Andrei Negut - Problems For The Mathematical Olympiads (From The First Team Selection Test To The IMO) - GIL (2005) PDF
Andrei Negut - Problems For The Mathematical Olympiads (From The First Team Selection Test To The IMO) - GIL (2005) PDF
Andrei Negut - Problems For The Mathematical Olympiads (From The First Team Selection Test To The IMO) - GIL (2005) PDF
'
Problems
For the Mathematical
Ol~mpiads
PROBLEMS
FOR THE MATHEMATICAL
OLYMPIADS
ISBN 973-9417-52-3
51(075.35)(079.1)
FOREWORD ....................................................................... 7
I. Problems
Chapter 1. GEOMETRY ........................................................ 9
Chapter 2. NUMBER THEORY ............................................... 17
Chapter 3. COMBINATORICS ................................................ 23
Chapter 4. ALGEBRA ......................................................... 31
II. Solutions
Chapter 1. GEOMETRY ....................................................... 39
Chapter 2. NUMBER THEORY ............................................... 75
Chapter 3. COMBINATORICS ................................................ 95
Chapter 4. ALGEBRA ........................................................ 121
III. APPENDIX 1: USEFUL FACTS ............................................. 145
IV. APPENDIX 2: SOURCES OF PROBLEMS .................................. 158
FOREWORD
This book consists of a number of math problems, all of which are meant primarily as
preparation for competitions such as the International Mathematical Olympiad. They are
therefore of IMO level, and require only elementary notions of math; however, since the
International Mathematical Olympiad is perhaps the most difficult exam in elementary
mathematics, any participant should have with him a good knowledge and grasp of what
he is dealing with. This book is not meant to teach elementary math at an IMO level, but
to help a prospective participant train and enhance his understanding of these concepts.
There are many such collections of problems that are directed at IMO participants.
Yet in my view there are two things which individualize this book: the first of these is
in the selected problems. These are some of the most beautiful problems I encountered
in my four-year preparation for such high-level math competitions, and they have been
presented to me by many great professors and instructors over the years. They are neither
boring nor tedious, but require a certain amount of insight and ingenuity, which I find
to be the necessary quality of any 'beautiful' math problem. Moreover, I have tried not
to add here very well-known problems (such as problems from past IMO's or from other
important contests), as these very likely become known to every student in the first year
of his Olympiad career. Instead, there is a smaller chance that the reader might already
know the problems presented here, and the basis of any good preparation is to work as
many new problems as possible. Any one of these problems could be given at an IMO,
and I hope that this book might help them come to light.
I have branded each problem with one of 3 degrees of difficulty: E for easy, M for
medium and D for difficult (the reader can find the category of a problem at the beginning
of its solution). But these are just relative, as an E problem is of the level of a problem
1 at an IMO, an M problem is similar to what one should expect from problem 2 and
D could be a problem 3. Therefore, a novice in the Olympiad world should not feel
frustrated if he has trouble with an E problem, because on an absolute scale it can be
quite difficult. Such problems are far beyond the level of regular school work.
The second thing that is important about this book is the solutions. Any good
participant at an IMO needs to !mow not so much theory as tricks to be employed in
elementaiy problems. It is far less useful as far as IMO's are concerned (and far more
difficult) for a student to learn multi variable calculus and Lagrange multipliers than to
Foreword
know how to apply geometrical inversion. That is why I emphasize on all these methods,
lemmas and propositions in my solutions, and I have often sacrificed succinctness of a
proof to the educational value of presenting one of these methods. I have also presented
a few of the concepts I have employed throughout the book in an appendix. Thus I can
state that in my.opinion, a potential IMO participant needs two things: ingenuity on one
hand and a firm grasp on all these 'toys' and tricks on the other. Which of them is more
important, I do not know yet, and I can only guess.
I would like to thank the people who invented these wonderful problems. While most
of the solutions are my own work, the problems are not mine and I am in great debt to
their creators. Since these problems were mainly taken from my notes and papers, their
exact origin is unknown to me, and I have replaced the name of their authors by the
symbols ***. It is not a fitting homage, and I apologize for this.
But because I have encountered these problems during my own preparation as an
IMO participant, they have also become a part of me. Each one of them is associated
to the person who showed it to me, the friends who told me their beautiful solutions,
or the contests in which I have or have not solved them. I would like to thank all of
the professors who helped me and who made me into what I am, and though I can't
name them all, it is people like Radu Gologan, Severius Moldoveanu, Dorela Fainisi, Dan
Schwartz, Calin Popescu, Mihai Baluna, Bogdan Enescu, Dinu Serbanescu and Mircea
Becheanu who have shown me the most beautiful and subtle art there is. I will also not
forget all the comrades and friends that have passed through the Olympiad experience
with me, but they will forgive me for not naming them. They know who they are. I also
cannot forget my family, who has stood by me and given me that priceless moral support
which is indifferent of how well or how badly I behaved in the competitions.
I would like to thank Mircea Lascu and the Gil Publishing House for supporting me
and this book on the long journey to publishing, and Prof. Radu Gologan for a great
deal of help and useful advice. I also want to thank Gabriel Kreindler, Andrei Stefanescu,
Andrei Ungureanu and Adrian Zahariuc for providing some of the solutions present in
this book.
I wish you the best of luck in all your endeavors, mathematical or not.
Andrei Negut
Chapter 1
GEOMETRY
PROBLEMS
2 Rp 1
- < - - - < -2
5 - 2aR+bc
***
5. Let us consider an angle of vertex 0 and a circle tangent to in the points A and
B. The parallel through A at 0 B will cut the circle again in C, and the line
OC intersects the circle again in P. Show that the line AP splits the segment
0 B into two equal parts.
***
6. A triangle and a square are circumscribed around the unit square. Prove that
the common area of these two figures is at least 3.4.
All-Soviet Union Olympiad 1986
7. Let ABCD be a convex quadrilateral and M be the intersection of its diagonals.
If 0 1 and 02 are the circumcenters of ABM and CDM, show that 40 1 0 2 >
AB+CD.
***
11
i2 Problems
***
13. Let Ai ... A. be a polygon. Prove that we cu have A.;A.;+1 be a diameter for all
1 ~ i ~ n (where An+l =Ai) iff n is odd.
•••
14. We are given the circles r, f1 and r2 of radii r,r1,r2 respectively, any two of
them exterior and such that r > r 1 and r > r2. The cOmmon exterior tangents
of rand ri intersect in Ai and those of r and r2 intersect in A •. The tangents
from Ai to r 2 and those from A• to r i determine a quadrilateral; prove that
this quadrilateral has an inscribed circle and compute its radius in terms of
17. If ABCDEF is acyclic hexagon and l(A, BDF), l(B, ACE), l(D, ABF), l(E,ABC)
are concurrent (where l(M,XYZ) is the Simson line of M with respect to
XY Z), prove that CD E F is a rectangle.
***
18. Let P be a convex hexagon which can be split up into 27 disjoint parallelograms.
Prove that P has a center of symmetry and that it can be broken up into 21
disjoint parallelograms.
•••
19. Let Ai, Bi, Ci be the midpoints of the sides of the acute-angled triangle ABC.
The six lines which pass through ea.ch of these points and are perpendicular to
the other two sides determine a hexagon. Prove that the area of the hexagon is
half the area of ABC.
77 de ture
20. Let Pi ... Pn be a convex polygon in the plane, such that for any i of j there exists
some k such that LP;P.P; = 60°. Prove that our polygon is just a equilateral
triangle.
***
21. Let Ha and Hb be the feet of the altitudes from A and B in the acute-angled
triangle ABC, and let w., Wb be the feet of the bisectors from A and B. Prove
that the incenter of ABC lies on HaHb iff the circumcenter of ABC lies on
w.wb.
German Olympiad 2002
22. Consider the lines di and d2 concurrent in P. For any 0 on di - {P}, consider
the circle Ci of center 0 and which is tangent to d2, and the circle C2 which is
tangent to di, d2 and C,. Find the locus of the intersection of these two circles
when 0 travels on d,.
***
23. Let AoAiA2 be a triangle, and wi a circle that passes through Ai and A2. For
any k ;::: 1, let us construct Wk tangent to Wk- i and which passes through the
points Ak and Ak+i, where the indices are considered modulo 3 (meaning that
Ao = A, = Ao, Ar = A. = Ai and Ag = Ag = A2)· Prove that the circles w7
and w1 are identical.
