
2 2
1
= 6 QA0 R = 6 CAB.
2
Similarly, 6 A0 B 0 C 0 = 6 ABC and 6 B 0 C 0 A0 = 6 BCA. So, triangles
A0 B 0 C 0 and ABC are similar.

71. Hexagon ABCDEF is inscribed in a circle so that AB = CD = EF.

Let P, Q, R be the points of intersection of AC and BD, CE and DF,
EA and F B respectively. Prove that triangles P QR and BDF are
similar.

Solution. (Due to Ng Ka Wing) Let O be the center of the circle

and let L, M, N be the projections of O on BD, DF, F B, respectively.
Then L, M, N are midpoints of BD, DF, F B, respectively. Let S be the
projection of O on AE. Since AB = EF, we get F B = AE and hence

76
 ON = OS. Let 6 AOB = 6 COD = 6 EOF = 2. Then 6 RON =
16 16 1 6
2 SON = 2 ARB = 4 ( AOB + EOF ) = . Hence, ON/OR =
6
cos . Similarly, 6 P OL = 6 QOM = and OL/OP = OM/OQ =
cos .
Next rotate 4P QR around O at angle so that the image Q0
of Q lies on the line OM, the image R0 of R lies on the line ON
and the image P 0 of P lies on line OL. Then ON/OR0 = OL/OP 0 =
OM/OQ0 = cos . So 4P 0 Q0 R0 , 4LM N are similar. Since L, M, N
are midpoints of BD, DF, F B, respectively, we have 4LM N, 4BDF
are similar. Therefore, 4P QR, 4BDF are similar.

72. (1998 IMO shortlisted problem) Let ABCD be a cyclic quadrilateral.

Let E and F be variable points on the sides AB and CD, respectively,
such that AE : EB = CF : F D. Let P be the point on the segment
EF such that P E : P F = AB : CD. Prove that the ratio between the
areas of triangles AP D and BP C does not depend on the choice of E
and F.

Solution. Let [U V W ] denote the area of 4U V W and let d(X, Y Z)

denote the distance from X to line Y Z. We have AE : EB = CF :
F D = a : b, where a + b = 1. Since P E : P F = AB : CD, we have

CD AB
d(P, AD) = d(E, AD) + d(F, AD),
AB + CD AB + CD

CD AB
[AP D] = [AED] + [AF D]
AB + CD AB + CD
a CD b AB
= [ABD] + [ACD],
AB + CD AB + CD
CD AB
d(P, BC) = d(E, BC) + d(F, BC),
AB + CD AB + CD
CD AB
[BP C] = [BEC] + [BF C]
AB + CD AB + CD
b CD a AB
= [BAC] + [BDC].
AB + CD AB + CD

77
 Since A, B, C, D are concyclic, sin 6 BAD = sin 6 BCD and sin 6 ABC=
sin 6 ADC, So,

[AP D] a CD [ABD] + b AB [ACD]

=
[BP C] b CD [BAC] + a AB [BDC]
a CD AB AD sin 6 BAD + b AB CD AD sin 6 ADC
=
b CD AB BC sin 6 ABC + a AB CD BC sin 6 BCD
AD a sin 6 BAD + b sin 6 ADC AD
= = .
BC b sin 6 ABC + a sin 6 BCD BC

Tangent Lines

73. Two circles intersect at points A and B. An arbitrary line through B

intersects the first circle again at C and the second circle again at D.
The tangents to the first circle at C and to the second circle at D
intersect at M. The parallel to CM which passes through the point
of intersection of AM and CD intersects AC at K. Prove that BK is
tangent to the second circle.

Solution. Let L be the intersection of AM and CD. Since

6 CM D + 6 CAD = 6 CM D + 6 CAB + 6 DAB
= 6 CM D + 6 BCM + 6 BDM = 180 ,

so A, C, M, D are concyclic. Since LK k M C,

6 LKC = 180 6 KCM = 180 6 KCL 6 LCM

= 180 6 ACB 6 CAB = 6 CBA = 6 LBA.

So A, B, L, K are concyclic. Then

6 KBA = 6 KLA = 6 CM A = 6 CDA = 6 BDA.

Therefore, BK is tangent to the circle passing through A, B, D.

74. (1999 IMO) Two circles 1 and 2 are contained inside the circle ,
and are tangent to at the distinct points M and N, respectively.

78
 1 passes through the center of 2 . The line passing through the two
points of intersection of 1 and 2 meets at A and B, respectively.
The lines M A and M B meets 1 at C and D, respectively. Prove that
CD is tangent to 2 .

Solution. (Due to Wong Chun Wai) Let X, Y be the centers of 1 , 2 ,

respectively. Extend Y X to meet 2 at Q. Join AN to meet 2 at E.
Since AB is the radical axis of 1 , 2 , so AC AM = AE AN. This
implies C, M, N, E are concyclic. Let U be the intersection of line CE
with the tangent to 1 at M. Then 6 U CM = 6 EN M = 6 AN M =
6 U M C. So CE is tangent to 1 . Similarly, CE is tangent to 2 . Now
Y E = Y Q and
1
6 CY E = 90 6 ECY = 90 6 CXY
2
= 90 (90 6 CY Q) = 6 CY Q.

These imply 4CY E, 4CY Q are congruent. Hence 6 CQY = 6 CEY =

90 . Similarly 6 DQY = 90 . Therefore, CD is tangent to 2 .

75. (Proposed by India for 1992 IMO) Circles G1 and G2 touch each other
externally at a point W and are inscribed in a circle G. A, B, C are
points on G such that A, G1 and G2 are on the same side of chord BC,
which is also tangent to G1 and G2 . Suppose AW is also tangent to
G1 and G2 . Prove that W is the incenter of triangle ABC.

Solution. Let P and Q be the points of tangency of G1 with BC and

arc BAC, respectively. Let D be the midpoint of the complementary
arc BC of G (not containing A) and L be a point on G1 so that DL is
tangent to G1 and intersects segment P C. Considering the homothety
with center Q that maps G1 onto G, we see that Q, P, D are collinear
because the tangent at P (namely BC) and the tangent at D are
parallel. Since 6 BQD, 6 CBD subtend equal arcs, 4BQD, 4P BD
are similar. Hence DB/DP = DQ/DB. By the intersecting chord
theorem, DB 2 = DP DQ = DL2 . So DL = DB = DC. Then D has
the same power DB 2 = DC 2 with respect to G1 and G2 . Hence D is
on the radical axis AW of G1 and G2 . So L = W and DW is tangent
to G1 and G2 .

79
 Since D is the midpoint of arc BC, so AW bisects 6 BAC. Also,
6 ABW = 6 BW D 6 BAD = 6 W BD 6 CBD = 6 CBW

and BW bisects 6 ABC. Therefore W is the incenter of 4ABC.

Comments. The first part of the proof can also be done by inversion
with respect to the circle centered at D and of radius DB = DC.
It maps arc BC onto the chord BC. Both G1 and G2 are invariant
because the power of D with respect to them is DB 2 = DC 2 . Hence
W is fixed and so DW is tangent to both G1 and G2 .

Locus

76. Perpendiculars from a point P on the circumcircle of 4ABC are drawn

to lines AB, BC with feet at D, E, respectively. Find the locus of the
circumcenter of 4P DE as P moves around the circle.

Solution. Since 6 P DB + 6 P EB = 180 , P, D, B, E are concyclic.

Hence, the circumcircle of 4P DE passes through B always. Then P B
is a diameter and the circumcenter of 4P DE is at the midpoint M of
P B. Let O be the circumcenter of 4ABC, then OM P B. It follows
that the locus of M is the circle with OB as a diameter.

77. Suppose A is a point inside a given circle and is different from the
center. Consider all chords (excluding the diameter) passing through
A. What is the locus of the intersection of the tangent lines at the
endpoints of these chords?

Solution. (Due to Wong Him Ting) Let O be the center and and r
be the radius. Let A0 be the point on OA extended beyond A such
that OA OA0 = r 2 . Suppose BC is one such chord passing through
A and the tangents at B and C intersect at D 0 . By symmetry, D 0 is
on the line OD, where D is the midpoint of BC. Since 6 OBD 0 = 90 ,
OD OD 0 = OB 2 (= OA OA0 .) So 4OAD is similar to 4OD 0 A0 .
Since 6 ODA = 90 , D 0 is on the line L perpendicular to OA at A0 .
Conversely, for D 0 on L, let the chord through A perpendicular
to OD 0 intersect the circle at B and C. Let D be the intersection of

80
 the chord with OD 0 . Now 4OAD, 4OD 0A0 are similar right triangles.
So OD OD 0 = OA OA0 = OB 2 = OC 2 , which implies 6 OBD 0 =
6 OCD 0 = 90 . Therefore, D 0 is on the locus. This shows the locus is
the line L.

78. Given 4ABC. Let line EF bisects 6 BAC and AE AF = AB AC.

Find the locus of the intersection P of lines BE and CF.

Solution. For such a point P, since AB/AE = AF/AC and 6 BAE =

6 F AC, so 4BAE, 4F AC are similar. Then 6 AEP = 6 P CA. So
A, E, C, P are concyclic. Hence 6 BP C = 6 CAE = 6 BAC/2. There-
fore, P is on the circle C whose points X satisfy 6 BXC = 6 BAC/2
and whose center is on the same side of line BC as A.
Conversely, for P on C, Let BP, CP intersect the angle bisector
of BAC at E, F, respectively. Since 6 BP C = 6 BCA/2, so 6 EP F =
6
6 EAC. Hence A, E, C, P are concyclic. So 6 BEA = 6 F CA. Also
6 BAE = 6 F AC. So 4BAE, 4F AC are similar. Then AB AC =
AE AF. Therefore, the locus of P is the circle C.

79. (1996 Putnam Exam) Let C1 and C2 be circles whose centers are 10
units apart, and whose radii are 1 and 3. Find the locus of all points
M for which there exists points X on C1 and Y on C2 such that M is
the midpoint of the line segment XY.

Solution. (Due to Poon Wai Hoi) Let O1 , O2 be the centers of C1 , C2 ,

respectively. If we fix Y on C2 , then as X moves around C1 , M will
trace a circle Y with radius 12 centered at the midpoint mY of O1 Y.
As Y moves around C2 , mY will trace a circle of radius 32 centered at
the midpont P of O1 O2 . So the locus is the solid annulus centered at
P with inner radius 32 12 = 1 and outer radius 32 + 12 = 2.

Collinear or Concyclic Points

80. (1982 IMO) Diagonals AC and CE of the regular hexagon ABCDEF

are divided by the inner points M and N, respectively, so that
AM CN
= = r.
AC CE
81
 Determine r if B, M and N are collinear.

Solution. (Due to Lee Tak Wing) Let AC = x, then BC = x/ 3,
CN = xr, CM = x(1 r). Let [XY Z] denote the area of 4XY Z.
Since 6 N CM = 60 , 6 BCM = 30 and [BCM ] + [CM N ] = [BCN ],
so
x2 (1 r) sin 30 x2 r(1 r) sin 60 x2 r
+ = .
2 3 2 2 3
1
Cancelling x2 and solving for r, we get r = .
3

81. (1965 Putnam Exam) If A, B, C, D are four distinct points such that
every circle through A and B intersects or coincides with every circle
through C and D, prove that the four points are either collinear or
concyclic.

Solution. Suppose A, B, C, D are neither concyclic nor collinear. Then

the perpendicular bisector p of AB cannot coincide with the perpen-
dicular bisector q of CD. If lines p and q intersect, their common point
is the center of two concentric circles, one through A and B, the other
through C and D, a contradiction. If lines p and q are parallel, then
lines AB and CD are also parallel. Consider points P and Q on p and
q, respectively, midway between the parallel lines AB and CD. Clearly,
the circles through A, B, P and C, D, Q have no common point, again
a contradiction.

82. (1957 Putnam Exam) Given an infinite number of points in a plane,

prove that if all the distances between every pair are integers, then the
points are collinear.

Solution. Suppose there are three noncollinear points A, B, C such

that AB = r and AC = s. Observe that if P is one of the other points,
then bt the triangle inequality, |P A P B| = 0, 1, 2, . . . , r. Hence P
would be on the line H0 joining A, B or on one of the hyperbolas
Hi = {X : |XAXB| = i} for i = 1, 2, . . . , r1 or on the perpendicular
bisector Hr of AB. Similarly, |P A P C| = 0, 1, 2, . . . , s. So P is on
one of the sets Kj = {X : |XA XC| = j} for j = 0, 1, . . . , s. Since

82
 lines AB and AC are distinct, every intersection Hi Kj is only a
finite set. So there can only be finitely many points that are integral
distances from A, B, C, a contradiction. Therefore, the given points
must be collinear.

83. (1995 IMO shortlisted problem) The incircle of triangle ABC touches
BC, CA and AB at D, E and F respectively. X is a point inside
triangle ABC such that the incircle of triangle XBC touches BC at
D also, and touches CX and XB at Y and Z respectively. Prove that
EF ZY is a cyclic quadrilateral.

Solution. If EF k BC, then AB = AC and AD is an axis of symmetry

of EF ZY. Hence EF ZY is a cyclic quadrilateral. If lines EF and BC
intersect at P, then by Menelaus theorem, (AF BP CE)/(F B P C
EA) = 1. Since BZ = BD = BF, CY = CD = CE and AF/EA = 1 =
XZ/Y X, we get (XZ BP CY )/(ZB P C Y X) = 1. By the converse
of the Menelaus theorem, Z, Y, P are collinear. By the intersecting
chord theorem, P E P F = P D 2 = P Y P Z. Hence EF ZY is a cyclic
quadrilateral by the converse of the intersecting chord theorem.

84. (1998 IMO) In the convex quadrilateral ABCD, the diagonals AC and
BD are perpendicular and the opposite sides AB and DC are not
parallel. Suppose the point P, where the perpendicular bisectors of
AB and DC meet, is inside ABCD. Prove that ABCD is a cyclic
quadrilateral if and only if the triangles ABP and CDP have equal
areas.

Solution. (Due to Leung Wing Chung) Set the origin at P. Suppose

A and C are on the line y = p and B and D are on the line x = q. Let
AP = BP = r, CP = DP = s. Then the coordinates of A, B, C, D are
p p p p
( r 2 p2 , p), (q, r 2 q 2 ), ( s2 p2 , p), (q, s2 q 2 ),

respectively. Now 4ABP, 4CDP have equal areas if and only if

 p  p 
1  r 2 p2 p  1  s2 p2
=  p 
,
p p
2 q r2 q2  2  q s2 q 2 

83
 p p p p
i.e. ( r 2p
p2 r 2p
q 2 pq)/2 = ( s2 p2 s2 q 2 pq)/2. Since
f (x) = ( x2 p2 x2 q 2 pq)/2 is strictly decreasing when x |p|
and |q|, the determinants are equal if and only if r = s, which is
equivalent to ABCD cyclic.

85. (1970 Putnam Exam) Show that if a convex quadrilateral with side-
lengths a, b, c, d and area abcd has an inscribed circle, then it is a
cyclic quadrilateral.

Solution. Since the quadrilateral has an inscribed circle, we have a +

c = b + d. Let k be the length of a diagonal and angles and selected
so that
k 2 = a2 + b2 2ab cos = c2 + d2 2cd cos .

If we subtract (a b)2 = (c d)2 and divide by 2, we get the equation

(*)
ab(1 cos ) = cd(1 cos ). From the area (ab sin +cd sin )/2 =
abcd, we get

4abcd = a2 b2 (1 cos2 ) + c2 d2 (1 cos2 ) + 2abcd sin sin .

Using (*), we can cancel abcd to obtain the equation

4 = (1 + cos )(1 cos ) + (1 + cos )(1 cos ) + 2 sin sin

= 2 2 cos( + ),

which implies + = 180 . Therefore, the quadrilateral is cyclic.

Concurrent Lines

86. In 4ABC, suppose AB > AC. Let P and Q be the feet of the per-
pendiculars from B and C to the angle bisector of 6 BAC, respectively.
Let D be on line BC such that DA AP. Prove that lines BQ, P C
and AD are concurrent.

Solution. Let M be the intersection of P C and AD. Let B 0 be the

mirror image of B with respect to line AP. Since BB 0 AP and

84
 AD AP, so BB 0 k AD. Then 4BCB 0 , 4DAC are similar. Since P
is the midpoint of BB 0 , so P C intersects AD at its midpoint M. Now

AQ MC AM AM
= = 0 = .
PQ PC BP BP

4BP Q, 4M AQ are similar. This implies 6 BQP = 6 M QA. So line

BQ passes through M, too.

87. (1990 Chinese National Math Competition) Diagonals AC and BD

of a cyclic quadrilateral ABCD meets at P. Let the circumcenters of
ABCD, ABP, BCP, CDP and DAP be O, O1, O2 , O3 and O4 , respec-
tively. Prove that OP, O1 O3 , O2 O4 are concurrent.

Solution. Let line P O2 intersect the circumcircle of 4BCP and seg-

ment AD at points Q and R, respectively. Now 6 P DR = 6 BCA =
6 P QB and 6 DP R = 6 QP B. So 6 DRP = 6 QBP = 90 and P O2
AD. Next circumcircles of ABCD and DAP share the common chord
AD, so OO4 AD. Hence P O2 and OO4 are parallel. Similarly,
P O4 and OO2 are parallel. So P O2 OO4 is a parallelogram and diag-
onal O2 O4 passes through the midpoint G of OP. Similarly, P O1 OO3
is a parallelogram and diagonal O1 O3 passes through G. Therefore,
OP, O1O3 , O2 O4 concur at G.