***
24. D, E, Fare the points of tangency of the incircle of triangle ABC with he sides
BC, AC and AB respectively. Let X be in the interior of ABC such that the
incircle of XBC touches XB,XC and BC in Z, Y and D respectively. Prove
that EFZ¥ is cyclic.
IM 0 1995 Shortlist
14 Problems
25. The point Plies in the interior of the square ABCD such that PA = 1, PB = 2
and PC= 3. What are the possible values of the side of the square?
***
26. The circle r2is interior to the circle r 1 and tangent to it in the point N. The
points C, S, T lie on r 1such that CS and CT a.re tangent to r2 in the points M
and K respectively. Let U and V be the midpoints of the arcs CS and CT, and
W the second point of intersection of the circles (U MC) and (VCK). Prove
that U CVW is a parallelogram.
***
27. Given is a quadrilateral with an inscribed circle, and the line that splits it up
into two parts of equal areas and perimeters. Prove that the line passes through
the center of the inscribed circle.
***
28. Let X be a point inside the triangle ABC and consider the lines through X
parallel to the sides of the triangle. The parallel to AB intersects CA in M, the
parallel to BC intersects AB in N and the parallel to CA intersects BC in P.
The lines AP, BM and CN split the triangle into 4 triangles and 3 quadrilat-
erals. Prove that the sum of the areas of 3 of the triangles equals the area of
the fourth.
1 ***
29. Prove that a finite number of squares with the sum of their areas 2 fit without
overlap in the unit square.
***
30. Let A be a point outside the circle w and let AB and C be the tangents from
it to the circle (with B and C on w). A line l is tangent to w and it intersects
AB and AC in P and Q respectively. If R is the intersection of BC with the
parallel to AC through P, then prove that as l varies, QR passes through a
fixed point.
MOSP 2004
31. If the angles between the three pairs of opposite sides of a tetrahedron are equal,
prove that these angles are right.
***
32. Let P be the intersection of the diagonals in the convex quadrilateral ABCD
with AB = AC = BD. Let 0 and I be the circumcenter and the incenter of
the triangle ABp. If these two points are different, then prove that OJ_]_ CD.
***
33. Given are 500 points inside a unit square. Show that there exist 12 of them
Ai, A2, ... , Ai2 such that AiA2 + A2Aa + ... + AnA12 < 1.
***
Geometry 15
34. Prove that for any points A,B, P, Q, R in the same plane we have
***
35. We have a set of 5 points in the plane, such that the area of any triangle formed
by these points is at least 2. Prove that there exists such a triangle of area at
least 3.
Romanian TST 2002
36. Let ABCD be a parallelogram and points M, N such that C E (AM) and
D E (BN). The lines NA and NC intersect the lines MB and MD in the
points E, F, G, H. Prove that EFGH is cyclic iff ABCD is a rhombus.
•••
37. Let us consider a convex polygon with n 2': 5 sides. Prove that there exist no
more than n( 2n3- 5) triangles of area 1 with the vertices among the vertices of
the polygon.
Andrei Negut
38. Given is a square of side-length 20, and a set formed by its four vertices and
1999 inside points. Prove that there exist 3 of those points which determine a
triangle of area at most Io.
JBMO 1999
39. In the tetrahedron ABCD we have that AB,CD < 1, AC= BD = 1 and
AD, BC > 1. Prove that the radius of the sphere inscribed in the tetrahedron
.
JS 1ess t han
8v'3 .
•••
40. The points A,B,C,D lie on a circle centered at 0. The lines AB and CD
intersect at M, and the circles (ACM) and (B D M) intersect ar M and N.
Prove that MN .l NO.
•••
Chapter 2
NUMBER THEORY
PROBLEMS
anan-1 +.j(a':,- l)(a~_ 1 -1). Prove that the number 2+~is a perfect
square.
***
2. Let x, y EN such that 3x 2 + x = 4y2 + y. Prove that x - y is a perfect square.
***
3. Let n E N, and consider all the polynomials with coefficients 0,1,2 or 3. How
many of these polynomials satisfy P(2) = n?
All-Soviet Union Olympiad 1986
4. Prove that for every n ;::: 2 there exists a positive number which can be written
as a sum of i squares, for all 2 ~ i :::;, n.
***
5. Let p;::: 5 be a prime, and prove that if 1 + ~ + ... +~=~,then p 4 lap- b.
***
6. Consider four natural numbers a, b, c, d and the set S formed by the points
(x, y) E (0, 1) x (0, 1) which have the property that ax + by and ex+ dy are
integers. If S has 2004 elements and (a, c) = 6, find (b, J).
Bulgarian Olympiad 2004
7. Find all integer numbers m, n such that (5 + 3V2°)m = (3 + 5V2°)".
All-Soviet Union 0 lympiad 1984
8. For every natural number n prove that the least common multiple of the num-
bers 1, 2, ... , n equals the least common multiple of ( ~) , (;), ... , (:) iff n + 1
is prime.
***
9. Solve in N the equation x 2 = y' - 3, for z - 1 non-divisible by 4.
***
10. We will call a lattice point invisible iff the segment which joins it to the origin
contains at least another lattice point. Prove that there are squares of any size
whose inside contains only invisible points.
Romanian TST 2003
19
20 Problems
***
14. Let ai, ... , a2005 be non-negative integers such that for every n the number
al + ... + a~005 is a perfect square. What is the minimum number of zeroes
among the ai's?
Team contest
15. Let p be a prime number and f be an integer polynomial of degree d such that
f(O) = 0, /(1) = 1 and f(n) congruent to 0 or 1 modulo p for every integer n.
Prove that d <". p - 1.
***
16. Find aJl the natural numbers n which have a multiple with the sum of decimal
digits n.
***
17. Consider the lattice in the plane, from which we may cut rectangles, but only by
making cuts along the lines of the lattice. Prove that for every m > 12 E N we
can cut a rectangle of area greater than m out of which we cannot cut further
a rectangle of area m.
All-Soviet Union Olympiad 1985
18. Prove that for an infinity of natural numbers n the numerator of the fraction
1+ 21 + ... + ;;:
1 h as at least two pnme
. f actors.
Russian Olympiad 2002
19. Prove that there exist an infinity of n cu such that S(3n) <". S(3n+l), where
S(a) is the sum of the digits of a in decimal notation. Moreover, there exist an
infinity of m such that S(2m) <". S(2m+').
***
20. Let n be a natural number greater than 1 and p be a prime such that nip - 1
and pln3 - 1. Prove that 4p - 3 is a perfect square.
MOSP 2004
21. Find aJl natural numbers n which satisfy n 213n + 1.
Number Theory 21
***
22. Find the natural numbers n for which
1. Let Ai, ... ,Ak be subsets of {1, 2, ... , n}, each of them with at least ~elements,
n n
and such that for every i of j we have #{A; A;} ~ "4. Prove that
77 de ture
2. Prove that the smallest number of colors required to color the finite regions of
a planar graph such that no two neighboring regions are colored in the same
way is 2 if and only if all the vertices of the graph have even degree.
***
3. There is an ant which walks on the infinite lattice, but it can only walk upwards
or to the right. There is an infinite sequence of points (which the ant knows)
Ann>O such that immediately after the ant takes its i-th step, the point A;
becomes poisoned. Prove that the ant can always find an infinite path which
always avoids stepping on a previously poisoned point.
***
4. A lOx 10 checkerboard has every field colored in one of 15 given colors, such that
each color is used for at least 6 squares. Prove that there are three identically
colored, equally-distanced fields on the same row or column.
IMAR test, 2004
5. What is the greatest number of knights which can be placed on a 5 x 5 chessboard
such that each of them attacks exactly 2 of the others?
***
6. A (2n + 1) x (2n + 1) chessboard is covered by pieces o.f the shapes 8J,
EEl %, and which we can always rotate by 90° and reflect across the
horizontal and vertical axis. Prove that at least 4n + 2 trominoes (3.square
pieces) are used.
***
25
26 Problems
7. Let Pi, ... , Pn be distinct 2-element subsets of {1, 2, ... ,n} such that if P; nP; f
O; then {i,j} = Pk for some k. Prove that every one of the a;'s appears in
exactly 2 of the P;'s.
***
8. The numbers 1, 2, ... , 2005 are divided into 6 disjoint subsets. Prove that there
are numbers a, b, c (not necessarily distinct) belonging to one of the subsets
such that a + b = c.
Romanian-Hungarian Training Camp
9. Given a graph with n vertices, not bipartite and without triangles, prove that
. a vertex wit
t here exists . h degree at most 2n .
5
Bulgarian Olympiad 2004
10. The natural numbers are colored with two colors. Prove that there exists a
sequence of natural numbers k1 < k2 < ... < kn < ... with the property that
the sequence 2k1 < k1 + k2 < 2k2 < k2 + k, < 2k3 < ... is of the same color.
***
11. We have n + 1 natural weights, the sum of which is 2n. We are given a scale
and at every step we place one of the weights (in order from the heaviest to the
lightest weight) in that arm of the scale which has the less weight on it. If at one
step, both arms have the same weight on them, we put the next weight on any
of them. Prove that at the end of the process, the scale will be in equilibrium.
All-Soviet Union Olympiad 1984
12. In the square with vertices A = (0,0), B = (O,n), C = (n,n), D = (n,O)
consider a broken line which goes from A to C by passing only once though all
thelattice points inside or on the frontier of our square. We will color in black
all the interior regions from which we can reach the si~es AB or BC without
crossing the broken line. Show that the black area is ;-.
***
13. The numbers from 1.to n are placed in an arbitrary order. We will perform the
following operation: if the first number in the sequence is k, then we switch the
order if the first k numbers in our sequence (so the first number becomes the
k-th, the second becomes the k - 1-th etc.) Prove that after a finite number of
operations the first number will be 1.
77 de ture
14. Consider a regular hexagon which we will divide into 6 equilateral triangles
via its diagonals. Then, each of these triangles will be further divided into 4
congruent equilateral triangles, thus obtaining a network of 24 triangles. In the
vertices of this network, let us write the numbers 1, 2, ... , 19. We will call a small
triangle with the numbers in its Vertices a < b < c if we can go from a to c via
Combinatorics 27
20. The numbers 1, 2, ... , 2n are divided into two sequences: ai < a2 < ... < an and
b1 > b2 > ... > bn. Prove that la1 - b1I + ... +Ian - bnl = n 2.
All-Soviet Union Olympiad 1985
21. Given are 4 congruent right triangles, and a move consists of selecting a triangle
and splitting it into two others by cutting along the altitude (the one that
corresponds to the right angle). Prove that after a finite number of paths, we
will always have two congruent triangles, regardless of how the cuts were done.
•••
22. Let m, n be non-zero natural numbers such that m > 1 and n > 2m. Find the
greatest natural number k for which there exist disjoint subsets A 1 , .•. ,A. of
the set {l, 2, ... ,n} such that #A; E {m, m- l} for all i.
77 de ture
23. Prove that the set of numbers {l, 2, ... , 2005} can be colored with two colors
such that any of its arithmetic progressions with 18 terms contains both colors.
***
24. Let us consider the lattice in the plane,and in every point of y-coordinate non-
positive we place a checker. At every step, we can make a "jump" by taking
a checker, removing one of its neighbors and placing the checker two units in
the direction of the removed neighbor. Show that none of our checkers will ever
make it to the line y = 5.
•••
25. Can the natural numbers from 0 to 2" - 1 be written on a circle such that the
last n - 1 binary digits 2 of one of them are the first n - 1 binary digits of his
neighbor in the clockwise direction?
Romanian-Hungarian Training Camp
26. Out of the 77-element subsets of {l, 2, ... , 2005}, are there more of them those
with even element-sum or those with odd element-sum?
***
27. We have 3 chess clubs with n players each, such that every player has played at
least n + 1 others from the other two clubs. Prove that there are three players
from different clubs which have all played each other.
•••
28. A good n-path is one that starts from (0, 0) and is composed of n climbs of the
form (x,y)--> (x+l, y+l) and n descents of the form (x, y)--> (x+l,y-1) such
that the path never goes below the x-axis. A return is a sequence of descents
which has ~ climb right before it and which terminates on the x-axis. Prove
that the number of good n - 1-paths equals the number of good n-paths with
no return of even length.
Putnam 2003
Combinatorics 29
29. In a n x n matrix with non-zero natural entries we are allowed to perform the
following operations: multiplying a row by 2 or subtracting 1 from all the entries
of a column. Show that we can obtain the zero matrix after a sequence of such
steps.
•••
30. Let a,b be distinct natural numbers. A set {x,y,z} is called (a,b)-adapted if
{ z - y, y - x} = {a, b}. Prove that the set of natural numbers can be written
as a disjoint union of {a, b}-adapted sets.
77 de ture
31. We have a 100 x 100 board, all of whose fields are either infected or healthy.
Every day, all healthy cells which have at least two infected neighbors become
infected. Find the smallest n such that there is an initial distribution of n
infected fields which will eventually infect the entire board.
***
32. In a graph G, any subgraph G' of order 4 contains a vertex connected by exactly
one other vertex of G'. Find the greatest possible number .of vertices that G
can have.
Team contest, 2004
33. A rectangle may be split up into disjoint rectangles, all of which have an integer
dimension. Prove that the big rectangle has an integer dimension.
***
34. Prove that in a m X n matrix with elements 0 and 1, the maximal number of
l's which are all on different lines and columns equals the minimum number of
lines and columns which will cover all the l's in the matrix.
***
35. All the digits are used in the decimal expression of x, and there is an > 1 such
that there are at most n + 8 different segments of n consecutive digits in the
decimal expression of x. Prove that x is rational.
All-Soviet Union Olympiad 1983
36. Two players play a game in which they alternatively write down numbers be-
tween 1 and 1000 such that tl1ey never write a divisor of a number previously
written. Which one of them has a winning strategy no matter what the other
does?
All-Soviet Union Olympiad 1987
37. Prove that a graph with 2n sides and n 2 + 1 edges contains at least n triangles.
***
Chapter 4
ALGEBRA
PROBLEMS
1. All the roots of the polynomial xn + an_,xn-l + ... + ao are in the interval
n-2
(0, 1), where n 2:: 3. Prove that L kak > 0.
k=O
***
2. Let .>. be a real number such that the inequality
0<v'2002-~<~
b ab
holds for an infinity of palrs (a, b) of natural numbers. Prove that .>. 2:: 5.
77 de ture
3. For any natural number n ~ 4 and non-negative numbers ai, ... ,an with sum
1, prove that
•••
4. If a, b, c are the side-lengths of a triangle, then prove that
(ab
2 -+-+- a c c)
b
> -+-+-+3
b c a -c b a
***
5. For any 0 < ai :5 a2 :5 ... :5 an and 0 < b1 :5 b2 :5 ... :5 bn prove that
***
6. Prove that for any a, b > 0 we have that
33
34 Problems
IMO 1999
8. The positive numbers a, b, c, A, B, C satisfy a+ A= b + B = c + C = k. Prove
that aB + bC + cA,.:; k 2 •
All-Soviet Union Olympiad 1987
9. Consider the natural numbers a, < a2 < ... < an such that 'L....J°"' _.!:_ ,.:; 1. Prove
ai
that for any x > 0 we have that
2
1 ) 1 1
(
La~+x ,.:;2a1(a1-l)+x
***
10. Consider the positive numbers xi, x2, ... , Xn which satisfy L Xi= L ~-Prove
that
"--1-<1
L..- n-1 +xk -
***
11. Given positive numbers x, y, z with xy+yz+zx = 3, prove that for any a, b, c >0
the following relation holds
Lb:c(y+z) 2'. 3
***
12. Find all integer polynomials P for which (x 2 + 6x + 10)P2 (x) - 1 is the square
of an integer polynomial.
Vietnam Olympiad 2002
13. Find the functions f : JR --> JR with the property that for every x, y E JR we have
***
14. Find the largest k such that a, b, c > 0 and kabc > a 3 + b3 + c 3 implies that
a, b, c are the side-lengths of a triangle.
***
15. Prove that if (an)neN is a strictly increasing and unbounded sequence of positive
numbers, then the sequence
is unbounded.