88. (1995 IMO) Let A, B, C and D be four distinct points on a line, in that
order. The circles with diameters AC and BD intersect at the points
X and Y. The line XY meets BC at the point Z. Let P be a point on
the line XY different from Z. The line CP intersects the circle with
diameter AC at the points C and M, and the line BP intersects the
circle with diameter BD at the points B and N. Prove that the lines
AM, DN and XY are concurrent.
Solution 1. (Due to Yu Chun Ling) Let AR be parallel to BP and
DR0 be parallel to CP, where R and R0 are points on line XY. Since
BZ ZD = XZ ZY = CZ ZA, we get BZ/AZ = CZ/DZ. Since
4CZP is similar to 4DZR0 and 4BZP is similar to 4AZR, so

ZP BZ CZ ZP
= = = .
ZR AZ DZ ZR0
85
 Hence R and R0 must coincide. Therefore, 4BP C is similar to 4ARD.
Since XY AD, AM CM, CM k DR, DN BN and
BN k AR, the lines AM, DN, XY are the extensions of the altitudes
of 4ARD, hence they must be concurrent.
Solution 2. (Due to Mok Tze Tao) Set the origin at Z and the x-
axis on line AD. Let the coordinates of the circumcenters of triangles
AM C and BN D be (x1 , 0) and (x2 , 0), and the circumradii be r1 and
r2 , respectively. Then the coordinates of A and C are (x1 r1 , 0) and
(x1 + r1 , 0), respectively. Let the coordinates of P be (0, y0 ). Since
y0
AM CP and the slope of CP is , the equation of AM
x1 + r1
works out to be (x1 + r1 )x y0 y = x21 r12 . Let Q be the intersection of
r12 x21
AM with XY, then Q has coordinates (0, ). Similarly, let Q0 be
y0
0 r22 x22
the intersection of DN with XY, then Q has coordinates (0, ).
y0
Since r12 x21 = ZX 2 = r22 x22 , so Q = Q0 .
Solution 3. Let AM intersect XY at Q and DN intersect XY at
Q0 . Observe that the right triangles AZQ, AM C, P ZC are similar, so
AZ/QZ = P Z/CZ. Then QZ = AZ CZ/P Z = XZ Y Z/P Z. Simi-
larly, Q0 Z = XZ Y Z/P Z. Therefore Q = Q0 .

89. AD, BE, CF are the altitudes of 4ABC. If P, Q, R are the midpoints
of DE, EF, F D, respectively, then show that the perpendicular from
P, Q, R to AB, BC, CA, respectively, are concurrent.

Solution. Let Y be the foot of the perpendicular from Q to BC and

H be the orthocenter of 4ABC. Note that 6 ACF = 90 6 CAB =
6 ABE. Since 6 CEH = 90 = 6 CDH, C, D, H, E are concyclic. So
6 ACF = 6 EDH. Now AH k QY, ED k QR, so 6 EDH = 6 RQY.
Hence, 6 RQY = 6 EDH = 6 ACF. Similarly, 6 ABE = 6 F DH =
6 P QY. Next, since 6 ACF = 6 ABE, QY bisects 6 P QR. From these,
it follows the perpendiculars from P, Q, R to AB, BC, CA concur at
the incenter of 4P QR.

90. (1988 Chinese Math Olympiad Training Test) ABCDEF is a hexagon

86
 inscribed in a circle. Show that the diagonals AD, BE, CF are concur-
rent if and only if AB CD EF = BC DE F A.

Solution. (Due to Yu Ka Chun) Suppose AD, BE, CF concurs at X.

From similar triangles ABX and EDX, we get AB/DE = BX/DX.
Similarly, CD/F A = DX/F X and EF/BC = F X/BX. Multiplying
thse, we get (AB CD EF )/(DE F A BC) = 1, so AB CD EF =
BC DE F A.
For the converse, we use the so-called method of false position.
Suppose (*) AB CD EF = BC DE F A and AD intersect BE
at X. Now let CX meet the circle again at F 0 . By the first part, we
get AB CD EF 0 = BC DE F 0 A. Dividing this by (*), we have
EF 0 /EF = F 0 A/F A. If F 0 is on open arc AF, then F 0 A < F A and
EF < EF 0 yielding F 0 A/F A < 1 < EF 0 /EF, a contradiction. If
F 0 is on the open arc EF, then F A < F 0 A and EF 0 < EF yielding
EF 0 /EF < 1 < F 0 A/F A, a contradiction. So F 0 = F.
Alternatively, we can use Cevas theorem and its converse. Let
AC and BE meet at G, CE and AD meet at H, EA and CF meet at
I. Let h, k be the distances from A, C to BE, respectively. Then

AG h AB sin 6 ABG
= = .
CG k BC sin 6 CBG

Similarly,

CH CD sin 6 CDH EI EF sin 6 EF I

= and = .
EH DE sin 6 EDH AI F A sin 6 AF I

Now 6 ABG = 6 EDH, 6 CBG = 6 EF I, 6 CDH = 6 AF I. By Cevas

theorem and its converse, AD, BE, CF are concurrent if and only if

AG CH EI AB CD EF
1= = .
CG EH AI BC DE F A

91. A circle intersects a triangle ABC at six points A1 , A2 , B1 , B2 , C1 , C2 ,

where the order of appearance along the triangle is A, C1 , C2 , B, A1 , A2 ,

87
 C, B1 , B2 , A. Suppose B1 C1 , B2 C2 meets at X, C1 A1 , C2 A2 meets at
Y and A1 B1 , A2 B2 meets at Z. Show that AX, BY, CZ are concurrent.

Solution. Let D be the intersection of AX and B1 C2 . Since AX, B1 C1 ,

B2 C2 are concurrent, by (the trigonometric form of) Cevas theorem,
DC2 B1 B2 AC1 sin C2 AD sin B2 B1 C1 sin B1 C2 B2
1= = .
DB1 AB2 C1 C2 sin DAB1 sin C1 B1 C2 sin B2 C2 C1
sin BAX sin C2 AD sin C1 B1 C2 sin B2 C2 C1
Then = = . Similarly,
sin XAC sin DAB1 sin B2 B1 C1 sin B1 C2 B2
sin CBY sin A1 C1 A2 sin C2 A2 A1
= ,
sin Y BA sin C2 C1 A1 sin C1 A2 C2
sin ACZ sin B1 A1 B2 sin A2 B2 A1
= .
sin ZCB sin A2 A1 B1 sin A1 B2 A2
Using 6 C1 B1 C2 = 6 C1 A2 C2 and similar angle equality, we see that
the product of the three equations involving X, Y, Z above is equal to
1. By the converse of the trigonometric form of Cevas theorem, we see
that AX, BY, CZ are concurrent.

92. (1995 IMO shortlisted problem) A circle passing through vertices B

and C of triangle ABC intersects sides AB and AC at C 0 and B 0 ,
respectively. Prove that BB 0 , CC 0 and HH 0 are concurrent, where H
and H 0 are the orthocenters of triangles ABC and AB 0 C 0 , respectively.

Solution. (Due to Lam Po Leung) Let d(X, L) denote the distance

from a point X to a line L. For the problem, we will use the following
lemma.
Lemma. Let lines L1 , L2 intersect at P (forming four angles with ver-
tex P ). Suppose H, H 0 lie on an opposite pair of these angles. If
d(H, L1)/d(H 0 , L1 ) = d(H, L2 )/d(H 0 , L2 ), then H, P, H 0 are collinear.
Proof. Let HH 0 intersect L1 , L2 at X, Y, respectively. Then
HH 0 HX d(H, L1 )
= + 1 = +1
H 0X H 0X d(H 0 , L1 )
d(H, L2 ) HY HH 0
= +1= 0 +1= 0 .
d(H 0 , L2 ) HY HY

88
 So X = Y is on both L1 and L2 , hence it is P. Therefore, H, P, H 0 are
collinear.
For the problem, let BB 0 , CC 0 intersect at P. Since 6 ABH =
90 6 A = 6 AC 0 H 0 , so BH k C 0 H 0 . Similarly, CH k B 0 H 0 . Let
BH, CC 0 intersect at L and CH, BB 0 intersect at K. Now

6 P BH = 6 ABH 6 C 0 BP = (90 6 A) 6 B 0 CP
= 6 ACH 6 B 0 CP = 6 P CH.

So, K, B, C, L are concyclic. Then 4LHK, 4BHC are similar. Also,

4BHC, 4B 0 H 0 C 0 are similar because

6 CBH = 90 6 ACB = 90 6 AC 0 B 0 = 6 C 0 B 0 H 0

and similarly 6 BCH = 6 B 0 C 0 H 0 . Therefore 4LHK, 4B 0 H 0 C 0 are

similar. So KH/B 0 H 0 = LH/C 0 H 0 . Since BH k C 0 H 0 and CH k B 0 H 0 ,
so d(H, BB 0 )/d(H 0 , BB 0 ) = d(H, CC 0 )/d(H 0 , CC 0 ). By the lemma,
HH 0 also passes through P.

Perpendicular Lines

93. (1998 APMO) Let ABC be a triangle and D the foot of the altitude
from A. Let E and F be on a line passing through D such that AE
is perpendicular to BE, AF is perpendicular to CF, and E and F are
different from D. Let M and N be the midpoints of the line segments
BC and EF, respectively. Prove that AN is perpendicular to N M.

Solution. (Due to Cheung Pok Man) There are many different pic-
tures, so it is better to use coordinate geometry to cover all cases.
Set A at the origin and let y = b 6= 0 be the equation of the line
through D, E, F. (Note the case b = 0 implies D = E = F, which
is not allowed.) Let the coordinates of D, E, F be (d, b), (e, b), (f, b),
respectively. Since BE AE and slope of AE is b/e, so the equa-
tion of line AE is ex + by (b2 + e2 ) = 0. Similarly, the equation of
line CF is f x + by (b2 + f 2 ) = 0 and the equation of line BC is
dx + by (b2 + d2 ) = 0.

89
 From these, we found the coordinates of B, C are (d+e, b de b ), (d+
df e+f
f, b b ), respectively. Then the coordinates of M, N are (d + 2 , b
de+df
2b
), ( e+f
2
, b), respectively. So the slope of AN is 2b/(e + f ) and the
slope of M N is ( de+df e+f
2b )/d = 2b . The product of these slopes is
1. Therefore, AN M N.

94. (2000 APMO) Let ABC be a triangle. Let M and N be the points
in which the median and the angle bisector, respectively, at A meet
the side BC. Let Q and P be the points in which the perpendicular at
N to N A meets M A and BA, respectively, and O the point in which
the perpendicular at P to BA meets AN produced. Prove that QO is
perpendicular to BC.

Solution 1. (Due to Wong Chun Wai) Set the origin at N and the
x-axis on line N O. Let the equation of line AB be y = ax + b, then
1
the equation of lines AC and P O are y = ax b and y = x + b,
a
respectively. Let the equation of BC be y = cx. Then B has co-
b bc b bc
ordinates ( , ), C has coordinates ( , ), M has
ca ca c+a c+a
ab abc b
coordinates ( 2 2
, 2 2
), A has coordinates ( , 0), O has co-
c a c a a
ab
ordinates (ab, 0) and Q has coordinates (0, ). Then BC has slope c
c
1
and QO has slope . Therefore, QO BC.
c
Solution 2. (Due to Poon Wai Hoi) The case AB = AC is clear.
Without loss of generality, we may assume AB > AC. Let AN intersect
the circumcircle of 4ABC at D. Then

1
6 DBC = 6 DAC = 6 BAC = 6 DAB = 6 DCB.
2
So DB = DC and M D BC.
With A as the center of homothety, shrink D to O, B to B 0 and C
to C 0 . Then 6 OB 0 C 0 = 12 6 BAC = 6 OC 0 B 0 and BC k B 0 C 0 . Let B 0 C 0
cut P N at K. Then 6 OB 0 K = 6 DAB = 6 OP K. So P, B 0 , O, K are
concyclic. Hence 6 B 0 KO = 6 B 0 P O = 90 and so B 0 K = C 0 K. Since

90
 BC k B 0 C 0 , this implies A, K, M are collinear. Therefore, K = Q.
Since 6 B 0 KO = 90 and BC k B 0 C 0 , we get QO BC.

95. Let BB 0 and CC 0 be altitudes of triangle ABC. Assume that AB 6=

AC. Let M be the midpoint of BC, H the orthocenter of ABC and D
the intersection of B 0 C 0 and BC. Prove that DH AM.

Solution. Let A0 be the foot of the altitude from A to BC. Since

A0 , B 0 , C 0 M lie on the nine-point circle of 4ABC, so by the intersecting
chord theorem, DB 0 DC 0 = DA0 DM. Since 6 AC 0 H = 90 = 6 AB 0 H,
points A, C 0 , H, B 0 lie on a circle 1 with the midpoint X of AH as
center. Since 6 HA0 M = 90 , so the circle 2 through H, A0, M has
the midpoint Y of HM as center. Since DB 0 DC 0 = DA0 DM, the
powers of D with respect to 1 and 2 are the same. So D (and H)
are on the radical axis of 1 , 2 . Then DH XY. By the midpoint
theorem, XY k AM. Therefore, DH AM.

96. (1996 Chinese Team Selection Test) The semicircle with side BC of
4ABC as diameter intersects sides AB, AC at points D, E, respec-
tively. Let F, G be the feet of the perpendiculars from D, E to side
BC respectively. Let M be the intersection of DG and EF. Prove that
AM BC.

Solution. Let H be the foot of the perpendicular from A to BC. Now

6 BDC = 90 = 6 BEC. So DF = BD sin B = BC cos B sin B and
similarly EG = BC cos C sin C. Now
GM EG cos C sin C cos C AB
= = = .
MD FD cos B sin B cos B AC
Since BH = AB cos B, HG = AE cos C, we get
BH AB cos B AC cos B BH GM DA
= = and = 1.
HG AE cos C AD cos C HG M D AB
By the converse of Menelaus theorem on 4BDG, points A, M, H are
collinear. Therefore, AM BC.

97. (1985 IMO) A circle with center O passes through the vertices A and
C of triangle ABC and intersects the segments AB and AC again at

91
 distinct points K and N, respectively. The circumcircles of triangles
ABC and KBN intersect at exactly two distinct points B and M.
Prove that OM M B.

Solution. Let CM intersect the circle with center O at a point L.

Since 6 BM C = 180 6 BAC = 180 6 KAC = 6 KLC, so BM is
parallel to KL. Now
6 LKM = 6 LKN + 6 N KM = 6 LCN + 6 N BM
= 180 6 BM C = 6 BAC = 6 KLM.

Then KM = LM. Since KO = LO, so OM KL. Hence OM BM.

98. (1997 Chinese Senoir High Math Competition) A circle with center O
is internally tangent to two circles inside it at points S and T. Suppose
the two circles inside intersect at M and N with N closer to ST. Show
that OM M N if and only if S, N, T are collinear.

Solution. (Due to Leung Wai Ying) Consider the tangent lines at S

and at T. (Suppose they are parallel, then S, O, T will be collinear so
that M and N will be equidistant from ST, contradicting N is closer
to ST.) Let the tangent lines meet at K, then 6 OSK = 90 = 6 OT K
implies O, S, K, T lie on a circle with diameter OK. Also, KS 2 = KT 2
implies K is on the radical axis M N of the two inside circles. So
M, N, K are collinear.
If S, N, T are collinear, then

6 SM T = 6 SM N + 6 T M N = 6 N SK + 6 KT N = 180 6 SKT.

So M, S, K, T, O are concyclic. Then 6 OM N = 6 OM K = 6 OSK =

90 .
Conversely, if OM M N, then 6 OM K = 90 = 6 OSK implies
M, S, K, T, O are concyclic. Then
6 SKT = 180 6 SM T
= 180 6 SM N 6 T M N
= 180 6 N SK 6 KT N.

92
 Thus, 6 T N S = 360 6 N SK 6 SKT 6 KT N = 180 . Therefore,
S, N, T are collinear.

99. AD, BE, CF are the altitudes of 4ABC. Lines EF, F D, DE meet lines
BC, CA, AB in points L, M, N, respectively. Show that L, M, N are
collinear and the line through them is perpendicular to the line joining
the orthocenter H and circumcenter O of 4ABC.

Solution. Since 6 ADB = 90 = 6 AEB, A, B, D, E are concyclic. By

the intersecting chord theorem, N A N B = N D N E. So the power of
N with respect to the circumcircles of 4ABC, 4DEF are the same.
Hence N is on the radical axis of these circles. Similarly, L, M are also
on this radical axis. So L, M, N are collinear.
Since the circumcircle of 4DEF is the nine-point circle of 4ABC,
the center N of the nine-point circle is the midpoint of H and O. Since
the radical axis is perpendicular to the line of centers O and N, so the
line through L, M, N is perpendicular to the line HO.

Geometric Inequalities, Maximum/Minimum

100. (1973 IMO) Let P1 , P2 , . . . , P2n+1 be distinct points on some half of

the unit circle centered at the origin O. Show that

|OP1 + OP2 + + OP2n+1 | 1.

Solution. When n = 0, then |OP1 | = 1. Suppose the case n = k is true.
For the case n = k + 1, we may assume P1 , P2 , . . . , P2k+3 are arranged

clockwise. Let OR = OP2 + + OP2k+2 and OS = OP1 + OP2k+3 .



By the case n = k, |OR | 1. Also, OR lies inside 6 P1 OP2k+3 . Since

|OP1 | = 1 = |OP2k+3 |, OS bisects 6 P1 OP2k+3 . Hence 6 ROS 90 .