All-Soviet Union Olympiad 1985
Algebra 35
16. Find all the functions f : JR - {O} -+ JR with the property that for all x, y we
have
77 de ture
19. Does there exist a finite set M of non-zero real numbers such that for every nat-
ural n, there exists a polynomial of degree greater than n with all its coefficients
and roots in M?
77 de ture
20. Let P(z) E <C[X] be a monic polynomial with all its roots strictly inside the
unit disk. Prove that there exists a z with lzl = 1 such that IP(z)I 2'. 1.
***
21. Let x, y, z 2'. 0 be real numbers such that x + y + z + xyz = 4. Prove that
for n ~ 1. Prove that there exists k, 1 :-:; k :-:; 2005 such that ak < 0.
77 de ture
25. Consider the real polynomials f and g of degree 2. Prove that if f(x) E Z
implies g(x) E Z, then there are a, b E Z such that g = af + b.
***
26. Given a function f: JR--> JR which satisfies f(xy+x+y) = f(xy) + f(x) + f(y)
for all x, y E JR, prove that f(x + y) = f(x) + f(y) also holds for all x, y.
***
27. Are there functions f : Q+ --> Q+ such that f(xf(y)) = f(x) holds for all
y
x,yEQ+?
***
28. Prove that for any positive numbers ai, ... ,an we have:
(:Ea,)
iES
2
:-:; L
l~i$j$n
(a;+ ... +a3) 2
1. M We can suppose WLOG that b <:: c. Let T be the second meeting point of
line PM with circle (ABC). Let us denote KP= MP= a and PT= (3; we
know that KB= p- c and MC= p- b. Because of the power of the points K
and M with respect to the circle (ABC), we have that
K
I
I
I
I
2 p(2p-b-c) pa
"'= 2 =2
39
40 Solutions
· 2bc- (b2 + c2 - a 2)
2pa=KM2 =2p2 -2p2 cosA=?a=p 2bc =?abc=2p(p-b)(p-c) (1)
But by making the substitution abc = 2p(p- b)(p- c) from (1), we get:
which is true again, by (1). What follows is that N = N', BO the points 0,
N and I are collinear.
2. D We will use inversion to solve the problem. First of all, note that LA :5
360° - 2 · 135° = 90°. From the quadrilaterals ANGE and AMCD we have
that LANG = LAMC = 135° - LA < 135° and therefore B and D will be
situated on the segments AM and AN. Apply inversion around A and denote
the inverses of our points by '.
The points B' and M' will be on the line AB, with M' between A and
B'; similarly, D' and N' will be on AD, with N' between A and D'. The
fact that the angle between AB and BC equals 135° tells us that the angle
between AB' and the circle (AB'C') is equal to the same 135°. This angle will
be LB'C'A, and therefore LB'C'A = 135°; similarly, LD'C'A = 135° and
therefore LB' C' D' = 90°.
Geometry 41
M B'
K'
But the angle LM CD = goo tells us that the angle between the circles
(AC'M') and (AC'D') isgo0 , and that is exactly the angleLAM'C'+LAD'C'.
So we will have LM'C' D' = LAM'C' + LAD'C' +A = goo +A, and simi-
larly LN'C'B' =goo+ A. Because LB'C'D' = 90°, this gives us LM'C'B' =
LN'C'D' =A and therefore LN'C'A = LM'C'A = 135° -LA. And finally, K
(the intersection of the circles (AMN) and (A.ED)) turns into K', the inter-
section of the lines M'N' and B'D'. Let T' = K'B'nAc'.
The angle between AK and KC equals the angle between AK' and the
circle (AK'C'), meaning the angle LK'C'A. To prove that AK l. KC it is
therefore enough to show that AC' l. C' K'.
Menelaos' Theorem for the triangle AB' D' with the transversal K' M' N'
gives us
N'D' sinLA
C 1N 1 sinLC'D'A
N' A sin(135 - LA)
C'N' = sinLC'AD'
sinLC'D'A AC'
sinLC1 AD1 C 1D 1
.
By di vidi·ng t h e first two rel at1ons d l . sinLO'D'A . h h val
an rep acmg sin LC' AD' wit t e ue
we obtained for it from the third equality, we get that
K'B' C'B'
(2)
K'D' C'IY
But in triangle 0' B' D', C' A = C'T' is a bisector, and so we have
C'B' T'B'
(3)
C'D' T'D'
From (2) and (3) we have that the points C', K', T' are all on the same
Apollonius circle of the segment B' D'. But as K', T' E B' D', they are on the
same diameter in that circle, and therefore C' K' J_ C'T' = C' A, exactly what
we needed to prove.
3. E Note that ON J_ CD, as both 0 and N are on the perpendicular bisector
CD. But let KMncD = P and we will have
4. M Let D be the foot of the altitude from A, and D will be on the side BC
since the triangle has only acute angles. Now, by summing up AD+ BD > AB
and CD+AD >AC we get
28 abc
2ha +a> b + c =?ha> p - a=?-;;-> p - a=? 28 > a(p- a)=? 2.R > a(p- a)=?
~=5(a+b+c)>5(a+~)>
4 8 - 8
h 5 ~>3 Sh
- a+ a {::} V ~''ti + a- - a+ a {::}
¢> 100h~ + 25a2 ;:: 64h~ + 9a2 + 48ah 0 ¢> 36h~ + 16a2 ;:: 48ah0
which holds for all positive a and h. because of the AM - GM inequality.
This tells us that (1) is true, and thus so is our conclusion.
5. E Let M be the intersection of AP and OB and note that L.OPM = L.APC =
" - L.OAC = L.AOB. This means that the triangles OMP and AMO are
similar, and therefore
6. D First of all, if we remove the circle from the inside of the triangle, we will
be left with 3 sectors, each of which we be called "the sector of a particular
vertex". The same observation goes for the square, except that we get 4 sectors.
SECTORS
44 Solutions
Let the triangle be ABC and the square MNPQ. We will study two cases:
whether or not there are vertices of the triangle inside the square. If there are
such vertices, then the angle corresponding in the triangle to that vertex must
be non-acute (since it is tangent to the circle and inside the right angle of the
square). So there can only be one such vertex; let it be A, and let M be that
vertex such that A lies in the sector of M.
c
We will now show that N and Q are situated inside our triangle. If N
weren't inside the triangle, then the point C would have to be on the same side
of NP as the circle. But in this case, the second tangents from B and C to the
circle (i.e. those tangents not containing A) could only meet on different sides
of MN as the circle. This is a contradiction since the circle is inscribed inside
the triangle, so N and Q must lie inside the triangle.
But N and Q cannot lie in the sector of A, so one must lay in each of the
sectors of B and C (obviously, they cannot both lie in the same sector or the
square wouldn't be circumscribed around the circle). So the sectors of N and Q
are completely inside the triangle. Thus, the common part of the two polygons
is at least rr (the circle) plus 2(1- ~) (the areas of the sectors of N and Q).
This means that the common part has area at least
11'
2+2>3.4
The second case that must be studied is when none of the A, B, C lie inside
the square. Then, the lines AB, BC, CA break the plane up into 7 parts. Call
the parts that cont sin only one triangle vertex on their boundary of "type 1"
and those that contain exactly two vertices of "type 2". In the three of them of
type 1 there can be no vertices of the square (as then we would have vertices
of the triangle inside the square). Moreover, there can be no two vertices of
Geometry 45
the square in the same region of type 2 (because then the square wouldn't
be circumscribed around the circle anymore). Since there are 4 vertices of the
square, one of them must lie inside the 7th part, which is the triangle itself. Let
M lie inside the triangle, and if there were any other vertex of the square inside
the triangle we would be done (because of the argument in part 1). So we can
safely assume that M is the only vertex of the square inside the triangle.
Now, the three sides of the triangle will each cut one the sectors determined
by the square except for M's sector. Let us estimate the area of such a cut. If
we had a sector (say N's) and a triangle-side passing through it and cutting
the tangents from N in X and Y, what would be the area between XY and the
circle? Well, that area is 1 - ~ - N X ~NY. But since XY is tangent to the
circle1 we have
and so our area would be <-: 1 - ~ - (.,/2 - 1) 2. Thus, our total area of
intersection would be at least three such areas plus the area of the circle plus
the area of M's sector, which adds up to
3rr -3(v2-1)
3- 4 r,; 2 +rr+l- " =4-3·(0.41) 2 =3.49 .. .>3.4
4
Our problem is thus solved.