Then |OP1 + + OP2k+3 | = |OR + OS | |OR | 1.

101. Let the angle bisectors of 6 A, 6 B, 6 C of triangle ABC intersect its

circumcircle at P, Q, R, respectively. Prove that

AP + BQ + CR > BC + CA + AB.

93
 Solution. (Due to Lau Lap Ming) Since 6 ABQ = 6 CBQ, we have
AQ = CQ. By cosine law,

AQ2 = AB 2 + BQ2 2AB BQ cos 6 ABQ

CQ2 = CB 2 + BQ2 2CB BQ cos 6 CBQ.

If AB 6= CB, then subtracting these and simplifying, we get AB +

CB = 2BQ cos 6 ABQ < 2BQ. If AB = CB, then BQ is a diameter
and we again get AB +CB = 2AB < 2BQ. Similarly, BC +AC < 2CR
and CA + BA < 2AP. Adding these inequalities and dividing by 2, we
get the desired inequality.

102. (1997 APMO) Let ABC be a triangle inscribed in a circle and let la =
ma /Ma , lb = mb /Mb , lc = mc /Mc , where ma , mb , mc are the lengths
of the angle bisectors (internal to the triangle) and Ma , Mb , Mc are
the lengths of the angle bisectors extended until they meet the circle.
Prove that
la lb lc
2 + 2 + 3,
sin A sin B sin2 C
and that equality holds iff ABC is equilateral.

Solution. (Due to Fung Ho Yin) Let A0 be the point the angle bisector
of 6 A extended to meet the circle. Applying sine law to 4ABA0 , we
get AB/ sin C = Ma / sin(B + A 2
). Applying sine law to 4ABD, we get
AB/ sin(C + A2
) = ma / sin B. So

ma sin B sin C
la = = sin B sin C.
Ma sin(B + A2
) sin(C + A
2
)

By the AM-GM inequality,

la lb lc sin B sin C sin C sin A sin A sin B

+ + + + 3
sin2 A sin2 B sin2 C sin2 A sin2 B sin2 C

with equality if and only if sin A = sin B = sin C and C + A

2
= B+ A
2
=

= 90 , which is equivalent to 6 A = 6 B = 6 C.

94
103. (Mathematics Magazine, Problem 1506) Let I and O be the incen-
ter and circumcenter of 4ABC, respectively. Assume 4ABC is not
equilateral (so I 6= O). Prove that

6 AIO 90 if and only if 2BC AB + CA.

Solution. (Due to Wong Chun Wai) Let D be the intersection of ray AI

and the circumcircle of 4ABC. It is well-known that DC = DB = DI.
(DC = DB because 6 CAD = 6 BAD and DB = DI because 6 BID =
6 BAD + 6 ABI = 6 CAD + 6 CBI = 6 DBC + 6 CBI = 6 DBI.) Since
ABDC is a cyclic quadrilateral, by Ptolemys theorem, AD BC =
AB DC +AC DB = (AB+AC)DI. Then DI = ADBC/(AB+AC).
Since 4AOD is isosceles, 6 AIO = 90 if and only if DI AD/2, which
is equivalent to 2BC AB + AC.
Comments. In the solution above, we see 6 AIO = 90 if and only if
2BC = AB + AC. Also, the converse of the well-known fact is true,
i.e. the point I on AD such that DC = DB = DI is the incenter of
4ABC. This is because 6 BID = 6 DBI if and only if 6 CBI = 6 ABI,
since 6 DBC = 6 BAD always.)

104. Squares ABDE and ACF G are drawn outside 4ABC. Let P, Q be
points on EG such that BP and CQ are perpendicular to BC. Prove
that BP + CQ BC + EG. When does equality hold?

Solution. Let M, N, O be midpoints of BC, P Q, EG, respectively. Let

H be the point so that HEAG is a parallelogram. Translating by GA,
then rotating by 90 about A, 4GHA will coincide with 4ABC and O
will move to M. So HA = BC, HA BC, OE = OG = M A, EG
M A. Let L be on M N such that AL k EG. Since N L k P B, P B
BC, BC HA, so LN OA is a parallelogram. Then AO = LN. Since
M A EG, so M A AL, which implies M L M A. Therefore

BP + CQ = 2M N = 2(LN + M L)
2(AO + M A) = 2(BM + OE) = BC + EG.

Equality holds if and only if L coincides with A, i.e. AB = AC.

95
105. Point P is inside 4ABC. Determine points D on side AB and E on
side AC such that BD = CE and P D + P E is minimum.

Solution. The minimum is attained when ADP E is a cyclic quadri-

lateral. To see this, consider the point G such that G lies on the
opposite side of line AC as B, 6 ABP = 6 ACG and CG = BP.
Let E be the intersection of lines AC and P G. Let D be the in-
tersection of AB with the circumcircle of AP E. Since ADP E is a
cyclic quadrilateral, 6 BDP = 6 AEP = 6 CEG. Using the defini-
tion of G, we have 4BDP, 4CEG are congruent. So BD = CE and
P D + P E = GE + P E = GP.
For other D 0 , E 0 on sides AB, AC, respectively such that BD 0 =
CE 0 , by the definition of G, we have 4BP D 0 , 4CGE 0 are congruent.
Then P D 0 = GE 0 and P D 0 + P E 0 = GE 0 + P E 0 > GP.

Solid or Space Geometry

106. (Proposed by Italy for 1967 IMO) Which regular polygons can be ob-
tained (and how) by cutting a cube with a plane?

Solution. (Due to Fan Wai Tong, Kee Wing Tao and Tam Siu Lung)
Observe that if two sides of a polygon is on a face of the cube, then
the whole polygon lies on the face. Since a cube has 6 faces, only
regular polygon with 3, 4, 5 or 6 sides are possible. Let the vertices
of the bottom face of the cube be A, B, C, D and the vertices on the
top face be A0 , B 0 , C 0 , D 0 with A0 on top of A, B 0 on top of B and so
on. Then the plane through A, B 0 , D 0 cuts an equilateral triangle. The
perpendicular bisecting plane to edge AA0 cuts a square. The plane
through the midpoints of edges AB, BC, CC 0 , C 0 D 0 , D 0 A0 , A0 A cuts a
regular hexagon. Finally, a regular pentagon is impossible, otherwise
the five sides will be on five faces of the cube implying two of the
sides are on parallel planes, but no two sides of a regular pentagon are
parallel.

107. (1995 Israeli Math Olympiad) Four points are given in space, in general
position (i.e., they are not coplanar and any three are not collinear).

96
A plane is called an equalizing plane if all four points have the same
distance from . Find the number of equalizing planes.

Solution. The four points cannot all lie on one side of an equalizing
plane, otherwise they would lie in a plane parallel to the equalizing
plane. Hence either three lie on one side and one on the other or two
lie on each side. In the former case, there is exactly one equalizing
plane, which is parallel to the plane P containing the three points and
passing through the midpoint of the segment joining the fourth point x
and the foot of the perpendicular from x to P. In the latter case, again
there is exactly one equalizing plane. The two pair of points determine
two skew lines in space. Consider the two planes, each containing one
of the line and is parallel to the other line. The equalizing plane is
the plane midway between these two plane. Since there are 4 + 3 = 7
ways of dividing the four points into these two cases, there are exactly
7 equalizing planes.

97
 Solutions to Number Theory Problems

Digits

108. (1956 Putnam Exam) Prove that every positive integer has a multiple
whose decimal representation involves all ten digits.

Solution. Let n be a positive integer and p = 1234567890 10k , where

k is so large that 10k > n. Then the n consecutive integers p + 1, p +
2, , p+n have decimal representations beginning with 1234567890
and one of them is a multiple of n.

109. Does there exist a positive integer a such that the sum of the digits
(in base 10) of a is 1999 and the sum of the digits (in base 10) of a2 is
19992 ?

Solution. Yes. In fact, there is such a number whose digits consist of

1 2 k
0s and 1s. Let k = 1999. Consider a = 102 + 102 + + 102 . Then
the sum of the digits of a is k. Now
2 3 k+1
X i
+2j
a2 = 102 + 102 + + 102 +2 102 .
1i<jk

Observe that the exponent are all different by the uniqueness of base
2 representation. Therefore, the sum of the digits of a2 in base 10 is
k + 2C2k = k 2 .

110. (Proposed by USSR for 1991 IMO) Let an be the last nonzero digit
in the decimal representation of the number n!. Does the sequence
a1 , a2 , . . . , an , . . . become periodic after a finite number of terms?

Solution. Suppose after N terms, the sequence becomes periodic with

period T. Then ai+jT = ai for i N, j = 1, 2, 3 . . . . By the pigeonhole
principle, there are two numbers among 10N+1 , 10N+2 , 10N+3 , . . . that
have the same remainder when divided by T, say 10m 10k (mod T )
with N < m < k. Then 10k 10m = jT for some integer j.

98
 Observe that 10k ! = 10k (10k 1)! implies a10k = a10k 1 . Let
n = 10k 1+jT. Since 10k 1 N, an+1 = a10k +jT = a10k 1+jT = an .
Since (n + 1)! = (2 10k 10m )(n!) = 199 90 0(n!), so 9an =
9an1 an (mod 10). This implies an = 5. However, in the prime
factorization of n!, the exponent of 2 is greater than the exponent of
5, which implies an is even, a contradiction.

Modulo Arithmetic

111. (1956 Putnam Exam) Prove that the number of odd binomial coeffi-
cients in any row of the Pascal triangle is a power of 2.
m m
Solution. By induction, (1 + x)2 1 + x2 (mod 2). If we write n
in base 2, say n = 2a1 + 2a2 + + 2ak , where the ai s are distinct
nonnegative integers, then
a1 ak a1 ak
(1 + x)n = (1 + x)2 (1 + x)2 (1 + x2 ) (1 + x2 ) (mod 2).

nS
In expanding the expression in front of (mod 2), Xwe get the sum of x ,
where for each subset S of {1, 2, . . . , k}, nS = 2ai . Since there are 2k
iS
k
subsets of {1, 2, . . . , k}, there are exactly 2 terms, each with coefficient
1. This implies there are exactly 2k odd binomial coefficients in the
n-th row of the Pascal triangle.

112. Let a1 , a2 , a3 , . . . , a11 and b1 , b2 , b3 , . . . , b11 be two permutations of the

natural numbers 1, 2, 3, . . . , 11. Show that if each of the numbers a1 b1 ,
a2 b2 , a3 b3 , . . . , a11 b11 is divided by 11, then at least two of them will
have the same remainder.

Solution. Suppose a1 b1 ,a2 b2 , . . . ,a11 b11 have distinct remainders when

divided by 11. By symmetry, we may assume a1 b1 0 (mod 11). Let
x = (a2 b2 ) (a11 b11 ). On one hand, x 10! 10 (mod 11). On the
other hand, since ai bi 6 0 (mod 11), for i = 2, . . . , 11, we get a1 =
11 = b1 . So x = (a2 a11 )(b2 b11 ) = (10!)2 102 1 (mod 11), a
contradiction.

99
113. (1995 Czech-Slovak Match) Let a1 , a2 , . . . be a sequence satisfying a1 =
2, a2 = 5 and
an+2 = (2 n2 )an+1 + (2 + n2 )an
for all n 1. Do there exist indices p, q and r such that ap aq = ar ?

Solution. (Due to Lau Lap Ming) The first few terms are 2, 5, 11, 8, 65,
766, . . . . Since the differences of consecutive terms are multiples of 3,
we suspect an 2 (mod 3) for all n. Clearly, a1 , a2 2 (mod 3). If
an , an+1 2 (mod 3), then

an+2 (2 n2 )2 + (2 + n2 )2 = 8 2 (mod 3).

So by induction, all an 2 (mod 3). Then ap aq 6= ar for all p, q, r as

4 6 2 (mod 3).

Prime Factorization

114. (American Mathematical Monthly, Problem E2684) Let An be the set

of positive integers which are less than n and are relatively prime to n.
For which n > 1, do the integers in An form an arithmetic progression?

Solution. Suppose An is an arithmetic progression. If n is odd and

n 3, then 1, 2 An implies An = {1, 2, . . . , n 1}, which implies n is
prime. If n is even and not divisible by 3, then 1, 3 An , 2 6 An imply
An = {1, 3, 5, . . . , n 1}, which implies n is a power of 2. Finally, if
n is even and divisible by 3, then let p be the smallest prime not
dividing n. Either p 1 (mod 6) or p 5 (mod 6). In the former
case, since 1, p are the first two elements of An and n 1 An , so
1 + k(p 1) = n 1 for some k. This implies n 2 (mod 6), a
contradiction. So p 5 (mod 6). Then 2p 1 is divisible by 3 and
so 2p 1 6 An . Consequently, An = {1, p}, which implies n = 6 by
considering the prime factorization of n. Therefore, An is an arithmetic
progression if and only if n is a prime, a power of 2 or n = 6.

115. (1971 IMO) Prove that the set of integers of the form 2k 3 (k =
2, 3, . . .) contains an infinite subset in which every two members are
relatively prime.

100
 Solution. We shall give a recipe for actually constructing an infinite
set of integers of the form ai = 2ki 3, i = 1, 2, . . . , each relatively
prime to all the others. Let a1 = 22 3 = 1. Suppose we have n pairwise
relatively prime numbers a1 = 2k1 3, a2 = 2k2 3, . . . , an = 2kn 3.
We form the product s = a1 a2 an , which is odd. Now consider the
s + 1 numbers 20 , 21 , 22 , . . . , 2s . At least two of these will be congruent
(mod s), say 2 2 (mod s), or equivalently 2 (2 1) = ms
for some integer m. The odd number s does not divide 2 , so it must
divide 2 1; hence 2 1 = ls for some integer l. Since 2 1
is divisible by s and s is odd, 2 3 is relatively prime to s. This
implies 2 3 6= 2ki 3 for i = 1, 2, . . . , n. So we may define an+1 =
2 3. This inductive construction can be repeated to form an infinite
sequence.
Comments. By Eulers theorem, we may take the exponent to be
(s), the Euler -function of s, which equals the number of positive in-
tegers less than s that are relatively prime to s, then 2(s) 1 (mod s).

116. (1988 Chinese Math Olympiad Training Test) Determine the smallest
value of the natural number n > 3 with the property that whenever
the set Sn = {3, 4, . . . , n} is partitioned into the union of two sub-
sets, at least one of the subsets contains three numbers a, b and c (not
necessarily distinct) such that ab = c.

Solution. (Due to Lam Pei Fung) We first show that 35 = 243 has the
property, then we will show it is the least solution.
Suppose S243 is partitioned into two subsets X1 , X2 . Without loss
of generality, let 3 be in X1 . If 32 = 9 is in X1 , then we are done.
Otherwise, 9 is in X2 . If 92 = 81 is in X2 , then we are done. Otherwise,
81 is in X1 . If 81/3 = 27 is in X1 , then we are done. Otherwise, 27 is
in X2 . Finally, either 3 81 = 243 is in X1 or 9 27 = 243 is in X2 .
In either case we are done.
To show 243 is the smallest, we will show that S242 can be par-
titioned into two subsets, each of which does not contain products of
its elements. Define C to be prime in S242 if C is not the product
of elements of S242 . The primes in S242 consist of 4, 8, p, 2p where
p < 242 is a usual prime number. Since the smallest prime in S242 is

101
 3, no number in S242 is the product of more than four primes. Put
all the primes and numbers that can be written as products of four
primes in one subset X1 , and let X2 = S242 \ X1 .
No products in X2 are in X2 because numbers in X2 have at least
two prime factors, so their products can be written with at least four
prime factors. Next looking at the product of 4, 8, p, 2p (p odd prime
< 242), we see that a product of two primes cannot be factored into
a product of four primes. So no products in X1 are in X1 .

Base n Representations

117. (1983 IMO) Can you choose 1983 pairwise distinct nonnegative integers
less than 105 such that no three are in arithmetic progression?

Solution. We consider the greedy algorithm for constructing such a

sequence: start with 0, 1 and at each step add the smallest integer
which is not in arithmetic progression with any two preceding terms.
We get 0, 1, 3, 4, 9, 10, 12, 13, 27, 28, . . . . In base 3, this sequence is

0, 1, 10, 11, 100, 101, 110, 111, 1000, 1001, . . . .

(Note this sequence is the nonnegative integers in base 2.) Since 1982 in
base 2 is 11110111110, so switching this from base 3 to base 10, we get
the 1983th term of the sequence is 87843 < 105 . To see this sequence
works, suppose x, y, z with x < y < z are three terms of the sequence
in arithmetic progression. Consider the rightmost digit in base 3 where
x differs from y, then that digit for z is a 2, a contradiction.

118. (American Mathematical Monthly, Problem 2486) Let p be an odd

prime number and r be a positive integer not divisible by p. For any
positive integer k, show that there exists a positive integer m such that
the rightmost k digits of mr , when expressed in the base p, are all 1s.

Solution. We prove by induction on k. For k = 1, take m = 1. Next,

suppose mr , in base p, ends in k 1s, i.e.

mr = 1 + p + + pk1 + (apk + higher terms).

102
 Clearly, gcd(m, p) = 1. Then

(m + cpk )r = mr + rmr1 cpk + + cr pkr i

= 1 + p + + pk1 + (a + rmr1 c)pk + higher terms.

Since gcd(mr, p) = 1, the congruence a + rmr1 c 1 (mod p) is solv-

able for c. If c0 is a solution, then (m + c0 pk )r will end in (k + 1) 1s
as required.