7. M Let us draw the two lines d1 and d2 through M which are perpendicular
to AC and BD, and let the angle between them (which will equal LAMB) be
x. Then because 01 and 02 are the circumcenters of the triangles ABM and
CD M, it is easy to see that they cannot lie in the shaded region between d1 and
d 2 (as then we would have an impossible inequality of the form L.01M B <-: 90°).
So L.01M02 > x and then what we need to prove becomes
c
B
8. M First of all, we should see what it means for BE to cut AC in half: let their
point of intersection be L and applying the Law of Sines in triangles ALB and
BLC will give us the ratio~=~~·=:~!~· But AE = 2RsinABE and
EC= 2RsinEBC, where R is the radius of the circle (ABCDE). Thus, we
have
AL AB AE
LC BC EC
and for L to be the midpoint of AC we would then need to prove that
AB· AE =CB· CE. But since AC II ED we have EC= AD and AE =CD,
so it would suffice to prove that AB· CD= AD· BC.
B
Let us now turn to just tbe points A, B, C, D. Applying the Law of Sines
in the triangles ABM and BCM give us
~=~=AM·2R
sinAMB sinABD AD
_!!£_=~=CM·~
sinBMC sinCBD CD
Dividing the two yields
AB AM CD AM AB AD
BC = MC . AD ~MC = BC. CD (1)
Now, we have that
Geometry 47
AD CD
AB= CB =?AD-CB= AB-CD
which is exactly what we needed to prove.
9. D We will prove by induction after the number n of points inside the stronger
statement that there is such a broken line uniting these points and whose end-
points are the two endpoints of the hypothenuse. It is true for n = 1, as an easy
computation shows. Now, assume it true for any number of points less than n
and let us prove it for n points. Let the triangle be ABC with A being the right
angle.
Consider the altitude AD which breaks the original triangle into two smaller
ones similar to it and let X1 1 ••• 1 X 11 be our points. Suppose that there are points
on both sides of the altitude; then apply the induction hypothesis for both of
the smaller triangles, and you will have two broken lines CX1X2 ... XkA and
AXk+l···XnB whose total sum of the squares is at most the square of the
hypothenuse. We can replace the segments XkA and AXk+ 1 by the segment
XkXk+l and obtain our broken line; the sum of the squares now becomes even
less, since LXkAXk+I !> 90° =? XkX~+i < AX~ + AX~+i ·
The other case is when one of the two triangles (say ACD) contains none of
the points. Then we will apply the above algorithm for the triangle ABD and
get a broken line BX1 ... XnA whose sum of the squares is at most AB 2. But
XnA 2 < XnC 2 + AC2 so the line BX1... XnC will be valid, as its sum of the
squares is at most AB 2 + AC2 = BC2. If we can't apply the algorithm in the
previous paragraph to ABD, it must be because its own altitude leaves all the
points only one side. In that case, break it into two triangles with its altitude
48 Solutions
and repeat what is said in this paragraph. Eventually, we will get one of the
smaller and smaller triangles to contain more than 0 and less that n points.
And then the problem will be solved.
10. E Let our polygon be A,. .. A,, and we will prove the statement by induction
over n. For n = 3 it is obvious and now we will assume it true for n - 1 and
prove it for n. First of all, if there existed one of the points Ai, ... , An (say A;)
who were the endpoint of at most one segment 1 then consider the other n - 1
points. Because of the induction hypothesis 1 among them there are at most
n - 1 segments, and plus our at most one segment from A; would yield at most
n segments in total.
Now, assume that each point is the endpoint of at least 2 segments and
assume for the purpose of contradiction the we have at least n + 1 segments.
These n + 1 segments have together at least 2n + 2 endpoints, and by the
pigeonhole principle there must be a vertex which is an endpoint to at least
three segments. Let that vertex be Ai and assume that [A1A;], [A1A;] and
[A1Ak] are segments, with i < j < k. But it's easy to see that if we had another
segment out of A; except [A1A;] that segment would not intersect either [Ai.A;]
or [A1 A•l· Contradiction, and thus our assumption is wrong and the problem
is thus solved.
Observe that this tells us that any set of n points has at most n diameters.
This is true because any two diameters of a set must intersect; indeed1 if di-
ameters AB and CD did not intersect, then either ABCD would be a convex
quadrilateral (and then AC+ BD > AB+ CD contradicts the fact that AB
and CD are diameters) or one of A,B,C,D lies inside the triangle formed by
the other three. In this last case, suppose that D lies inside ABC and then note
that at least one of CA or CB is greater than the diameter CD (which is again
impossible).
11. SOLUTION BY ANDREI UNGUREANU
M Let us assume the circles k1 , k2, ks of radius ro as being tangent to the
angles A, B and C respectively; let r1 be the radius of k•. Let Ai, Bi, Ci and
T be the centers of the circles ki, k., ks and k•.
Geometry 49
But the points Ai, B 1 , C1 a.re on the bisectors of the angles A, B and C
respectively, and therefore AA 1 n BB1 n CC1 =I.
Let A2 and C2 be the tangency points of the circles ki and kg with AC.
We have that AiA2 II C1 C2, because both a.re perpendicular with AC, and that
AiA2 = C1 C2 (both a.re radii in the congruent circles ki and kg). So AiA2C2C1
is a rectangle and this implies Ai C, II AC.
Similarly, AiB1 I AB and BiC1 II BC, so that the triangles ABC and
A1B1C1 can be obtained one from the other by a dilation. As I is the inter-
section of tbe lines AA 1 , BB1 and CC1 , it will be the two triangles' center of
dilation. But Tis thecircumcenterof AiB1C1 (as TA1 = TA2 = TAs = ro+r1)
and through the dilation that transforms AiB1C1 into ABC, Twill be trans-
formed into 0. Since the center of dilation is I, we will have that I, T, 0 a.re
collinear.
12. M Let our polygon be AiA2 ... An and let 8; = AiA.+i + A2Ai+2 + ... + AnA.+n
(where An+• means A,). Therefore, d = 82 + ... + 8k if n = 2k + 1 and
d = 82 + ... + 8k-1 + ~ if n = 2k (because if n = 2k, each of the diagonals
A;A;+k is counted twice in 8•)·
Obviously, we have
2d
d=82 + ... +8•<p(2+ ... +k)=?p-<k(k+l)-2= - 2 -lJ -2
lnJ2 ln+
Fbr n = 2k we have
50 Solutions
k 2d
d<p(2+ ... +k-1)+p2=>-p<k(k-1)+k-2= lnJ2 ln+
-2-lJ -2
The left pa.rt of the inequality is trickier to prove. Fbr that, we need the
following
LEMMA 1 If ABCD is a convex quadrilateral, then AC+ BD > AB+ CD.
Let M be the intersection of AC and BD and we will have AM+ BM > AB
and CM+ DM >CD. Summing these up yields the claim of Lemma 1.
LEMMA 2If1<k<n-1 then we have 8k > 81.
This lemma can be proven easily by noting that for every
A;Ai+1Ai+kA.+k+I is convex and thus we can apply Lemma 1 to it, thus yield-
ing A;A;+k + A.+1.A.+k+I > AiA.+1 + Ai+kA.+k+I · Summing these up after all
i gives us the desired 8k > 8,.
+ +. . +
But 2d = 82 83 8n-2, so we have 2d > (n - 3)81. But 81 = p, so
~ >n-3.
p
13. M If n = 2k + 1 is odd, then it is easy to see that by putting n points at even
spaces around a circle and uniting each of them to that point which is k points
away in clockwise direction, we get the desired polygon.
Now conversely, we must show that this is impossible in a polygon with
an even number of vertices. First of all, note than a polygon with n diameters
must have none of its points inside its convex hull; indeed, if it weren't so 1 then
one of its vertices would be inside a triangle formed by three other vertices
(just take the convex hull of the polygon and break it up into triangles). That
vertex (say A1) situated inside a triangle A;A;A1 cannot be an endpoint of
any diameter1 because for any veftex A:i: 1 the vector ~ can be written as
--+ -----+ ~ .
k;A,A;+k;A,,A;+k1A,A1, whereO::; k;,k;,k1 and k;+k;+kt = 1 (barycentric
coordinates). Now, that would mean that A,A1 < k,A,A;+ k;A,,A; + k1A,A,,
so A,A1 cannot be a diameter, because at least one of A,,A;, A,A; and A,A1
would be greater than it.
So a vertex like A 1 situated inside a triangle cannot be an endpoint to any
diameter. Thus, the remaining n - 1 points form all the n diameters, which is
impossible because of the remark at the end of problem 12 (that a set of m
points has at most m diameters). Therefore, in order for it to have n diameters,
our polygon needs to have none of its vertices inside its convex hull, and let the
vertices of the convex hull be B1 B2 ... Bn in this order. Let us now suppose, as
the hypothesis claims, that the n diameters form a cycle. It is easy to see that
any two diameters need to have a common point (otherwise, we could find even
longer segments than the diameters).
Geometry 51
Let B 1 B; be a diameter, and call the two sets of vertices that are split by
this diameter M and N. The other diameters that leave from Bl (say B 1 B,),
respectively B; (say B;By) must both be directed into one of the sets Mor N
(otherwise, these diameters would not meet). Let's assume that they both are
directed into M. All the other n - 3 diameters will now unite a point from M
with a point from N (because each of them must intersect BlB;). So a chain
of n - 3 diameters must leave B, and travel back and forth between M and N,
ultimately reaching By. But n - 3 is odd, so that chain cannot start and end in
M (where B, and By both are). Contradiction.
14. M SOLUTION BY GABRIEL KREINDLER
Let us denote by 0, 01 and 02 the centers of the circles
r, r 1 and r2 respectively; suppose the lines A102 and A201 meet at T.
A,
We will try to prove that T is the center of the inscribed circle in the
quadrilateral determined by the intersections of the tangents from Al and A2
to the circles r2 and r,.
We will find the distance d1 from T to the tangents drawn from Al to circle
d AlT
r2. We have that ;:; = A102 (1), but we know that
02A2 r2
A20 r
Applying the theorem of Menelaos applied in triangle A1002 with the
secant 0 1T A2 we find that
---·-----
y,
vertex which opposes direction d.). We will now look at the regions T' - T.,
T' - T. and T' - T,. Let Ya, z. be the points of T' -T. the closest to AB and
AC, respectively; define similarly Yb, Zb, Ye, Zc.
The 9 points Xa, Xb, Xe, Ya, Yb, Ye, Za, Zb, Zc can be covered by two trans-
lations of T. That is why two of Xa, x.,x, (suppose them Xa and X•) must be
covered by the same translation To. Note that because To covers Xa and Xb we
n
will have that To T' c T, (since To and T, have the same size). That is why
instead of To we can consider Tc, which covers at least as many points (outside
of T' there are no points of our set, so this area isn't worth covering).
So T, is one of the translations. Therefore, the other one (call it T 1 ), must
cover the points X., Y., Z, (which are by definition in T -T,). But any point
of our set inside T' - Tc is covered by Ti, because the points Xe, Ye, Zc a.re
extremal in this set with respect to the directions da,db,dc.
We have thus constructed two translation that covered the points of T, and
T' - T., so all the points are covered. The problem is therefore proven.
17. M We will use an important facts concerning Simson lines, which is proved in
the appendix.
LEMMA If ABCD are points on a circle, the 4 Simson lines of one with
respect to the other three are concurrent in the point determined by the complex
number
a+b+c+d .
(where a, b, c, d are the complex numbers associated to
4
our points).
Now the problem becomes easy. Let the common point of those 4 lines be
X. The two Simson lines l(A, BDF) and l(D, ABF) intersect in the point given
by the complex number
a+b+d+f .
(as the Lemma says). The other two lines
2
·
mtersect · a+b+c+e , and &nee
m · all4rmes are concurrent, t h ese two pomts·
2
must be one and the same. But equating the two yields c + e = d + f, and
so CDEF is a parallelogram. But since it is inscribed in a circle, it must be a
rectangle.
18. D Let us consider a certain side, which we will call So and which we will assume
to be at the "bottom" of the polygon (just imagine the polygon to be a rigid
body sitting on the side So). A number of parallelograms are adjacent to Bo,
and let 8 1 be the set of sides of these parallelograms that are above Bo and
parallel to it. Obviously, all the segments in 8 1 are parallel among themselves
and the sum of their lengths will be at least that of Bo. A number of other
parallelograms will be adjacent to the sides in 8 1 , and let 82 be the set of their
sides which are parallel to those of 8 1 and lie above it. Again, the sum of the
lengths of the segments of 82 will be at least that of 81. Repeat this algorithm,
and when we reach a side of the polygon with a side of a parallelogram, consider
54 Solutions
that side further in the next level S;. Ultimately, we will reach an S; which
comprises only of polygon sides, and because the segments of the S; are all
parallel, they can only lie on one side. If that side is S&, we just proved that
So 11 S& and IB&I <". IBol; but the same reasoning, starting with S& will yield
IBol <". IB&I =? IBol = IB&l-
So we just proved that each side of the polygon has another side, parallel
and equal to it. In particular, this means that the polygon has an even number
of sides A1A2 ... A2k· Let A;Ai+1 be the side parallel to A 1A2 and we will show
that i = k+l. Because the polygon is convex, A2Ao, ... , A.-1A. must be pairwise
parallel to A;+1A.+o, ... , A2kA1 respectively. Therefore, i-2 = 2k-i =? i = k+l.
So the side A 1 A2 will be parallel to that side which is k sides away from it, and
the same remark goes for any one of the sides. Let 0 be the midpoint of A 1A•+i-
Because A1A2A•+1Ak+2 is a parallelogram (equal and parallel opposite sides)
then 0 will also be the midpoint of A2Ak+2· In this way, we can prove that 0
is the midpoint of AoAk+o, A.Ak+4, ... A2kAk-I and thus 0 will be the center
of symmetry.
We proved before that our polygon must be centrally symmetric, and there-
fore it must have 2k sides. Let us prove that the minimum number of paral-
lelograms in which such a 2k sided n-gon can be broken up is G). First, we
will show that this number is necessary. Take any two pairs of parallel sides
a II b and c II d and two sequences of pairwise adjacent parallelograms, one
going from a to b and the other from c to d. Eventually, these two 'strings'of
parallelograms will intersect, and since the parallelograms are all distinct, they
can only intersect in a whole parallelogram. So for any 2 pairs of opposite sides,
there is a parallelogram with sides parallel to them. Since there are ( ~) such
pairs, we have at least so many parallelograms.
A,
Aic+1 Ak
Now, since 27 < G), the above claims tell us that our polygon has at
most 2 · 7 sides. But we just proved that to pave it, there are enough G) = 21
parallelograms.
19. E SOLUTION BY ADRlAN ZAHARIUC
Let Hi,H2,H3 be the orthocenters of the triangles AB1Ci, A1BC1 and
A 1 B 1 C respectively, and let H be that of the original triangle. The hexagon
the problem refers to is A1HsB1H1C1H2. We will have that
But AB1C1 and ABC may be dilated into each other by a factor of 2, and
through this dilation H B 1 C1 will be transformed into HBC by the same factor.
So we will have [H1B1C1] = [H!C] and the corresponding equalities for the
other vertices. Therefore,
[HBC]+ [HAC]+[HAB] _
1 1 1I+
[AH B H CHI [ABC
1s1112= -
4
angles are 60°, so AB= BC= CA. Similarly, if DE is the segment of shortest
length, there will be an F such that DEF is equilateral.
But let us go back to the triangle ABC. Draw lines through A, B and C
parallel to BC, AC and AB respectively, and assume that they meet in A',
B' and C'. It is easy to note that if we had vertices of our polygon outside of
A'B'C' the we would contradict the maximality of AB,BC and CA. Now, if
one the sides of the triangle DEF (say DE) cuts one of the sides of ABC (say
AB), it will have to also cut AC (as there are no points of our convex polygon
inside ABC) and then LDAE > LBAC = 60°. But DE is the shortest side
in the triangle DAE (by assumption), so it will oppose the smallest angle. So
LD AE > 60° is a contradiction.