119. (Proposed by Romania for 1985 IMO) Show that the sequence {an }
defined by an = [n 2] for n = 1, 2, 3, . . . (where the brackets denote
the greatest integer function) contains an infinite number of integral
powers of 2.

Solution.
Write 2 in base 2 as b0 .b1 b2 b3 . . . , where each bi = 0 or 1.
Since 2 is irrational,
there are infinitely many bk =1. If bk = 1, then
k1
in base 2, 2 2 = b0 bk1 .bk Let m = [2k1 2], then
1
2k1 2 1 < [2k1 2] = m < 2k1 2 .
2

2
Multiplying by 2 and adding 2, we get 2k < (m + 1) 2 < 2k + .
2
Then [(m + 1) 2] = 2k .

Representations

120. Find all (even) natural numbers n which can be written as a sum of
two odd composite numbers.

Solution. Let n 40 and d be its units digit. If d = 0, then n =

15 + (n 15) will do. If d = 2, then n = 27 + (n 27) will do. If d = 4,
then n = 9 + (n 9) will do. If d = 6, then n = 21 + (n 21) will do. If
d = 8, then n = 33+(n33) will do. For n < 40, direct checking shows
only 18 = 9 + 9, 24 = 9 + 15, 30 = 9 + 21, 34 = 9 + 25, 36 = 9 + 27
can be so expressed.

121. Find all positive integers which cannot be written as the sum of two
or more consecutive positive integers.

103
 Solution. (Due to Cheung Pok Man) For odd integer n = 2k + 1 3,
n = k + (k + 1). For even integer n 2, suppose n = m + (m + 1) +
+ (m + r) = (2m + r)(r + 1)/2 with m, r 1. Then 2m + r, r + 1 2
and one of 2m + r, r + 1 is odd. So n must have an odd divisor greater
than 1. In particular, n = 2j , j = 0, 1, 2, . . . , cannot be written as
the sum of consecutive positive integers. For the other even integers,
n = 2j (2k +1) with j, k 1. If 2j > k, then n = (2j k)+(2j k +1)+
+(2j +k). If 2j k, then n = (k2j +1)+(k2j +2)+ +(k+2j ).

122. (Proposed by Australia for 1990 IMO) Observe that 9 = 4+5 = 2+3+4.
Is there an integer N which can be written as a sum of 1990 consecutive
positive integers and which can be written as a sum of (more than one)
consecutive integers in exactly 1990 ways?
1989
X
Solution. For such N , we have N = (m + i) = 995(2m + 1989). So
i=0
N is odd and is divisible by 995 = 5 199. Also, there are exactly 1990
X k
(k + 1)(n + 2k)
positive integer pairs (n, k) such that N = (n + i) = .
2
i=0
Hence 2N can be factored as (k+1)(2n+k) in exactly 1990 ways. (Note
if 2N = ab with 2 a < b, then n = (1 + b a)/2, k = a 1.) This
means 2N has exactly 2 1991 = 2 11 181 positive divisors. Now
write 2N in prime factorization as 2 5e1 199e2 . Then we get
2 11 181 = 2(e1 + 1)(e2 + 1) . So {e1 , e2 } = {10, 180}. Therefore,
N = 510 199180 or 5180 19910 . As all the steps can be reversed, these
are the only answers.

123. Show that if p > 3 is prime, then pn cannot be the sum of two positive
cubes for any n 1. What about p = 2 or 3?

Solution. Suppose n is the smallest positive integer such that pn is the

sum of two positive cubes, say pn = a3 +b3 = (a+b)(a2 ab+b2). Then
a + b = pk and a2 ab + b2 = pnk . Since a + b 2, so k > 0. Since
a2 ab + b2 ab > 1, so n > k. Now 3ab = (a + b)2 (a2 ab + b2 ) =
p2k pnk and 0 < k < n, so p | 3ab. Since p > 3, so p | a or p | b.
Since a + b = pk , so p | a and p | b, say a = pA and b = pB. Then
A3 + B 3 = pn3 , contradicting the smallest property of n.

104
 For p = 2, suppose a3 + b3 = 2n . If a + b > 2, then 2 | a, 2 | b and
( a2 )3 + ( 2b )3 = 2n3 . So a = b = 2k and n = 3k + 1.
For p = 3, suppose a3 + b3 = 3n . If a + b = 3k and a2 ab + b2 =
3nk 32 , then 9 | 3ab implies 3 | a, 3 | b and ( a3 )3 + ( 3b )3 = 3n3 .
Otherwise we have 3 = a2 ab + b2 ab and then a + b = 3. So in this
case, a, b are 2 3k , 3k and n = 3k + 2.

124. (Due to Paul Erdos and M. Suranyi) Prove that every integer k can be
represented in infinitely many ways in the form k = 12 22 m2
for some positive integer m and some choice of signs + or .

Solution. We first show every integer can be so represented in at least

one way. If k can be represented, then changing all the signs, we see
k also can be represented. So it suffices to do the nonnegative cases.
The key observation is the identity

(m + 1)2 (m + 2)2 (m + 3)2 + (m + 4)2 = 4.

Now 0 = 4 4 = (12 22 + 32 ) 42 + 52 + 62 72 , 1 = 12 , 2 =
12 22 32 +42 , 3 = 12 +22 . By the identity, if k can be represented,
then k + 4 can be represented. So by induction, every nonnegative
integer (and hence every integer) can be represented. To see there
are infinitely many such representations, we use the identity again.
Observe 0 = 4 4 = (m + 1)2 (m + 2)2 (m + 3)2 + (m + 4)2 (m +
5)2 + (m + 6)2 + (m + 7)2 (m + 8)2 . So for every representation, we
can add 8 more terms to get another representation.

125. (1996 IMO shortlisted problem) A finite sequence of integers a0 , a1 , . . . ,

an is called quadratic if for each i {1, 2, . . . , n}, |ai ai1 | = i2 .
(a) Prove that for any two integers b and c, there exists a natural
number n and a quadratic sequence with a0 = b and an = c.
(b) Find the least natural number n for which there exists a quadratic
sequence with a0 = 0 and an = 1996.

Solution. Part (a) follows from the last problem by letting k = c

b. For part (b), consider ak in such a quadratic sequence. We have

105
 ak 12 + 22 + + k 2 = k(k + 1)(2k + 1)/6. So a17 1785. Also
ak 12 + 22 + + k 2 (mod 2). Since 12 + 22 + + 182 is odd,
n 19. To construct such a quadratic sequence with n = 19, first note
12 + 22 + + 192 = 2470. Now we write (2470 1996)/2 = 237 =
142 + 52 + 42 . Then

1996 = 12 + 22 + 32 42 52 + 62 + + 132 142 + 152 + + 192 .

126. Prove that every integer greater than 17 can be represented as a sum of
three integers > 1 which are pairwise relatively prime, and show that
17 does not have this property.

Solution. (Due to Chan Kin Hang and Ng Ka Wing) Let k 3. From

18 = 2+3+13, we see 2+3+(6k5) works for 6k. From 20 = 3+4+13,
we see 3 + 4 + (6k 4) works for 6k + 2. From 22 = 2 + 3 + 17, we see
2 + 3 + (6k 1) works for 6k + 4.
For 6k + 1, we split into cases 12k 0 + 1 and 12k 0 + 7. We have
12k 0 +1 = 9+(6k 0 1)+(6k 0 7) and 12k 0 +7 = 3+(6k 0 1)+(6k 0 +5).
For 6k + 3, we split into cases 12k 0 + 3 and 12k 0 + 9. We have
12k 0 +3 = 3+(6k 0 1)+(6k 0 +1) and 12k 0 +9 = 9+(6k 0 1)+(6k 0 +1).
For 6k + 5, we split into cases 12k 0 + 5 and 12k 0 + 11. We have
12k 0 +5 = 9+(6k 0 5)+(6k 0 +1) and 12k 0 +11 = 3+(6k 0 +1)+(6k 0 +7).
Finally, 17 does not have the property. Otherwise, 17 = a + b + c,
where a, b, c are relatively prime and a < b < c. Then a, b, c are odd. If
a = 3, then 3 + 5 + 7 < a + b + c < 3 + 5 + 11 shows this is impossible.
If a 5, then b 7, c 9 and a + b + c 21 > 17, again impossible.

Chinese Remainder Theorem

127. (1988 Chinese Team Selection Test) Define xn = 3xn1 + 2 for all
positive integers n. Prove that an integer value can be chosen for x0 so
that x100 is divisible by 1988.

106
 xn 2
Solution. Let yn = , then y n = y n1 + , which implies
3n 3n
2 2 2
yn = y0 + + 2 + + n.
3 3 3
This gives xn = (x0 + 1)3n 1. We want x100 = (x0 + 1)3100 1 to be
divisible by 1988 = 4 7 71, which means
x0 0 (mod 4), x0 1 (mod 7), x0 45 (mod 71).
Since 4, 7, 71 are pairwise relatively prime, by the Chinese remainder
theorem, such x0 exists.

128. (Proposed by North Korea for 1992 IMO) Does there exist a set M
with the following properties:
(a) The set M consists of 1992 natural numbers.
(b) Every element in M and the sum of any number of elements in M
have the form mk , where m, k are positive integers and k 2?

Solution. (Due to Cheung Pok Man) Let n = 1+2+ +1992. Choose

n distinct prime numbers p1 , p2 , . . . , pn . Let d = 2e2 3e3 4e4 nen , where
ei is a solution of the n equations x 1 (mod pi ) and x 0 (mod pj )
for every 1 j n, j 6= i. (Since the pi s are pairwise relatively
prime, such a solution exists by the Chinese remainder theorem.) Since
e2 , e3 , . . . , en 0 (mod p1 ), d is a pi -th power. Since e2 +1, e3 , . . . , en
0 (mod p2 ), 2d is a p2 -th power and so on. It follows d, 2d, . . . , 1992d
are all perfect powers and any sum of them is a multiple of d, less than
or equal to nd, hence is also a perfect power.

Divisibility

129. Find all positive integers a, b such that b > 2 and 2a + 1 is divisible by
2b 1.

Solution. Since b > 2, so 2b 1 < 2a + 1, hence b < a. Let a =

qb + r, 0 r < b, then by division, we get
2a + 1 ab a2b aqb 2r + 1
= 2 + 2 + + 2 + .
2b 1 2b 1

107
 2r + 1
Since 0 < < 1, there are no solutions.
2b 1

130. Show that there are infinitely many composite n such that 3n1 2n1
is divisible by n.

Solution. We use the fact xy | xk y k for positive integer k. Consider

t t t
n = 32 22 for t = 2, 3, . . . . By induction, we can show 2t | 32
t t t
1 = n 1 + 22 . (Alternatively, by Eulers theorem, 32 = (3(2 ) )2
t t t t
1 (mod 2t ).) Then n 1 = 2t k. So n = 32 22 | (32 )k (22 )k =
3n1 2n1 .

131. Prove that there are infinitely many positive integers n such that 2n +1
is divisible by n. Find all such ns that are prime numbers.

Solution. Looking at the cases n = 1 to 10 suggest for n = 3k , k =

0, 1, 2, . . . , we should have n | 2n + 1. The case k = 0 is clear. Suppose
k+1 k k k
case k is true. Now 23 + 1 = (23 + 1)(23 2 23 + 1). By case k,
k k
23 1 (mod 3), so 23 2 23k + 1 (1)2 (1) + 1 0 (mod 3). So
k+1
23 + 1 is divisible by 3k+1 , completing the induction.
If a prime n divides 2n +1, then by Fermats little theorem, n | 2n
2, too. Then n | (2n + 1) (2n 2) = 3, so n = 3.

132. (1998 Romanian Math Olympiad) Find all positive integers (x, n) such
that xn + 2n + 1 is a divisor of xn+1 + 2n+1 + 1.

Solution. (Due to Cheng Kei Tsi and Leung Wai Ying) For x = 1,
2(1n +2n +1) > 1n+1 +2n+1 +1 > 1n +2n +1. For x = 2, 2(2n +2n +1) >
2n+1 + 2n+1 + 1 > 2n + 2n + 1. For x = 3, 3(3n + 2n + 1) > 3n+1 +
2n+1 + 1 > 2(3n + 2n + 1). So there are no solutions with x = 1, 2, 3.
For x 4, if n 2, then we get x(xn + 2n + 1) > xn+1 + 2n+1 + 1.
Now
xn+1 + 2n+1 + 1
=(x 1)(xn + 2n + 1)
+ xn (2n + 1)x + 3 2n + 2
>(x 1)(xn + 2n + 1)

108
 because for n = 2, xn (2n + 1)x + 2n+1 = x2 5x + 8 > 0 and for
n 3, xn (2n + 1)x x(4n1 2n 1) > 0. Hence only n = 1
and x 4 are possible. Now xn + 2n + 1 = x + 3 is a divisor of
xn+1 + 2n+1 + 1 = x2 + 5 = (x 3)(x + 3) + 14 if and only if x + 3
is a divisor of 14. Since x + 3 7, x = 4 or 11. So the solutions are
(x, y) = (4, 1) and (11, 1).

133. (1995 Bulgarian Math Competition) Find all pairs of positive integers
x2 + y 2
(x, y) for which is an integer and divides 1995.
xy

Solution. Suppose (x, y) is such a pair. We may assume x > y, oth-

erwise consider (y, x). Then x2 + y 2 = k(x y), where k | 1995 =
3 5 7 19. If p = 3 or 7 or 19 divides k, then by the fact that
prime p 3 (mod 4) dividing x2 + y 2 implies p divides x and y,
we may cancel p2 to get an equation x20 + y02 = k0 (x0 y0 ) with
k0 not divisible by 3, 7, 19. Since x20 + y02 > x20 > x0 > x0 y0 ,
we must have x20 + y02 = 5(x0 y0 ). Completing squares, we get
(2x0 5)2 + (2y0 + 5)2 = 50, which gives (x0 , y0 ) = (3, 1) or (2, 1).
It follows (x, y) = (3c, c), (2c, c), (c, 3c), (c, 2c), where c is a positive
divisor of 3 7 19.

134. (1995 Russian Math Olympiad) Is there a sequence of natural numbers

in which every natural number occurs just once and moreover, for any
k = 1, 2, 3, . . . the sum of the first k terms is divisible by k?

Solution. Let a1 = 1. Suppose a1 , . . . , ak has been chosen to have

the property. Let n be the smallest natural number not yet appeared.
By the Chinese remainder theorem, there is an integer m such that
m a1 ak (mod k +1) and m a1 ak n (mod k +2).
We can increase m by a large multiple of (k + 1)(k + 2) to ensure it
is positive and not equal to anyone of a1 , . . . , ak . Let ak+1 = m and
ak+2 = n. The sequence constructed this way have the property.

135. (1998 Putnam Exam) Let A1 = 0 and A2 = 1. For n > 2, the number
An is defined by concatenating the decimal expansions of An1 and
An2 from left to right. For example, A3 = A2 A1 = 10, A4 = A3 A2 =

109
 101, A5 = A4 A3 = 10110, and so forth. Determine all n such that An
is divisible by 11.

Solution. The Fibonacci numbers Fn is defined by F1 = 1, F2 = 1 and

Fn = Fn1 + Fn2 for n > 2. Note An has Fn digits. So we have the
recursion An = 10Fn2 An1 +An2 (1)Fn2 An1 +An2 (mod 11).
By induction, the sequence Fn (mod 2) is 1, 1, 0, 1, 1, 0, . . . . The first
eight terms of An (mod 11) are 0, 1, 1, 2, 1, 1, 0, 1. (Note the numbers
start to repeat after the sixth term.) In fact, the recursion implies
An+6 An (mod 11) by induction. So An is divisible by 11 if and only
if n = 6k + 1 for some positive integer k.

136. (1995 Bulgarian Math Competition) If k > 1, show that k does not
divide 2k1 + 1. Use this to find all prime numbers p and q such that
2p + 2q is divisible by pq.

Solution. Suppose k | 2k1 + 1 for some k > 1. Then k is odd. Write

k = pe11 perr , where pi s are distinct primes. Let pi 1 = 2mi qi
withe qi odd. Let mj = min{m1 , . . . , mr }. Since mi 1 (mod mj ), we
get pei i 1 (mod mj ) and so k = 2mj q + 1 for some positive integer
mj
q. Since pj | k and k | 2k1 + 1, so 22 q = 2k1 1 (mod pj ).
mj
Then 2(pj 1)q = (22 q )qj 1 (mod pj ) because qj is odd. However,
2pj 1 1 (mod pj ) by Fermats little theorem since gcd(2, pj ) = 1. So
2(pj 1)q 1 (mod pj ), a contradiction.
Suppose p, q are prime and 2p + 2q is divisible by pq. Then 2p
2q (mod p). If p, q are odd, then 2p 2 (mod p) by Fermats little
theorem and 2q 2p 2 (mod p). So 2pq (2)p 2 (mod p).
Similarly, 2pq 2 (mod q). Then 2pq1 1 (mod pq), contradict-
ing the first part of the problem. If p = 2, then q = 2 or q > 2. If
q > 2, then 22 2q 2 (mod q) by Fermats little theorem, which
implies q = 3. Therefore, the solutions are (p, q) = (2, 2), (2, 3), (3, 2).

137. Show that for any positive integer n, there is a number whose decimal
representation contains n digits, each of which is 1 or 2, and which is
divisible by 2n .

Solution. We will prove that the 2n numbers with n digits of 1s or

2s have different remainders when divided by 2n . Hence one of them is

110
 divisible by 2n . For n = 1, this is clear. Suppose this is true for n = k.
Now if a, b are (k + 1)-digit numbers, where each digit equals 1 or 2,
and a b (mod 2k+1 ), then the units digits of a, b are the same. If
a = 10a0 + i, b = 10b0 + i, where i is the units digit, then 2k+1 divides
a b = 10(a0 b0 ) is equivalent to 2k divides a0 b0 . Since a0 , b0 are
k-digit numbers (with digits equal 1 or 2), we have a0 = b0 . So a = b,
completing the induction.

138. For a positive integer n, let f (n) be the largest integer k such that 2k
divides n and g(n) be the sum of the digits in the binary representation
of n. Prove that for any positive integer n,
(a) f (n!) = n
g(n);

2n (2n)!
(b) 4 divides = if and only if n is not a power of 2.
n n!n!

Solution. (Due to Ng Ka Man and Poon Wing Chi) (a) Write n in

base 2 as (ar ar1 a0 )2 . Then
n  n 
ai = (ar ai+1 ai )2 (ar ai+1 0)2 = 2 .
2i 2i+1

So
r 
r
X X n  n  r
X n
g(n) = ai = 2 i+1 = n = n f (n!).
2i 2 2i
i=0 i=0 i=0

(b) Let Mn = (2n)!/(n!)2 . Since g(2n) = g(n), using (a), we get

f (Mn ) = f ((2n)!) 2f (n!) = 2g(n) g(2n) = g(n).

So the largest k such that 2k divides Mn is k = g(n). Now 4 divides

Mn if and only if g(n) 2, which is equivalent to n not being a power
of 2.

139. (Proposed by Australia for 1992 IMO) Prove that for any positive in-
teger m, there exist an infinite number of pairs of integers (x, y) such
that

111
 (a) x and y are relatively prime;
(b) y divides x2 + m;
(c) x divides y 2 + m.

Solution. Note (x, y) = (1, 1) is such a pair. Now, if (x, y) is such

a pair with x y, then consider (y, z), where y 2 + m = xz. Then
every common divisor of z and y is a divisor of m, and hence of x. So
gcd(z, y) = 1. Now

x2 (z 2 + m) = (y 2 + m) + x2 m = y 4 + 2my 2 + m(x2 + m)

is divisible by y. Since gcd(x, y) = 1, y | z 2 + m, so (y, z) is another

such pair with y y 2 /x < z. This can be repeated infinitely many
times.

140. Find all integers n > 1 such that 1n + 2n + + (n 1)n is divisible

by n.

Solution. For odd n = 2j + 1 > 1, since (n k)n k n (mod n) for

1 k j, so 1n + 2n + + (n 1)n is divisible by n. For even n, write
n = 2s t, where t is odd. Then 2s | 1n + 2n + + (n 1)n . Now if k
is even and less than n, then 2s | k n . If k is odd and less than n, then
s
by Eulers theorem, k 2 1 1 (mod 2s ), so k n 1 (mod 2s ). Then
0 1n + 2n + + (n 1)n n2 (mod 2s ), which implies 2s+1 | n, a
contradiction. So only odd n > 1 has the property.

141. (1972 Putnam Exam) Show that if n is an integer greater than 1, then
n does not divide 2n 1.

Solution. Suppose n | 2n 1 for some n > 1. Since 2n 1 is odd, so

n is odd. Let p be the smallest prime divisor of n. Then p | 2n 1, so
2n 1 (mod p). By Fermats little theorem, 2p1 1 (mod p). Let k
be the smallest positive integer such that 2k 1 (mod p). Then k | n
(because otherwise n = kq + r with 0 < r < k and 1 2n = (2k )q 2r
2r (mod p), contradicting k being smallest). Similarly k | p 1. So
k | gcd(n, p 1). Now d = gcd(n, p 1) must be 1 since d | n, d p 1
and p is the smallest prime divisor of n. So k = 1 and 2 = 2k
1 (mod p), a contradiction.

112
142. (Proposed by Romania for 1985 IMO) For k 2, let n1 , n2 , . . . , nk be
positive integers such that
   
n2  (2n1 1), n3  (2n2 1), . . . , nk  (2nk1 1), n1  (2nk 1).

Prove that n1 = n2 = = nk = 1.

Solution. Observe that if ni = 1 for some i, then ni+1 will equal 1

and the chain effect causes all of them to be 1. So assume no ni is 1.
Let pk be the smallest prime number dividing nk . Then pk | 2nk1 1.
So 2nk1 1 (mod pk ). Let mk be the smallest positive integer m
such that 2m 1 (mod pk ). Then mk | nk1 and mk | pk 1 by
Fermats little theorem. In particular 1 < mk pk 1 < pk and so
the smallest prime divisor pk1 of nk1 is less than pk . Then we get
the contradiction that pk > pk1 > > p1 > pk .

143. (1998 APMO) Determine the largest of all integer n with theproperty
that n is divisible by all positive integers that are less than 3 n.

Solution. (Due to LauLap Ming) The largest n is 420. Since 420 =

3 4 5 7 and 7 < 3 420 < 8, 420 has the property. Next, if n
has the property and n > 420, then 3, 4, 5, 7 divide n. Hence n
840 > 729 = 93 . Then 5, 7, 8, 9 divide n, so n 5 7 8 9 = 2420 >
2197 = 133 . Then 5, 7, 8, 9, 11, 13 divide n, so n 5 7 8 9 11
13 > 8000 > 193 . Let k be the integer such that 19k < 3 n 19k+1 .
Then 5k , 7k , 9k , 11k , 13k , 16k , 17k , 19k divide n, and we get the following
contradiction

n 5k 7k 9k 11k 13k 16k 17k 19k

= 19k (4 5)k (3 7)k (2 11)k (2 13)k (3 17)k > 193k+3 n.

144. (1997 Ukrainian Math Olympiad) Find the smallest integer n such that
among any n integers (with possible repetitions), there exist 18 integers
whose sum is divisible by 18.

Solution. Taking seventeen 0s and seventeen 1s, we see that the

smallest such integer n cannot be 34 or less. We will show 35 is the

113
answer. Consider the statement among any 2k1 integers, there exist
k of them whose sum is divisible by k. We will first show that if the
statement is true for k = k1 and k2 , then it is true for k = k1 k2 .
Suppose it is true for k = k1 and k2 . Since the case k = k1 is true,
for 2k1 k2 1 integers, we can take out 2k1 1 of them and pick k1 of
them with sum divisible by k1 to form a group. Then return the other
k1 1 integers to the remaining integers and repeat the taking and
picking. Totally we we will get 2k2 1 groups. Since the case k = k2
is true, from the 2k2 1 sums s1 , . . . , s2k2 1 of the groups, considering
the numbers di = si /gcd(k1 , k2 ), we can get k2 of them whose sum is
divisible by k2 . The union of the k2 groups with sum si s consists of
k1 k2 numbers whose sum is then divisible by k1 k2 .
To finish the problem, since 18 = 2 32 , we have to show the
statement is true for k = 2 and 3. Among 2 2 1 = 3 numbers,
there are two odd or two even numbers, their sum is even. Among
2 3 1 = 5 integers, consider (mod 3) of the integers. If 0,1,2 each
appears, then the sum of those three will be 0 (mod 3), otherwise there
are two choices for 5 integers and three of them will be congruent (mod
3), their sum is 0 (mod 3).
Comments. The statement is true for every positive integer k. All we
have to consider is the case k = p is prime. Suppose 2p 1 integers
are given. There are
 
2p 1 (2p 1)(2p 2) (p + 1)
m= =
p (p 1)!

ways in picking p of them. If no p of them have a sum divisible by p,

then consider X
S= (a1 + + ap )p1 ,

where the sum is over all m pickings a1 , . . . , ap . By Fermats little

theorem,
S 1 + + 1 = m 6 0 (mod p).
e
On the other hand, in expansion, the terms ae11 app have exponent
sum e1 + + ep = p 1. Hence the numbers of nonzero exponents ei

114
 in the terms will be positive integers j p 1. Since p j of the ei is
0, the coefficient of the term in the full expansion of S is
    
2p 1 j p1 (2p 1 j) p (p j + 1) p1
= ,
pj e1 , . . . , ep (p j)! e1 , . . . , ep

which is divisible by p. So all coefficients are divisible by p, hence

S 0 (mod p), a contradiction.

Perfect Squares, Perfect Cubes

1 1 1
145. Let a, b, c be positive integers such that + = . If the greatest
a b c
common divisor of a, b, c is 1, then prove that a + b must be a perfect
square.

1 1 1 ac c
Solution. By algebra, + = is equivalent to = . Sup-
a b c c bc
ac c p
pose = = , where p, q are positive integers and gcd(p, q) =
c bc q
a c b c
1. Then = and = by simple algebra. So
p+q q p+q p
a b c
= = .
p(p + q) q(p + q) pq

Now gcd(p, q) = 1 implies gcd(p(p+q), q(p+q), pq) = 1. Since gcd(a, b, c)

= 1, we have a = p(p + q), b = q(p + q) and c = pq. Therefore
a + b = (p + q)2 .

146. (1969 Eotvos-Kurschak

Math Competition) Let n be a positive integer.
Show that if 2 + 2 28n2 + 1 is an integer, then it is a square.

Solution. If 2 + 2 28n2 + 1 = m, an integer, then 4(28n2 + 1) = (m
2)2 . This implies m is even, say m = 2k. So 28n2 = k 2 2k. This implies
k is even, say k = 2j. Then 7n2 = j(j 1). Since gcd(j, j 1) = 1,
either j = 7x2 , j 1 = y 2 or j = x2 , j 1 = 7y 2 . In the former
case, we get 1 y 2 (mod 7), which is impossible. In the latter case,
m = 2k = 4j = 4x2 is a square.

115
147. (1998 Putnam Exam) Prove that, for any integers a, b, c, there exists a
positive integer n such that n3 + an2 + bn + c is not an integer.

Solution. Let P (x) = x3 + ax2 + bx + c and n = |b| + 1. Observe

that P (n) P (n + 2) (mod 2). Suppose both P (n) and P (n + 2) are
perfect squares. Since perfect squares are congruent to 0 or 1 (mod 4),
so P (n) P (n + 2) (mod 4). However, P (n + 2) P (n) = 2n2 + 2b
is not divisible by 4, a contradiction. p
So either P
p(n) or P (n + 2) is
not a perfect square. Therefore, either P (n) or P (n + 2) is not an
integer.

148. (1995 IMO shortlisted problem) Let k be a positive integer. Prove that
there are infinitely many perfect squares of the form n2k 7, where n
is a positive integer.

Solution. It suffices to show there is a sequence of positive integers

ak such that a2k 7 (mod 2k ) and the ak s have no maximum. Let
a1 = a2 = a + 3 = 1. For k 3, suppose a2k 7 (mod 2k ). Then
either a2k 7 (mod 2k+1 ) or a2k 2k 7 (mod 2k+1 ). In the former
case, let ak+1 = ak . In the latter case, let ak+1 = ak + 2k1 . Then since
k 3 and ak is odd,

a2k+1 = a2k + 2k ak + 22k2 a2k + 2k ak a2k + 2k 7 (mod 2k+1 ).

Since a2k 2k 7 for all k, the sequence has no maximum.

a b c
149. Let a, b, c be integers such that + + = 3. Prove that abc is the
b c a
cube of an integer.

Solution. Without loss of generality, we may assume gcd(a, b, c) = 1.

(Otherwise, if d = gcd(a, b, c), then for a0 = a/d, b0 = b/d, c0 = c/d, the
equation still holds for a0 , b0 , c0 and a0 b0 c0 is a cube if and only if abc is a
cube.) Multiplying by abc, we get a new equation a2 c+b2 a+c2 b = 3abc.
If abc = 1, then we are done. Otherwise, let p be a prime di-
visor of abc. Since gcd(a, b, c) = 1, the new equation implies that p
divides exactly two of a, b, c. By symmetry, we may assume p divides

116
 a, b, but not c. Suppose the largest powers of p dividing a, b are m, n,
respectively.
If n < 2m, then n + 1 2m and pn+1 | a2 c, b2 c, 3abc. Hence
pn+1 | c2 b, forcing p | c, a contradiction. If n > 2m, then n 2m+1 and
p2m+1 | c2 b, b2 a, 3abc. Hence pY 2m+1
| a2 c, forcing p | c, a contradiction.
Therefore, n = 2m and abc = p3m is a cube.
p|abc

Diophantine Equations

150. Find all sets of positive integers x, y and z such that x y z and
xy + y z = z x .

Solution. (Due to Cheung Pok Man) Since 31/3 > 41/4 > 51/5 > ,
we have y z z y if y 3. Hence the equation has no solution if y 3.
Since 1 x y, the only possible values for (x, y) are (1, 1), (1, 2) and
(2, 2). These lead to the equations 1+1 = z, 1+2z = z and 4+2z = z 2 .
The third equation has no solution since 2z z 2 for z 4 and (2, 2, 3)
is not a solution to xy + y z = z x . The second equation has no solution
either since 2z > z. The first equation leads to the unique solution
(1, 1, 2).

151. (Due to W. Sierpinski in 1955) Find all positive integral solutions of

3x + 4y = 5z .

Solution. We will show there is exactly one set of solution, namely

x = y = z = 2. To simplify the equation, we consider modulo 3. We
have 1 = 0 + 1y 3x + 4y = 5z (1)z (mod 3). It follows that z
must be even, say z = 2w. Then 3x = 5z 4y = (5w + 2y )(5w 2y ).
Now 5w + 2y and 5w 2y are not both divisible by 3, since their sum
is not divisible by 3. So, 5w + 2y = 3x and 5w 2y = 1. Then,
(1)w + (1)y 0 (mod 3) and (1)w (1)y 1 (mod 3). From
these, we get w is odd and y is even. If y > 2, then 5 5w +2y = 3x 1
or 3 (mod 8), a contradiction. So y = 2. Then 5w 2y = 1 implies
w = 1 and z = 2. Finally, we get x = 2.

117
152. (Due to Euler, also 1985 Moscow Math Olympiad) If n 3, then prove
that 2n can be represented in the form 2n = 7x2 + y 2 with x, y odd
positive integers.

Solution. After working out solutions for the first few cases, a pattern
begins to emerge. If (x, y) is a solution to case n, then the pattern
suggests the following: If (x + y)/2 is odd, then ((x + y)/2, |7x y|/2)
should be a solution for the case n + 1. If (x + y)/2 is even, then
(|x y|/2, (7x + y)/2) should be a solution for the case n + 1. Before we
confirm this, we observe that since (x + y)/2 + |x y|/2 = max(x, y) is
odd, exactly one of (x + y)/2, |x y|/2 is odd. Similarly, exactly one
of (7x + y)/2, |7x y|/2 is odd. Also, if (x, y) is a solution and one of
x, y is odd, then the other is also odd.

Now we confirm the pattern by induction. For the case n = 3,

(x, y) = (1, 1) with (1 + 1)/2 = 1 leads to a solution (1, 3) for case
n = 4. Suppose in case n, we have a solution (x, y). If (x + y)/2 is
x + y
2 |7x y|
2
odd, then 7 + = 14x2 + 2y 2 = 2n+1. If (x + y)/2
2 2
|x y|
2 7x + y
2
is even, then 7 + = 14x2 + 2y 2 = 2n+1 . Therefore,
2 2
the pattern is true for all cases by induction.

153. (1995 IMO shortlisted problem) Find all positive integers x and y such
that x + y 2 + z 3 = xyz, where z is the greatest common divisor of x
and y.

Solution. Suppose (x, y) is a pair of solution. Let x = az, y = bz,

where a, b are positive integers (and gcd(a, b) = 1). The equation
implies a + b2 z + z 2 = abz 2. Hence a = cz for some integer c and
2 2 b2 + z
we have c + b + z = cbz , which gives c = 2 . If z = 1, then
bz 1
b2 + 1 2
c= = b+1+ . It follows that b = 2 or 3, so (x, y) = (5, 2)
b1 b1
16b2 + 32 33
or (5, 3). If z = 2, then 16c = = 4b + 1 + . It follows
4b 1 4b 1
that b = 1 or 3, so (x, y) = (4, 2) or (4, 6).

118
 2 b2 z 2 + z 3 b + z3
In general, cz = 2
= b+ 2 . Now integer cz 2 b =
bz 1 bz 1
3 2
b+z z z+1 z2 z + 1
1 implies b . If z 3, then < z + 1,
bz 2 1 z1 z1
b2 + z z2 + z
so b z. It follows that c = 2 2 < 2, so c = 1. Now b is an
bz 1 z 1
integer solution of w 2 z 2 w + z + 1 = 0. So the discriminant z 4 4z 4
is a square. However, it is between (z 2 1)2 and (z 2 )2 , a contradiction.
Therefore, the only solutions are (x, y) = (4, 2), (4, 6), (5, 2) and (5, 3).

154. Find all positive integral solutions to the equation xy + yz + zx =

xyz + 2.

Solution. By symmetry, we may assume x y z. Dividing both

1 1 1 2
sides by xyz, we get + + = 1 + . So
z y x xyz
2 1 1 1 3
1 < 1+ = + + .
xyz z y x x
Then x = 1 or 2. If x = 1, then the equation implies y = z = 1. If x = 2,
1 1 1 1 1 1 1 1 1 2
then + = + . So < + = + . Then y < 4. Sim-
z y 2 yz 2 2 yz z y y
ple checkings yield y = 3, z = 4. Therefore, the required solutions are
(x, y, z) = (1, 1, 1), (2, 3, 4), (2, 4, 3), (3, 2, 4), (3, 4, 2), (4, 2, 3), (4, 3, 2).