A'
a b
( a+b+c'a+b+c'a+b+c c ) I · H H .ff . . b . di
· is on a b 1 it its arycentnc coor -
nates can be written as a convex sum of those Ha and Hb, so iff there exists an
a such that
a acosC b
a+b+c =a· o+ (l- a)-b-# l - " = (a+b+ c)cosC
and
b bcosC a
a+b+c = (l- a) ·O+a-a-# a= (a+b+c)cosC
We needn't write the third relation between the barycentric coordinates,
as it is implied by the first two (as the sum of the barycentric coordinates is
always 1). Such an a exists iff the two RHS above add up to 1, so our condition
for I being on H.H, is that
We thus see that the two conditions are the same, so 0 will be on Wa Wb
iff I is on H.H,.
22. E We will show that the locus is formed by two straight lines that pass through
P (without P itself). Indeed, take R to be the radius of the circle with center
0, let T be its tangency point with d,. The circle that is tangent to this circle
will have radius r, center Q, and it will touch d2 in S. Let M be the common
point of the two circles, and let M 1 and M2 be its projections on di and d,.
08 Solutions
MM,=_!!:____
r r+R
And from the right trapezoid OQSP we find
MM= 2rR
2 r+R
I
\
--- ---
24. D We will consider an inversion sround D, which will map the line BC into
itself, and our two circles into lines psrallel to BC. But note that E, Y, D and
are at the same distance from 0, so they are on a circle that obviously passes
through D. The inversion will turn this into a line perpendiculsr to BC (because
the center of the circle, namely C, is on BC) and the same thing will happen
to the circle determined by the points F, Z, D.
A
So if the images of E, F, Y, Z sre E', F', Y', Z' then we just showed that
E'F'llBC, Y'Z'llBC, E'Y' ..L BC and F'Z' ..L BC. Therefore, E'F'Y'Z' is a
rectangle, so the four images are on the same circle. Inverting the figure again
to obtain the original, this will turn into a circle, so our four points E, F, Z, Y
are on the same circle.
25. E The easiest way to solve this problem is to consider the vertices of the squsre
as being determined by the complex numbers A = il, B = l and C = -il. Then
the relations given by the hypothesis sre IP - ill = 1, Ip- ll = 2 and IP+ ill = 3.
These become
60 Solutions
(p-il)(p+il) = 1'*w+il(p-p)=1-12
and similarly w-
l(p + p) = 4 - 12 and w-
il(p - p) = 9 -1 2 • If we let
-p-p = 28, i(p-p) = 2D, then pp= S2+D2 and the above relations become
S2 +D 2 +2lD = 1-12
By summing up the first and the last one, we get 8 2 + D 2 = 5 - 12 and the
first and second equations then give us D = - ~ and S = -ft. Therefore
But 212 is precisely the square of the side of our square. If P is inside the
square, the angle APC will be obtuse, so AC2 ~ 32 + 1 = 10 =? AB2 ~ 5. Of
course, neither of our solutions satisfy this, so we must conclude that there are
no such points P.
26. D Let us consider an inversion a.round C and denote the inverses of the points
in our figure by 1 • The circles (CTS), (CVK) and (CMU) will transform into
the lines T'S', V' K' and M' U', and W' will be the intersection of these last
two lines. The circle (NKM) transforms into a circle tangent to the lines C'T',
C'S' and T'S', but it will be escribed since r 1 and r 2 were initially one inside
the other.
Geometry 61
V'
But U' is ou the line T'S', ou the other side of S' thau T' such that
1 S'U'
CU =SU*> CU' = CU'CS' =} S'U' = CS'
Similarly, V' is ou T'S' such that T'V' =CT'. Now, to prove that CUWV
is a parallelogram, it suffices to show that CV II UW (CU II VW is proven in
the same way). This is equivalent to proving that the angle between the lines
CV aud UW is 0, which in turn iu equivalent to showing that the augle between
line CV' and circle (CU'W') is 0 (since inversion preserves augles). So we need
to prove that CV' is tangent to circle (CU'W'); let us denote CT'= s, CS'= t,
T'S'= c aud CK'= V'N' = U'N' =CM'= s+~+u =p.
BecauseT'C = T'V' and T'N' = T'K' it's easy to see that CN'K'V' is au
isosceles trape.wid, and the same goes for CN' M'U'. Let L' be the intersection
of CW' aud U'V'. Apply the Law of Sines in triaugles V' L'W' aud C K'W' and
we get
L'V' L'W'
sinL'W1K 1 sinL1V 1W 1
p CW'
sin L1W 1K 1 sin V 1K'T'
Dividing these two yields
L'U' L'W' t
p= CW'p-t
and by dividing these two equations we get
L'V' p-t s
L'U' = p- s. t
Now,
Obviously, we cannot have both the above relations at the same time, so
we must have 0 on the line d.
28. M We will show that the area of the middle triangle equals the sum of the aress
of the other 3. Let the barycentric coordinates of X be (cr.,fJ,'Y) and then we
will have ~~ = fJ =? [A~C] = {J, where S denotes the area of the triangle. So
s,
Let s.<+, be the last square we place, so the total height of our layers will
be r1 + rk 1+1 + ... + rk" which we will assume for the purpose of contradiction
to be > 1 (if it were at most 1, we could pack our layers inside the square). But
we will have for all i
rt+2+ ... +r~,+i+I ;::: Tk.:+i+1(rk,+2+ ... +rk,+i+1) > rk.,+i+1(l-rk,+1) 2:: rk,+1+1(l-r1)
(1)
where we used the fact that the r/s form a descending sequence and that
Tk.:+2 + ... + rkHi+l > 1- rk.:+1 (because if this inequality weren't so, we could
put P«+i+l on the same level with P•,+1). But by summing (1) for all i's, we
would get on the LHS the total sum of our squares except for Pi, so
Now, we will have AB= rcot4 and BC= 2rcos4. By letting a,p,q be
the side-lengths of the triangle APQ, 2po be the perimeter of the triangle and
8 the area (r =~),equation (2) becomes
Po
LM =(po - q) (1- ~)
qsmA
= (p 0 -q) (1- po-28
2Sp ) = (po - q)(po -p)
. po
pq.2A 28 . 2A 2r .2A
LM= Posm 2 = posinA sm 2 = sinA sm 2
66 Solutions
which is constant, no matter what points P and Q we pick. All the lines
RQ will therefore pass through the point M.
31. M Let ABC D be our tetrahedron; we will denote by AB· GD the dot product
of the two vectors and by AB · CD the product of the lengths of the two
segments. Let a = DA, b = oB, c = OC, d = oD, where 0 is an arbitrary
point and letting o: be the common angle, we will have
B'C' < B'D'+C'D' '* A:·~C < A:·~D+ Ag·~D '*AD·BC < AB·CD+AC·BC
So the bracket in the RHS of (1) will always be non-zero, so we must have
7r
0:=2·
32. M LEMMA In any triangle with the usual notations, we will have
r
R = cosA+cosB +cosC-1 (1)
It is a well-known trigonometric relation that the RHS is equal to
. A . B. C N b ul. I. h . A j(p-a)(p-b)
4 sm
2 sm 2 sm 2 . ow, ;y m t1pymg toget er sm2 = y ab
and the corresponding expressions for B and C we get
s
4sin~sin~sin£=4~= p =.":._
2 2 2 abcpabcR
48
which yields (1).
Geometry 67
Back to the problem, where we will consider the triangle AB P (denote its
sides and angles by a, b,p, A, B, P), whose largest side is obviously AB. Then,
first note that cos A + cos B ;:: 1, which happens because cos A + cos B 2:
acosB + bcosA = 1. Now, take a Cartesian coordinate system which has AB
p
as its x-axis and the points P, C, D above it. We will compute the absolute
values of the slopes of the lines OJ and CD and show that their product is 1,
which will prove that the lines are perpendicular. The slope of OJ will be
Pl sinB - sin Al
p(cosB+cosA-1)
and it is easy to see that the product of the two slopes is 1, as required.