155. Show that if the equation x2 + y 2 + 1 = xyz has positive integral

solutions x, y, z, then z = 3.

Solution. (Due to Chan Kin Hang) Suppose the equation has positive
integral solutions x, y, z with z 6= 3. Then x 6= y (for otherwise 2x2 +1 =
x2 z would give x2 (z 2) = 1 and so x = 1, z = 3). As the equation is
symmetric in x, y, we may assume x > y. Among the positive integral
solutions (x, y, z) with x y and z 6= 3, let (x0 , y0 , z0 ) be a solution
with x0 least possible. Now x2 y0 z0 x + (y02 + 1) = 0 has x0 as
a root. The other root is x1 = y0 z0 x0 = (y02 + 1)/x0 . We have
0 < x1 = (y02 + 1)/x0 (y02 + 1)/(y0 + 1) y0 . Now (y0 , x1 , z0 ) is also
a positive integral solution with y0 x1 and z0 6= 3. However y0 < x0
contradicts x0 being least possible.

119
156. (1995 Czech-Slovak Match) Find all pairs of nonnegative integers x and
y which solve the equation px y p = 1, where p is a given odd prime.

Solution. If (x, y) is a solution, then

px = y p + 1 = (y + 1)(y p1 + y 2 y + 1)
and so y + 1 = pn for some n. If n = 0, then (x, y) = (0, 0) and p may
be arbitrary. Otherwise,
px = (pn 1)p + 1
   
np n(p1) p n(p2) p
=p pp + p + p2n + p pn .
2 p2
Since p is prime, all of the binomial coefficients are divisible by p. Hence
all terms are divisible by pn+1 , and all but the last by pn+2 . Therefore
the highest power of p dividing the right side is pn+1 and so x = n + 1.
We also have
   
np n(p1) p n(p2) p
0= p pp + p + p2n .
2 p2
For p = 3, this gives 0 = 33n
3 32n , which implies n = 1 and
p
(x, y) = (2, 2). For p 5, p2 is not divisible by p2 , so every term
but the lat on the right is divisible by p2n+2 , while the last term is not.
Since 0 is divisible by p2n+2 , this is a contradiction.
Therefore, the only solutions are (x, y) = (0, 0) for all odd prime
p and (x, y) = (2, 2) for p = 3.

157. Find all integer solutions of the system of equations

x+y+z = 3 and x3 + y 3 + z 3 = 3.

Solution. Suppose (x, y, z) is a solution. From the identity

(x + y + z)3 (x3 + y 3 + z 3 ) = 3(x + y)(y + z)(z + x),
we get 8 = (3 z)(3 x)(3 y). Since 6 = (3 z) + (3 x) +
(3 y). Checking the factorization of 8, we see that the solutions are
(1, 1, 1), (5, 4, 4), (4, 5, 4), (4, 4, 5).

120
 Solutions to Combinatorics Problems

Counting Methods

158. (1996 Italian Mathematical Olympiad) Given an alphabet with three

letters a, b, c, find the number of words of n letters which contain an
even number of as.
Solution 1. (Due to Chao Khek Lun and Ng Ka Wing) For a non-
negative even integer 2k n, the number of n letter words with 2k as
n n2k
is C2k 2 . The answer is the sum of these numbers, which can be
simplified to ((2 + 1)n + (2 1)n )/2 using binomial expansion.
Solution 2. (Due to Tam Siu Lung) Let Sn be the number of n letter
words with even number of as and Tn be the number of n letter words
with odd number of as. Then Sn + Tn = 3n . Among the Sn words,
there are Tn1 words ended in a and 2Sn1 words ended in b or c. So
we get Sn = Tn1 + 2Sn1 . Similarly Tn = Sn1 + 2Tn1 . Subtracting
these, we get Sn Tn = Sn1 Tn1 . So Sn Tn = S1 T1 = 2 1 = 1.
Therefore, Sn = (3n + 1)/2.

159. Find the number of n-words from the alphabet A = {0, 1, 2}, if any
two neighbors can differ by at most 1.

Solution. Let xn be the number of n-words satisfying the condition.

So x1 = 3, x2 = 7. Let yn be the number of n-words satisfying the
condition and beginning with 0. (By interchanging 0 and 2, yn is also
the number of n-words satisfying the condition and beginning with
2.) Considering a 0, 1 or 2 in front of an n-word, we get xn+1 =
3xn 2yn and yn+1 = xn yn . Solving for yn in the first equation, then
substituting into the second equation, we get xn+2 2xn+1 xn = 0.
For convenience,
set x0 = x2 2x1 = 1. Since r 2
2r 1 = 0 has
roots
1 2 and x0 = 1, x1 = 3, weget xn = (1 + 2) + (1 2)n ,
n

where = (1+ 2)/2, = (1 2)/2. Therefore, xn = ((1+ 2)n+1 +

(1 2)n+1 )/2.

160. (1995 Romanian Math Olympiad) Let A1 , A2 , . . . , An be points on a

circle. Find the number of possible colorings of these points with p
colors, p 2, such that any two neighboring points have distinct colors.

121
 Solution. Let Cn be the answer for n points. We have C1 = p, C2 =
p(p 1) and C3 = p(p 1)(p 2). For n + 1 points, if A1 and An
have different colors, then A1 , . . . , An can be colored in Cn ways, while
An+1 can be colored in p 2 ways. If A1 and An have the same
color, then A1 , . . . , An can be colored in Cn1 ways and An+1 can
be colored in p 1 ways. So Cn+1 = (p 2)Cn + (p 1)Cn1 for
n 3, which can be written as Cn+1 + Cn = (p 1)(Cn + Cn1 ).
This implies Cn+1 + Cn = (p 1)n2 (C3 + C2 ) = p(p 1)n . Then
Cn = (p 1)n + (1)n (p 1) for n > 3 by induction.

Pigeonhole Principle

161. (1987 Austrian-Polish Math Competition) Does the set {1, 2, . . . , 3000}
contain a subset A consisting of 2000 numbers such that x A implies
2x 6 A?

Solution. Let A0 be the subset of S = {1, 2, . . . , 3000} containing all

numbers of the form 4n k, where n is a nonnegative integer and k is
an odd positive integer. Then no two elements of A0 have ratio 2. A
simple count shows A0 has 1999 elements. Now for each x A0 , form
a set Sx = {x, 2x} S. Note the union of all Sx s contains S. So, by the
pigeonhole principle, any subset of S having more than 1999 elements
must contain a pair in some Sx , hence of ratio 2. So no subset of 2000
numbers in S has the property.

162. (1989 Polish Math Olympiad) Suppose a triangle can be placed inside
a square of unit area in such a way that the center of the square is not
inside the triangle. Show that one side of the triangle has length less
than 1.

Solution. (Due to To Kar Keung) Through the center c of the square,

draw a line L1 parallel to the closest side of the triangle and a second
line L2 perpendicular to L1 at c. The lines L1 and L2 divide the square
into four congruent quadrilaterals. Since c is not inside the triangle,
the triangle can lie in at most two (adjacent) quadrilaterals. By the
pigeonhole principle, two of the vertices of the triangle must belong to
the same quadrilateral. Now the furthest distance between two points

122
 in the quadrilateral is the distance between two opposite vertices, which
is at most 1. So the side of the triangle with two vertices lying in the
same quadrilateral must have length less than 1.

163. The cells of a 7 7 square are colored with two colors. Prove that
there exist at least 21 rectangles with vertices of the same color and
with sides parallel to the sides of the square.

Solution. (Due to Wong Chun Wai) Let the colors be black and white.
For a row, suppose there are k black cells and 7 k white cells. Then
there are C2k + C27k = k 2 7k + 21 9 pairs of cells with the same
color. So there are at least 79 = 63 pairs of cells on the same row with
the same color. Next there are C27 = 21 pairs of columns. So there are
212 = 42 combinations of color and pair of columns. For combination
i = 1 to 42, if there are ji pairs in the same combination, then there
are at least ji 1 rectangles for that combination. Since the sum of
42
X
the ji s is at least 63, so there are at least (ji 1) 63 42 = 21
i=1
such rectangles.

164. For n > 1, let 2n chess pieces be placed at the centers of 2n squares of
an n n chessboard. Show that there are four pieces among them that
formed the vertices of a parallelogram. If 2n is replaced by 2n 1, is
the statement still true in general?

Solution. (Due to Ho Wing Yip) Let m be the number of rows that

have at least 2 pieces. (Then each of the remaining nm rows contains
at most 1 piece.) For each of these m rows, locate the leftmost square
that contains a piece. Record the distances (i.e. number of squares)
between this piece and the other pieces on the same row. The distances
can only be 1, 2, . . . , n 1 because there are n columns.
Since the number of pieces in these m rows altogether is at least
2n (n m) = n + m, there are at least (n + m) m = n distances
recorded altogether for these m rows. By the pigeonhole principle, at
least two of these distances are the same. This implies there are at
least two rows each containing 2 pieces that are of the same distance
apart. These 4 pieces yield a parallelogram.

123
 For the second question, placing 2n 1 pieces on the squares of
the first row and first column shows there are no parallelograms.

165. The set {1, 2, . . . , 49} is partitioned into three subsets. Show that at
least one of the subsets contains three different numbers a, b, c such
that a + b = c.

Solution. By the pigeonhole principle, one of the subsets, say X, must

contain at least 49/3 elements, say x1 < x2 < . . . < x17 . Form the
differences x2 x1 , x3 x1 , . . . , x17 x1 and remove x1 (because a, b, c
are to be different) if it appears on the list. If one of the remaining
differences belongs to X, then we are done.

Otherwise, by the pigeonhole principle again, one of the subsets,

say Y (6= X), must contain at least 15/2 elements from these differences
yj = xij x1 , say y1 < y2 < . . . < y8 . Consider the differences y2
y1 , y3 y1 , . . . , y8 y1 and remove y1 and xi1 if they appear on the
list. If one of these differences belong to Y, then we are done. If one
of them, say yj y1 = xij xi1 (6= xi1 , xij ), belong to X, then let
xi1 , xij , xij xi1 are different elements of X and (xij xi1 ) + xi1 = xij
and we are done.

Thus, we may assume 5 of these differences zk = yjk y1 , belong

to the remainging subset Z and say z1 < z2 < . . . < z5 . Form the
difference z2 z1 , z3 z1 , z4 z1 , z5 z1 and remove z1 , yj1 , xij1 if they
appear on the list. The remaining difference zk z1 = yjk yj1 =
xijk yij1 must belong to one of X, Y or Z. As above, we get three
distinct elements a, b, c in one of X, Y or Z such that a + b = c.

Inclusion-Exclusion Principle

166. Let m n > 0. Find the number of surjective functions from Bm =

{1, 2, . . . , m} to Bn = {1, 2, . . . , n}.

Solution. For i = 1, 2, . . . , n, let Ai be the set of functions f : Bm

124
 Bn such that i 6= f (1), . . . , f (m). By the inclusion-exclusion principle,

|A1 An |
X X X
= |Ai | |Ai Aj | + |Ai A2 A3 |
1in 1i<jn 1i<j<kn
     
n m n m n
= (n 1) (n 2) + (n 3)m .
1 2 3

The number of surjections from Bm to Bn is

Xn  
n
nm |A1 An | = (1)i (n i)m .
i=0
i

167. Let A be a set with 8 elements. Find the maximal number of 3-element
subsets of A, such that the intersection of any two of them is not a 2-
element set.

Solution. Let |S| denote the number of elements in a set S. Let

B1 , , Bn A be such that |Bi | = 3, |Bi Bj | =
6 2 for i, j = 1, . . . , n.
If a A belongs to B1 , . . . , Bk , then |Bi Bj | = 1 for i, j = 1, . . . , k.
Since 8 = |A| |B1 Bk | = 1 + 2k, we get k 3. From this, we
see that every element of A is in at most 3 Bi s. Then 3n 8 3, so
n 8. To show 8 is possible, just consider

B1 = {1, 2, 3}, B2 = {1, 4, 5}, B3 = {1, 6, 7}, B4 = {8, 3, 4},

B5 = {8, 2, 6}, B6 = {8, 5, 7}, B7 = {3, 5, 6}, B8 = {2, 4, 7}.

168. (a) (1999 China Hong Kong Math Olympiad) Students have taken a
test paper in each of n (n 3) subjects. It is known that for any
subject exactly three students get the best score in the subject, and
for any two subjects excatly one student gets the best score in every
one of these two subjects. Determine the smallest n so that the above
conditions imply that exactly one student gets the best score in every
one of the n subjects.

125
(b) (1978 Austrian-Polish Math Competition) There are 1978 clubs.
Each has 40 members. If every two clubs have exactly one common
member, then prove that all 1978 clubs have a common member.

Solution. (a) (Due to Fan Wai Tong) For i = 1, 2, . . . , n, let Si be

the set of students who get the best score in the i-th subject. Suppose
nobody gets the best score in every one of the n subjects. Let x be
one student who is best in most number of subjects, say m (m < n)
subjects. Without loss of generality, suppose x is in S1 , S2 , . . . , Sm .
For i = 1, 2, . . . , m, let Si0 = Si \ {x}. Then the m sets Si0 are pairwise
disjoint and so each shares a (distinct) common member with Sm+1 .
Since Sm+1 has three members, so m 3. This means each student is
best in at most three subjects. By the inclusion-exclusion principle,

|S1 S2 Sn |
X X X
= |Si | |Si Sj | + |Si Sj Sk |
1in 1i<jn 1i<j<kn
 
n
3n + |S1 S2 Sn |,
2

which implies n 7. Therefore, if n 8, then there is at least one

student who get the best score in every one of the n subjects. There is
exactly one such students because only one student gets the best score
in a pair of subjects.
Finally, we give an example of the case n = 7 with nobody best in
all subjects:

S1 = {x1 , x2 , x3 }, S2 = {x1 , x4 , x5 }, S3 = {x2 , x4 , x6 },

S4 = {x3 , x5 , x6 }, S5 = {x1 , x6 , x7 }, S6 = {x2 , x5 , x7 },

S7 = {x3 , x4 , x7 }.

(b) Let n = 1978 and k = 40. Let C1 , C2 , . . . , Cn be the n clubs. For

each member of C1 , form a list of the indices of the other clubs that
this member also belongs to. Since C1 and any other club Ci have
exactly one common member, the k lists of the k members of C1 are

126
 disjoint and together contain all integers from 2 to n. By the pigeonhole
n1
principle, one of the lists, say xs list, will contain at least m = d e
k
numbers. (The notation means m is the least integer greater than or
n1
equal to .)
k
Next we will show this x is a member of all n clubs. Suppose x is
not a member of some club Ci . Then each of the m + 1 clubs that x
belong to will share a different member with Ci (otherwise two of the
m + 1 clubs will share a member y in Ci and also x, a contradiction).
Since Ci has k members, so k m + 1 n1 k
+ 1, which implies
2 2
k k + 1 n. Since k k + 1 = 1561 < n = 1978, this is a
contradiction. So x must be a member of all n clubs.
Comments. It is clear that the two problems are essentially the same.
As the number of members in the sets gets large, the inclusion-exclusion
principle in (a) will be less effective. The argument in part (b) is more
convenient and shows that for n sets, each having k members and each
pair having exactly one common member, if n > k 2 k + 1, then all n
sets have a common member.

Combinatorial Designs

169. (1995 Byelorussian Math Olympiad) In the begining, 65 beetles are

placed at different squares of a 9 9 square board. In each move, every
beetle creeps to a horizontal or vertical adjacent square. If no beetle
makes either two horizontal moves or two vertical moves in succession,
show that after some moves, there will be at least two beetles in the
same square.

Solution. (Due to Cheung Pok Man and Yung Fai) Assign an ordered
pair (a, b) to each square with a, b = 1, 2, . . . , 9. Divide the 81 squares
into 3 types. Type A consists of squares with both a and b odd, type
B consists of squares with both a and b even and type C consists of
the remaining squares. The numbers of squares of the types A, B and
C are 25, 16 and 40, respectively.
Assume no collision occurs. After two successive moves, beetles in
type A squares will be in type B squares. So the number of beetles in

127
 type A squares are at most 16 at any time. Then there are at most 32
beetles in type A or type B squares at any time. Also, after one move,
beetles in type C squares will go to type A or type B squares. So there
are at most 32 beetles in type C squares at any time. Hence there are
at most 64 beetles on the board, a contradiction.

170. (1995 Greek Math Olympiad) Lines l1 , l2 , . . . , lk are on a plane such

that no two are parallel and no three are concurrent. Show that we
can label the C2k intersection points of these lines by the numbers
1, 2, . . . , k 1 so that in each of the lines l1 , l2 , . . . , lk the numbers
1, 2, . . . , k 1 appear exactly once if and only if k is even.

Solution. (Due to Ng Ka Wing) If such labeling exists for an integer

k, then the label 1 must occur once on each line and each point labeled
1 lies on exactly 2 lines. Hence there are k/2 1s, i.e. k is even.

Conversely, if k is even, then the following labeling works: for

1 i < j k 1, give the intersection of lines li and lj the label
i + j 1 when i + j k, the label i + j k when i + j > k. For the
intersection of lines lk and li (i = 1, 2, . . . , k 1), give the label 2i 1
when 2i k, the label 2i k when 2i > k.

Alternatively, we can make use of the symmetry of an odd num-

ber sided regular polygon to construct the labeling as follows: for k
even, consider the k 1 sided regular polygon with the vertices labeled
1, 2, . . . , k 1. For 1 i < j k 1, the perpendicular bisector of
the segment joining vertices i and j passes through a unique vertex,
give the intersection of lines li and lj the label of that vertex. For the
intersection of lines lk and li (i = 1, 2, . . . , k 1), give the label i.

171. (1996 Tournaments of the Towns) In a lottery game, a person must

select six distinct numbers from 1, 2, 3, . . . , 36 to put on a ticket. The
lottery commitee will then draw six distinct numbers randomly from
1, 2, 3, . . . , 36. Any ticket with numbers not containing any of these six
numbers is a winning ticket. Show that there is a scheme of buying
9 tickets guaranteeing at least a winning ticket, but 8 tickets is not
enough to guarantee a winning ticket in general.

128
 Solution. Consider the nine tickets with numbers

(1, 2, 3, 4, 5, 6), (1, 2, 3, 7, 8, 9), (4, 5, 6, 7, 8, 9),

(10, 11, 12, 13, 14, 15), (10, 11, 12, 16, 17, 18), (13, 14, 15, 16, 17, 18),
(19, 20, 21, 22, 23, 24), (25, 26, 27, 28, 29, 30), (31, 32, 33, 34, 35, 36).
For the first three tickets, if they are not winning, then two of the
six numbers drawn must be among 1, 2, . . . , 9. For the next three tick-
ets, if they are not winning, then two of the six sumbers must be
10, 11, . . . , 18. For the last three tickets, if they are not winning, then
three of the six numbers must be among 19, 20, . . . , 36. Since only six
numbers are drawn, at least one of the nine tickets is a winning ticket.
For any eight tickets, if one number appears in three tickets, then
this number and one number from each of the five remaining tickets
may be the six numbers drawn, resulting in no winning tickets.
So of the 48 numbers on the eight tickets, we may assume (at least)
12 appeared exactly 2 times, say they are 1, 2, . . . , 12. Consider the two
tickets with 1 on them. The remaining 10 numbers on them will miss
(at least) one of the numbers 2, 3, . . . , 12, say 12. Now 12 appears in
two other tickets. Then 1, 12 and one number from each of the four
remaining tickets may be the six numbers drawn by the committee,
resulting in no winning tickets.

172. (1995 Byelorussian Math Olympiad) By dividing each side of an equi-

lateral triangle into 6 equal parts, the triangle can be divided into 36
smaller equilateral triangles. A beetle is placed on each vertex of these
triangles at the same time. Then the beetles move along different edges
with the same speed. When they get to a vertex, they must make a
60 or 120 turn. Prove that at some moment two beetles must meet
at some vertex. Is the statement true if 6 is replaced by 5?

Solution. We put coordinates at the vertices

so that (a, b), for 0 b
a
a 6, corresponds to the position of b in the Pascal triangle. First
mark the vertices

(0, 0), (2, 0), (2, 2), (4, 0), (4, 2), (4, 4), (6, 0), (6, 2), (6, 4), (6, 6).

129
 After one move, if no beetles meet, then the 10 beetles at the marked
vertices will move to 10 unmarked vertices and 10 other beetles will
move to the marked vertices. After another move, these 20 beetles will
be at unmarked vertices. Since there are only 18 unmarked vertices,
two of them will meet.
If 6 is replaced by 5, then divide the vertices into groups as follows:

{(0, 0), (1, 0), (1, 1)}, {(2, 0), (3, 0), (3, 1)},

{(2, 1), (3, 2), (3, 3), (2, 2)}, {(4, 0), (5, 0), (5, 1)},
{(4, 1), (5, 2), (5, 3), (4, 2)}, {(4, 3), (5, 4), (5, 5), (4, 4)}.
Let the beetles in each group move in the counterclockwise direction
along the vertices in the group. Then the beetles will not meet at any
moment.

Covering, Convex Hull

173. (1991 Australian Math Olympiad) There are n points given on a plane
such that the area of the triangle formed by every 3 of them is at most
1. Show that the n points lie on or inside some triangle of area at most
4.

Solution. (Due to Lee Tak Wing) Let the n points be P1 , P2 , . . . , Pn .

Suppose 4Pi Pj Pk have the maximum area among all triangles with
vertices from these n points. No Pl can lie on the opposite side of
the line through Pi parallel to Pj Pk as Pj Pk , otherwise 4Pj Pk Pl has
larger area than 4Pi Pj Pk . Similarly, no Pl can lie on the opposite side
of the line through Pj parallel to Pi Pk as Pi Pk or on the opposite side
of the line through Pk parallel to Pi Pj as Pi Pj . Therefore, each of the
n points lie in the interior or on the boundary of the triangle having
Pi , Pj , Pk as midpoints of its sides. Since the area of 4Pi Pj Pk is at
most 1, so the area of this triangle is at most 4.

174. (1969 Putnam Exam) Show that any continuous curve of unit length
can be covered by a closed rectangles of area 1/4.

130
 Solution. Place the curve so that its endpoints lies on the x-axis.
Then take the smallest rectangle with sides parallel to the axes which
covers the curve. Let its horizontal and vertical dimensions be a and
b, respectively. Let P0 and P5 be its endpoints. Let P1 , P2 , P3 , P4 be
the points on the curve, in the order named, which lie one on each of
the four sides of the rectangle. The polygonal line P0 P1 P2 P3 P4 P5 has
length at most one.
The horizontal projections of the segments of this polygonal line
add up to at least a, since the line has points on the left and right
sides of the rectangle. The vertical projections of the segments of this
polygonal line add up to at least 2b, since the endpoints are on the
x-axis and the line also has points on the top and bottom side of the
rectangle.

So the polygonal line has length at least a2 + 4b2 1. By the
AM-GM inequality, 4ab a2 + 4b2 1 and so the area is at most 1/4.

175. (1998 Putnam Exam) Let F be a finite collection of open discs in the
plane whose union covers a set E. Show that there is a pairwise disjoint
subcollection D1 , . . . , Dn in F such that the union of 3D1 , . . . , 3Dn
covers E, where 3D is the disc with the same center as D but having
three times the radius.

Solution. We construct such Di s by the greedy algorithm. Let D1

be a disc of largest radius in F . Suppose D1 , . . . , Dj has been picked.
Then we pick a disc Dj+1 disjoint from each of D1 , . . . , Dj and has
the largest possible radius. Since F is a finite collection, the algorithm
will stop at a final disc Dn . For x in E, suppose x is not in the union
of D1 , . . . , Dn . Then x is in some disc D of radius r in F . Now D is
not one of the Dj s implies it intersects some disc Dj of radius rj r.
By the triangle inequality, the centers is at most r + rj units apart.
Then D is contained in 3Dj . In particular, x is in 3Dj . Therefore, E is
contained in the union of 3D1 , . . . , 3Dn .

176. (1995 IMO) Determine all integers n > 3 for which there exist n points
A1 , A2 , . . . , An in the plane, and real numbers r1 , r2 , . . . , rn satisfying
the following two conditions:

131
 (a) no three of the points A1 , A2 , . . . , An lie on a line;
(b) for each triple i, j, k (1 i < j < k n) the triangle Ai Aj Ak has
area equal to ri + rj + rk .

Solution. (Due to Ho Wing Yip) For n = 4, note A1 = (0, 0), A2 =

(1, 0), A3 = (1, 1), A4 = (0, 1), r1 = r2 = r3 = r4 = 1/6 satisfy the
conditions. Next we will show there are no solutions for n 5. Suppose
the contrary, consider the convex hull of A1 , A2 , A3 , A4 , A5 . (This is the
smallest convex set containing the five points.) There are three cases.
Triangular Case. We may assume the points are named so A1 , A2 , A3
are the vertices of the convex hull, with A4 , A5 inside such that A5
is outside 4A1 A2 A4 and A4 is outside 4A1 A3 A5 . Denote the area of
4XY Z by [XY Z]. We get a contradiction as follows:

[A1 A4 A5 ] + [A1 A2 A3 ] = (r1 + r4 + r5 ) + (r1 + r2 + r3 )

= (r1 + r2 + r4 ) + (r1 + r3 + r5 )
= [A1 A2 A4 ] + [A1 A3 A5 ] < [A1 A2 A3 ].

Pentagonal Case. We may assume r1 = min{r1 , r2 , r3 , r4 , r5 }. Draw

line L through A1 parallel to A3 A4 . Since [A1 A3 A4 ] = r1 + r3 + r4
r2 + r3 + r4 = [A2 A3 A4 ], A2 is on line L or on the half plane of L
opposite A3 , A4 and similarly for A5 . Since A1 , A2 , A5 cannot all be on
L, we get 6 A2 A1 A5 > 180 contradicting convexity.
Quadrilateral Case. We may assume A5 is inside the convex hull. First
obseve that r1 + r3 = R2 + r4 . This is because

(r1 + r2 + r3 ) + (r3 + r4 + r1 ) = (r1 + r2 + r4 ) + (r2 + r4 + r3 )

is the area S of the convex hull. So 2S = 3(r1 + r2 + r3 + r4 ). Also

S = [A1 A2 A5 ] + [A2 A3 A5 ] + [A3 A4 A5 ] + [A4 A1 A5 ]

= 2(r1 + r2 r3 + r4 ) + r5 .

From the last equation, we get r5 = (r1 +r2 +r3 +r4 )/8 = S/12 < 0.
Next observe that A1 , A5 , A3 not collinear implies one side of
6 A1 A5 A3 is less than 180 . Then one of the quadrilaterals A1 A5 A3 A4

132
 or A1 A5 A3 A2 is convex. By the first observation of this case, r1 + r3 =
r5 + ri , where ri = r4 or r2 . Since r1 + r3 = r2 + r4 , we get r5 = r2 or
r4 . Similarly, considering A2 , A5 , A4 not collinear, we also get r5 = r1
or r3 . Therefore, three of the numbers r1 , r2 , r3 , r4 , r5 are negative, but
the area of the corresponding triangle is positive, a contradiction.

177. (1999 IMO) Determine all finite sets S of at least three points in the
plane which satisfy the following condition: for any two distinct points
A and B in S, the perpendicular bisector of the line segment AB is an
axis of symmetry of S.

Solution. Clearly, no three points of such a set is collinear (otherwise

considering the perpendicular bisector of the two furthest points of S
on that line, we will get a contradiction). Let H be the convex hull
of such a set, which is the smallest convex set containing S. Since S is
finite, the boundary of H is a polygon with the vertices P1 , P2 , . . . , Pn
belonging to S. Let Pi = Pj if i j (mod n). For i = 1, 2, . . . , n,
the condition on the set implies Pi is on the perpendicular bisector of
Pi1 Pi+1 . So Pi1 Pi = Pi Pi+1 . Considering the perpendicular bisector
of Pi1 Pi+2 , we see that 6 Pi1 Pi Pi+1 = 6 Pi Pi+1 Pi+2 . So the boundary
of H is a regular polygon.
Next, there cannot be any point P of S inside the regular polygon.
(To see this, assume such a P exists. Place it at the origin and the
furthest point Q of S from P on the positive real axis. Since the origin
P is in the interior of the convex polygon, not all the vertices can lie
on or to the right of the y-axis. So there exists a vertex Pj to the
left of the y-axis. Since the perpendicular bisector of P Q is an axis of
symmetry, the mirror image of Pj will be a point in S further than Q
from P, a contradiction.) So S is the set of vertices of some regular
polygon. Conversely, such a set clearly has the required property.
Comments. The official solution is shorter and goes as follows: Suppose
S = {X1 , . . . , Xn} is such a set. Consider the barycenter of S, which is
the point G such that

OX1 + + OXn
OG = .
n
133
(The case n = 2 yields the midpoint of segment X1 X2 and the case
n = 3 yields the centroid of triangle X1 X2 X3 .) Note the barycenter
does not depend on the origin. To see this, suppose we get a point G0

using another origin O 0 , i.e. O 0 G0 is the average of O 0 Xi for i = 1, . . . , n.

Subtracting the two averages, we get OG O 0 G0 = OO 0 . Adding O 0 G0

to both sides, we get OG = OG0 , so G = G0 .
By the condition on S, after reflection with respect to the perpen-
dicular bisector of every segment Xi Xj , the points of S are permuted
only. So G is unchanged, which implies G is on every such perpen-
dicular bisector. Hence G is equidistant from all Xi s. Therefore, the
Xi s are concyclic. For three consecutive points of S, say Xi , Xj , Xk ,
on the circle, considering the perpendicular bisector of segment Xi Xk ,
we have Xi Xj = Xj Xk . It follows that the points of S are the vertices
of a regular polygon and the converse is clear.

134
 Solutions to Miscellaneous Problems

178. (1995 Russian Math Olympiad) There are n seats at a merry-go-around.

A boy takes n rides. Between each ride, he moves clockwise a certain
number (less than n) of places to a new horse. Each time he moves a
different number of places. Find all n for which the boy ends up riding
each horse.

Solution. The case n = 1 works. If n > 1 is odd, the boys travel

1 + 2 + + (n 1) = n(n 1)/2 places between the first and the last
rides. Since n(n1)/2 is divisible by n, his last ride will repeat the first
horse. If n is even, this is possible by moving forward 1, n 2, 3, n
4, . . . , n1 places corresponding to horses 1, 2, n, 3, n1, . . . , n2 +1.

179. (1995 Israeli Math Olympiad) Two players play a game on an infinite
board that consists of 1 1 squares. Player I chooses a square and
marks it with an O. Then, player II chooses another square and marks
it with X. They play until one of the players marks a row or a column
of 5 consecutive squares, and this player wins the game. If no player
can achieve this, the game is a tie. Show that player II can prevent
player I from winning.

Solution. (Due to Chao Khek Lun) Divide the board into 2 2 blocks.
Then bisect each 2 2 block into two 1 2 tiles so that for every pair
of blocks sharing a common edge, the bisecting segment in one will be
horizontal and the other vertical. Since every five consecutive squares
on the board contains a tile, after player I chose a square, player II
could prevent player I from winning by choosing the other square in
the tile.

180. (1995 USAMO) A calculator is broken so that the only keys that still
work are the sin, cos, tan, sin1 , cos1 , and tan1 buttons. The dis-
play initially shows 0. Given any positive rational number q, show that
pressing some finite sequence of buttons will yield q. Assume that the
calculator does real number calculations with infinite precision. All
functions are in terms of radians.

135
 p
Solution. We will show that all numbers of the form m/n, where
m, n are positive
p integers, can be displayed by induction on k = m + n.
(Since r/s = r 2 /s2 , these include all positive rationals.)

For k = 2, pressing cos will display 1. Suppose the statement

is true for integer less than k. Observe that if x is displayed, then
using the facts = tan1 x implies cos1 (sin ) = (/2) and
tan((/2)
p ) = 1/x. So, we can display 1/x. Therefore, to display
m/n with k = m + n, p we may assume m < n. By the induc-
tion step, np< k implies (n m)/mpcan be displayed. Then p using
= tan1 (n m)/m and cos = m/n, we can display m/n.
This completes the induction.

181. (1977 Eotvos-Kurschak Math Competition) Each of three schools is

attended by exactly n students. Each student has exactly n + 1 ac-
quaintances in the other two schools. Prove that one can pick three
students, one from each school, who know one another. It is assumed
that acquaintance is mutual.

Solution. (Due to Chan Kin Hang) Consider a student who has the
highest number, say k, of acquaintances in another school. Call this
student x, his school X and the k acquaintances in school Y. Since
n+1 > n k, x must have at least one acquaintance, say z, in the third
school Z. Now z has at most k acquaintances in school X and hence z
has at least (n+1)k acquaintances in school Y. Adding the number of
acquaintances of x and z in school Y, we get k + (n + 1) k = n + 1 > n
and so x and z must have a common acquaintance y in school Y.

182. Is there a way to pack 250 1 1 4 bricks into a 10 10 10 box?

Solution. Assign coordinate (x, y, z) to each of the cells, where x, y, z =

0, 1, . . . , 9. Let the cell (x, y, z) be given color x + y + z (mod 4). Note
each 1 1 4 brick contains all 4 colors exactly once. If the packing
is possible, then there are exactly 250 cells of each color. However, a
direct counting shows there are 251 cells of color 0, a contradiction. So
such a packing is impossible.

136
183. Is it possible to write a positive integer into each square of the first
quadrant such that each column and each row contains every positive
integer exactly once?
 
Bn An
Solution. Yes, it is possible. Define A1 = (1) and An+1 = ,
An Bn
where the entries of Bn are those of An plus 2n1 . So

  4 3 2 1
2 1 3 4 1 2
A1 = (1), A2 = , A3 = , ....
1 2 2 1 4 3
1 2 3 4

Note that if every column and every row of An contain 1, 2, . . . , 2n1

exactly once, then every column and every row of Bn will contain
2n1 + 1, . . . , 2n exactly once. So, every column and every row of An+1
will contain 1, 2, . . . , 2n exactly once. Now fill the first quadrant using
the An s.

184. There are n identical cars on a circular track. Among all of them, they
have just enough gas for one car to complete a lap. Show that there is
a car which can complete a lap by collecting gas from the other cars
on its way around the track in the clockwise direction.

Solution. (Due to Chan Kin Hang) The case n = 1 is clear. Suppose

the case n = k is true. For the case n = k + 1, first observe that there
is a car A which can reach the next car B. (If no car can reach the
next car, then the gas for all cars would not be enough for completing
a lap.) Let us empty the gas of B into A and remove B. Then the k
cars left satisfy the condition. So there is a car that can complete a
lap. This same car will also be able to complete the lap collecting gas
from other cars when B is included because when this car gets to car
A, the gas collected from car A will be enough to get it to car B.

185. (1996 Russian Math Olympiad) At the vertices of a cube are written
eight pairwise distinct natural numbers, and on each of its edges is
written the greatest common divisor of the numbers at the endpoints

137
 of the edge. Can the sum of the numbers written at the vertices be the
same as the sum of the numbers written at the edges?

Solution. Observe that if a > b, then gcd(a, b) b and gcd(a, b) a/2.

So 3gcd(a, b) a + b. If the sum of the vertex numbers equals the sum
of the edge numbers, then we will have gcd(a, b) = (a + b)/3 for every
pair of adjacent vertex numbers, which implies a = 2b or b = 2a at the
two ends of every edge. At every vertex, there are 3 adjacent vertices.
The a = 2b or b = 2a condition implies two of these adjacent vertex
numbers must be the same, a contradiction.

186. Can the positive integers be partitioned into infinitely many subsets
such that each subset is obtained from any other subset by adding the
same integer to each element of the other subset?

Solution. Yes. Let A be the set of positive integers whose odd digit
positions (from the right) are zeros. Let B be the set of positive integers
whose even digit positions (from the right) are zeros. Then A and B
are infinite set and the set of positive integers is the union of a + B =
{a + b : b B} as a ranges over the elements of A. (For example,
12345 = 2040 + 10305 2040 + B.)

187. (1995 Russian Math Olympiad) Is it possible to fill in the cells of a

9 9 table with positive integers ranging from 1 to 81 in such a way
that the sum of the elements of every 3 3 square is the same?

Solution. Place 0, 1, 2, 3, 4, 5, 6, 7, 8 on the first, fourth and seventh

rows. Place 3, 4, 5, 6, 7, 8, 0, 1, 2 on the second, fifth and eigth rows.
Place 6, 7, 8, 0, 1, 2, 3, 4, 5 on the third, sixth and ninth rows. Then
every 3 3 square has sum 36. Consider this table and its 90 rotation.
For each cell, fill it with the number 9a + b + 1, where a is the number
in the cell originally and b is the number in the cell after the table is
rotated by 90 . By inspection, 1 to 81 appears exactly once each and
every 3 3 square has sum 9 36 + 36 + 9 = 369.

188. (1991 German Mathematical Olympiad) Show that for every positive
integer n 2, there exists a permutation p1 , p2 , . . . , pn of 1, 2, . . . , n
such that pk+1 divides p1 + p2 + + pk for k = 1, 2, . . . , n 1.

138
 Solution. (The cases n = 2, 3, 4, 5 suggest the following permutations.)
For even n = 2m, consider the permutation

m + 1, 1, m + 2, 2, . . . , m + m, m.

For odd n = 2m + 1, consider the permutation

m + 1, 1, m + 2, 2, . . . , m + m, m, 2m + 1.

If k = 2j 1, (1 j m) then (m + 1) + 1 + + (m + j) = j(m + j). If

k = 2j, (1 j m) then (m + 1) + 1 + + (m + j) + j = j(m + j + 1).

189. Each lattice point of the plane is labeled by a positive integer. Each
of these numbers is the arithmetic mean of its four neighbors (above,
below, left, right). Show that all the numbers are equal.

Solution. Consider the smallest number m labelled at a lattice point.

If the four neighboring numbers are a, b, c, d, then (a + b + c + d)/4 = m
and a, b, c, d m imply a = b = c = d = m. Since any two lattice
points can be connected by horizontal and vertical segments, if one
end is labelled m, then along this path all numbers will equal to m.
Therefore, every number equals m.

190. (1984 Tournament of the Towns) In a party, n boys and n girls are
paired. It is observed that in each pair, the difference in height is less
than 10 cm. Show that the difference in height of the k-th tallest boy
and the k-th tallest girl is also less than 10 cm for k = 1, 2, . . . , n.

Solution. (Due to Andy Liu, University of Alberta, Canada) Let b1

b2 bn be the heights of the boys and g1 g2 gn
be those of the girls. Suppose for some k, |bk gk | 10. In the case
gk bk 10, we have gi bj gk bk 10 for 1 i k and
k j n. Consider the girls of height gi , where 1 i k, and the
boys of height bj , where k j n. By the pigeonhole principle, two
of these n + 1 people must be paired originally. However, for that pair,
gi bj 10 contradicts the hypothesis. (The case bk gk 10 is
handled similarly.) So |bk gk | < 10 for all k.

139
191. (1991 Leningrad Math Olympiad) One may perform the following two
operations on a positive integer:
(a) multiply it by any positive integer and
(b) delete zeros in its decimal representation.
Prove that for every positive integer X, one can perform a sequence of
these operations that will transform X to a one-digit number.

Solution. By the pigeonhole principle, at least two of the X + 1 num-

bers
1, 11, 111, . . . , 111
| {z 1}
X+1 digits

have the same remainder when divided by X. So taking the difference

of two of these numbers, we get a number of the form 11 100 0,
which is a multiple of X. Perform operation (a) on X to get such a
multiple. Then perform operation (b) to delete the zeros (if any). If
the new number has more than one digits, we do the following steps:
(1) multiply by 82 to get a number 911 102, (2)delete the zero and
multiply by 9 to get a number 8200 08, (3) delete the zeros to get
828, (4) now 828 25 = 20700, 27 4 = 108, 18 5 = 90 and delete
zero, we get the single digit 9.

192. (1996 IMO shortlisted problem) Four integers are marked on a circle.
On each step we simultaneously replace each number by the difference
between this number and next number on the circle in a given direction
(that is, the numbers a, b, c, d are replaced by a b, b c, c d, d a).
Is it possible after 1996 such steps to have numbers a, b, c, d such that
the numbers |bc ad|, |ac bd|, |ab cd| are primes?

Solution. (Due to Ng Ka Man and Ng Ka Wing) If the initial numbers

are a = w, b = x, c = y, d = z, then after 4 steps, the numbers will be

a = 2(w 2x + 3y 2z), b = 2(x 2y + 3z 2w),

c = 2(y 2z + 3w 2x), d = 2(z 2w + 3y 2z).

From that point on, a, b, c, d will always be even, so |bc ad|, |ac
bd|, |ab cd| will always be divisible by 4.

140
193. (1989 Nanchang City Math Competition) There are 1989 coins on a
table. Some are placed with the head sides up and some the tail sides
up. A group of 1989 persons will perform the following operations:
the first person is allowed turn over any one coin, the second person is
allowed turn over any two coins, . . . , the k-th person is allowed turn
over any k coins, . . . , the 1989th person is allowed to turn over every
coin. Prove that
(1) no matter which sides of the coins are up initially, the 1989 persons
can come up with a procedure turning all coins the same sides up
at the end of the operations,
(2) in the above procedure, whether the head or the tail sides turned
up at the end will depend on the initial placement of the coins.

Solution. (Due to Chan Kin Hang) (1) The number 1989 may not
be special. So let us replace it by a variable n. The cases n = 1 and
3 are true, but the case n = 2 is false (when both coins are heads
up initially). So we suspect the statement is true for odd n and do
induction on k, where n = 2k 1. The cases k = 1, 2 are true. Suppose
the case k is true. For the case k + 1, we have n = 2k + 1 coins.
First, suppose all coins are the same side up initially. For i =
1, 2, . . . , k, let the i-th person flip any i coins and let the (2k + 1 i)-th
person flips the remaining 2k + 1 i coins. Then each coin is flipped
k + 1 times and at the end all coins will be the same side up.
Next, suppose not all coins are the same sides up initially. Then
there is one coin head up and another tail up. Mark these two coins.
Let the first 2k 1 persons flip the other 2k 1 coins the same side
up by the case k. Then there are exactly 2k coins the same side up
and one coin opposite side up. The 2k-th person flips the 2k coins the
same side up and the 2k + 1-st person flips all coins and this subcase
is solved.
So the k + 1 case is true in either way and the induction step is
complete, in particular, case n = 1989 is true.
(2) If the procedure does not depend on the initial placement, then
in some initial placements of the coins, the coins may be flipped with

141
 all heads up and may also be flipped with all tails up. Reversing the
flippings on the heads up case, we can then go from all coins heads up
to all tails up in 2(1 + 2 + + 1989) flippings. However, for each coin
to go from head up to tail up, each must be flipped an odd number of
times and the 1989 coins must total to an odd number of flippings, a
contradiction.

194. (Proposed by India for 1992 IMO) Show that there exists a convex
polygon of 1992 sides satisfying the following conditions:
(a) its sides are 1, 2, 3, . . . , 1992 in some order;
(b) the polygon is circumscribable about a circle.

Solution. For n = 1, 2, . . . , 1992, define

n1 if n 1, 3 (mod 4)
xn = 3/2 if n 2 (mod 4)

1/2 if n 0 (mod 4)

and an = xn + xn+1 with x1993 = x1 . The sequence an is

1, 2, 4, 3, 5, 6, 8, 7, . . . , 1989, 1990, 1992, 1991.

Consider a circle centered at O with large radius r and wind a polygonal

line A1 A2 A1992 A1993 with length Ai Ai+1 = ai around the circle so
that the segments Ai Ai+1 are tangent to the circle p at some point Pi
with Ai Pi = xi and Pi Ai+1 = xi+1 . Then OA1 = x21 + r 2 = OA1993.
Define
x1 x2 x1992
f (r) = 2 tan1 + 2 tan1 + + 2 tan1
r r r
= (6 A1 OP1 + 6 P1992 OA1993) + 6 P1 OP2 + 6 P2 OP3
+ + 6 P1991 OP1992 .

Now f is continuous, lim f (r) = 1992 and lim f (r) = 0. By the

r0+ r+
intermediate value theorem, there exists r such that f (r) = 2. For
such r, A1993 will coincide with A1 , resulting in the desired polygon.

142
 Comments. The key fact that makes the polygon exists is that there
is a permutation a1 , a2 , . . . , a1992 of 1, 2, . . . , 1992 such that the system
of equations

x1 + x2 = a1 , x2 + x3 = a2 , . . . , x1992 + x1 = a1992

have positive real solutions.

195. There are 13 white, 15 black, 17 red chips on a table. In one step, you
may choose 2 chips of different colors and replace each one by a chip of
the third color. Can all chips become the same color after some steps?

Solution. Write (a, b, c) for a white, b black, c red chips. So from

(a, b, c), in one step, we can get (a + 2, b 1, c 1) or (a 1, b + 2, c 1)
or (a 1, b 1, c + 2). Observe that in all 3 cases, the difference

(a + 2) (b 1), (a 1) (b + 2), (a 1) (b 1) a b (mod 3).

So a b (mod 3) is an invariant. If all chips become the same color,

then at the end, we have (45, 0, 0) or (0, 45, 0) or (0, 0, 45). So a b
0 (mod 3) at the end. However, a b = 13 15 6 0 (mod 3) in the
beginning. So the answer is no.

196. The following operations are permitted with the quadratic polynomial
ax2 + bx + c:
(a) switch a and c,
(b) replace x by x + t, where t is a real number.
By repeating these operations, can you transform x2 x 2 into x2
x 1?

Solution. Consider the discriminant = b2 4ac. After operation (a),

= b2 4ca = b2 4ac. After operation (b), a(x + t)2 + b(x + t) + c =
ax2 + (2at + b)x + (at2 + bt + c) and = (2at + b)2 4a(at2 + bt + c) =
b2 4ac. So is an invariant. For x2 x 2, = 9. For x2 x 1,
= 5. So the answer is no.

197. Five numbers 1, 2, 3, 4, 5 are written on a blackboard. A student may

erase any two of the numbers a and b on the board and write the

143
 numbers a + b and ab replacing them. If this operation is performed re-
peatedly, can the numbers 21, 27, 64, 180, 540 ever appear on the board?

Solution. Observe that the number of multiples of 3 among the five

numbers on the blackboard cannot decrease after each operation. (If
a, b are multiples of 3, then a + b, ab will also be multiples of 3. If one
of them is a multiple of 3, then ab will also be a multiple of 3.) The
number of multiples of 3 can increase in only one way, namely when one
of a or b is 1 (mod 3) and the other is 2 (mod 3), then a+b 0 (mod 3)
and ab 2 (mod 3). Now note there is one multiple of 3 in {1, 2, 3, 4, 5}
and four multiples of 3 in {21, 27, 64, 180, 540}. So when the number of
multiples of 3 increases to four, the fifth number must be 2 (mod 3).
Since 64 1 (mod 3), so 21, 27, 64, 180, 540 can never appear on the
board.

198. Nine 1 1 cells of a 10 10 square are infected. In one unit time, the
cells with at least 2 infected neighbors (having a common side) become
infected. Can the infection spread to the whole square? What if nine
is replaced by ten?

Solution. (Due to Cheung Pok Man) Color the infected cells black and
record the perimeter of the black region at every unit time. If a cell
has four, three, two infected neighbors, then it will become infected
and the perimeter will decrease by 4, 2, 0, respectively, when that cell
is colored black. If a cell has one or no infected neighbors, then it will
not be infected. Observe that the perimeter of the black region cannot
increase. Since in the beginning, the perimeter of the black region is
at most 9 4 = 36, and a 10 10 black region has perimeter 40, the
infection cannot spread to the whole square.
If nine is replaced by ten, then it is possible as the ten diagonal
cells when infected can spread to the whole square.

199. (1997 Colombian Math Olympiad) We play the following game with
an equilateral triangle of n(n + 1)/2 dollar coins (with n coins on each
side). Initially, all of the coins are turned heads up. On each turn, we
may turn over three coins which are mutually adjacent; the goal is to

144
 make all of the coins turned tails up. For which values of n can this be
done?

Solution. This can be done only for all n 0, 2 (mod 3). Below by
a triangle, we will mean three coins which are mutually adjacent. For
n = 2, clearly it can be done and for n = 3, flip each of the four
triangles. For n 0, 2 (mod 3) and n > 3, flip every triangle. Then
the coins at the corners are flipped once. The coins on the sides (not
corners) are flipped three times each. So all these coins will have tails
up. The interior coins are flipped six times each and have heads up.
Since the interior coins have side length n 3, by the induction step,
all of them can be flipped so to have tails up.
Next suppose n 1 (mod 3). Color the heads of each coin red,
white and blue so that adjacent coins have different colors and any
three coins in a row have different colors. Then the coins in the corner
have the same color, say red. A simple count shows that there are
one more red coins than white or blue coins. So the (odd or even)
parities of the red and white coins are different in the beginning. As
we flip the triangles, at each turn, either (a) both red and white coins
increase by 1 or (b) both decrease by 1 or (c) one increases by 1 and
the other decreases by 1. So the parities of the red and white coins
stay different. In the case all coins are tails up, the number of red and
white coins would be zero and the parities would be the same. So this
cannot happen.

200. (1990 Chinese Team Selection Test) Every integer is colored with one
of 100 colors and all 100 colors are used. For intervals [a, b], [c, d] having
integers endpoints and same lengths, if a, c have the same color and
b, d have the same color, then the intervals are colored the same way,
which means a + x and c + x have the same color for x = 0, 1, . . . , b a.
Prove that 1990 and 1990 have different colors.

Solution. We will show that x, y have the same color if and only if
x y (mod 100), which implies 1990 and 1990 have different colors.
Let the colors be 1, 2, . . . , 100 and let f (x) be the color (number)
of x. Since all 100 colors were used, there is an integer mi such that
f (mi ) = i for i = 1, 2, . . . , 100. Let M = min(m1 , m2 , . . . , m100 ) 1002 .

145
 Consider a fixed integer a M and an arbitrary positive integer
n. Since there are 1002 ways of coloring a pair of integers, at least two
of the pairs a + i, a + i + n (i = 0, 1, 2, . . . , 1002 ) are colored the same
way, which means f (a +i1 ) = f (a +i2 ) and f (a +i1 +n) = f (a +i2 +n)
for some integers i1 , i2 such that 0 i1 < i2 1002 . Let d = i2 i1 .
Since there are finitely many combinations of ordered pairs (i1 , d) and
n is arbitrary, there are infinitely many ns, say n1 , n2 , . . . , having the
same i1 s and ds.
Since these nk s may be arbitrarily large, the union of the intervals
[a + i1, a + i1 + nk ] will contain every integer x a + 1002 . So for every
such x, there is an interval [a + i1 , a + i1 + nk ] containing x. Since
f (a + i1 ) = f (a + i1 + d) and f (a + i1 + nk ) = f (a + i1 + d + nk ), so
intervals [a + i1 , a + i1 + nk ], [a + i1 + d, a + i1 + d + nk ] are colored
the same way. In particular, f (x) = f (x + d). So f (x) has period d
when x a + 1002 . Since a M, 100 colors are used for the integers
x a + 1002 and so d 100. Consider the least possible such period
d.
Next, by the pigeonhole principle, two of f (a + 1002 ), f (a + 1002 +
1), . . . , f (a + 1002 + 100) are the same, say f (b) = f (c) with a + 1002
b < c a + 1002 + 100. For every x a + 1002 + 100, choose a large
integer m so that x is in [b, b + md]. Since f (b + md) = f (b) = f (c) =
f (c + md), intervals [b, b + md], [c, c + md] are colored the same way. In
particular, f (x) = f (x + c b). So f (x) has period c b 100 when
x a + 1002 + 100. So the least period of f (x) for x a + 1002 + 100
must be 100. Finally, since a can be as close to as we like, f must
have period 100 on the set of integers. Since all 100 colors are used, no
two of 100 consecutive intgers can have the same color.

146