33. M Consider coordinate axes parallel to the sides of our square. Divide the
square into 22 even-spaced horizontal (horizontal meaning parallel to the x-
axis and vertical will mean parallel to the y-axis) strips of width ~· By the
pigeonhole principle, one of these will contain (at least) 23 points Ai, ... ,A23
and let their x-coordinates be x 1 ::; x 2 ::; ... ::; x 2a and their y-coordinates
y1, ... , Y23· Obviously, since x23 - x1 ::; 1, we will have either x12 - x1 ::; ~ or
1 1
x2s - x12::; 2. Assume that we have x12 - x1::; 2.
Now, because the height of the strip where A1, ... , A12 are located is ~,it
means that IY• - Y;I s; ~ for all i,j, and therefore
A1A2 + ... + A11A12 < lx2 - xii+ ... + lx12 - x11I + IY2 - Y1I+
1 11
+IYa - Y2I + ... + IY12 - Ynl s; 2 + 22 = 1.
34. M We shall first prove two very simple and useful lemmas.
LEMMA 1 If XYZT is a convex quadrilateral, thenXZ +YT> XY +ZT.
This follows easily if we let M be the intersection of the diagonals and sum up
the two inequalities XM + YM > XY and ZM +TM> ZT.
LEMMA 2 If Xis inside the triangle EFG, then EX+ FX < EG + FG.
Indeed, let EX and F X intersect the sides GF and GE in S and T. Then, we
will have EG +GS> ES and SF> FX - SX; by summing them up, we will
get the required inequality.
68 Solutions
The second case is when P and Q are on the opposite side of AB as R is.
In this case, because AB PQ is convex, Q will be inside AB P (or the other way
around), andit will lie in either APR or BPR. Assume it in APR and by doing
the exact same thing as in the previous case, we will get the desired result.
Geometry 69
35. D Let our pentagon be ABCDE. First of all, if the 5 points do not form a
convex polygon, then consider their convex hull. One of the other points will
be inside the hull, so it will be inside one of the triangles with vertices on
the convex hull. So four of our points will be A, B, C, X such that X is inside
the triangle ABC. Thus, the area of ABC is equal to the sum of the areas of
XAB,XBC,XCA, and thus at least 6.
We are left with the case when our points form a convex pentagon. Let
ABC be the triangle of maximum area among those with vertices in our points
and assume [ABC] < 3. Then draw lines through the vertices of ABC parallel
to the opposite sides, and let these lines determine a triangle A' B' C'. Because
of the maximality of the area of ABC, D and E must lie inside the triangle
A' B'C'. Moreover, they cannot lie inside the triangle ABC (as the pentagon is
convex) so they can only lie in one of the three triangular regions A' BC, AB'C
and ABC'. Suppose first of all that they lie in different regions, say E lies in
AB'C and D lies in AC' B.
which is impossible, so E must lie in the little shaded triangle. In the same
wey, D must lie in the shaded region inside the triangle ABC'. But then, since
B' A = C' A = a we will have DE ::; 2a and the distance from A to DE is at
most ~ (since any circle centered at A with the radius ~ always cuts DE for
any D, E in those regions). So
exactly what we needed. The astute reader will notice that the above argu-
ment does not hold when one of the angles B and C is obtuse. Then, a similar
argument can ~asily be formulated to show that the triangle ADE always has
area less than 3 of that of ABC, which as in (1) gives us [ABC] ~ 3 (it is not
hard to show that with the constraints placed on D and E, the area of ADE
is largest when D and E are as close to BC as possible in their little shaded
regions, and even then it will be less than ~[ABC]).
We still have to study the case when Jj and E are in the same region, say
inside A'BC. Then, we will have
Similarly, applying it in the triangle MOB with the line N AF will give us
ME·MF MB 2
(3)
MG·MH= MD 2
Now, if EFGH is cyclic, the LHS becomes 1 and then MB= MD. But
then, MO will be a median in the isosceles triangle M BD and this would imply
MO .L BD, and so ABCD is a rhombus. Conversely, if ABCD is a rhombus,
then MB= MD and (3) will tell us that ME·MF = MG·MH, which implies
that EFGH is cyclic.
37. E Let our polygon be AiA2 ... An and we will consider three categories of seg-
ments A;A;: the edges (of which there are n), the diagonals of the form A;A.+2
(of which there are also n) and the other diagonals (of which we have n(n 2- 5)).
Let T be the number of triangles of area 1, and 3T will be the number
of segments A;A; involved in these triangles. But each segment from the first
category (the edges) cannot be part of more than 2 triangles of area 1 (because
if it were part of three, then the vertices of those at least three triangles would
be at equal distance from our edge and thus on a line, contradicting convexity).
Similarly, any diagonal of the form A;A;+ 2 appears in at most 3 triangles, and
any other diagonal in 4 of them. So this means that
72 Solutions
n(n-5) 2 n(2n-5)
3T~ 2n+3n+4· - -2- = 2n -5n=} T~ - -3- -
3V ha v'3
r~ 481=4<8
which ends our proof.
40. M Let us consider an inversion around M and let us denote by X' the inverse of
any point X. The points A', B', C', D' will still be on a circle and the lines A' B'
and C' D' will pass through M. The circles (ACM) and (BDM) will transform
into the lines A' C' and B' D', which will meet in N'. What we need to prove
is that the angle between the lines MN and 0 N, meaning the angle between
MN' and the circle (MO'N'), is 90". This angle will be the angle MO'N'; the
greatest problem we face is finding O', the image of 0.
Assume that Mis outside the segments [AB] and [CD] (if it lies inside, the
proof is identical). But
sinp sinq
show that sin( a _ p) = sin( a _ q) . But the Law of Sines in triangle A' N' E'
and E' N' B' gives us
1. D For all natural numbers :1; = k 2 l, where l is a square-free number, call l the
non-square part of x. Now, note that as = 2 · 972 - 1; we will try to turn our
recurrence into a simpler one:
(4)
So the product of nm+a + 1 and am + 1 is always a square. Therefore, the
non-square part of am + 1 is the same as the non-square part of am+3 + 1. But
the non-square parts of a 1 +1, a 2 + 1, a 3 + 1 are 2, and therefore by induction
the non-square part of any an + 1 will be 2. So an + 1 = 2k~, and taking the
square root of (4) gives us that
75
76 Solutions
(5)
But now subtract (2) from (3), and we will have the same remark as above
for am - 1, that it has the same non-square part for all m. Therefore, am - 1 =
61;;, for some Im 's (since the non-square part of a 1 - 1, a 2 - 1, as - 1 is 6) and
we will thus have that 61;;, + 2 = 2k;;, =? (km - l)(km + 1) = 31;;,. It is easy
to prove by induction (from the relation (2)) that the an 's are all of the form
4k + 1. Therefore, kn is odd, and thus the greatest common divisor of km - 1
and km + 1 will be 2. Thus, from (km - 1) (km + 1) = 31;;, we get that one of
km - 1 and km + 1 is of the form 2b2 and the other is of the form 6c2 (since the
two numbers have no common divisors except for 2). But from (5) is easy to
prove by induction that km is always of the form 6q + 1, and thus km+ 1 must
be the one of the form 2b 2 •
Therefore, we will have J2 + ,/2 + 2am = ,/2 + 2km = 2b, and this is all
we had to prove.
2. E Let x = y + h and the relation becomes
"? 1 "? 1 ¥1
Pl t;(p- l)!i(p-i) t; i(p-i) =• 0 t; j2 =• 0
# #
where .!:. refers to the reciprocal of x in the field z,. But the num-
x 2
bers 12 ,2 2 , ... ,(p;l) are all the quadratic residues (since any two of
p-1
them are non-congruent and there are - 2 - of them). So the numbers
~, f., . .,(p ~ 1) 2 are also residues, and they will be all the residues. But
1 1 2 x
1 +2+ ... +p-l =p ·-y
where x, y a.re integers with y non-divisible by p. So
~=
p 2 • :'. + .!:_ =} ap - b = p 2x =} ap _ b = p 3 bx (1)
b y p bp y y
But plb, because it is the denominator of the fraction
But the total number of points in S will be this number plus the number
of points of S which are not in any rectangle. But the points which are not
k
contained in any rectangle either have the x coordinate il or the y coordinate
.!__ And it is easy to see that for a point such as this to be in S, it needs to have
m
both coordinates of the given form (so x = ~ and y = ~). But the number
of these points contained in the unit square is 5(m- 1) and we will have that
2004 = 6ma + 5(m- 1) =} 6mj2004 - 5(m- 1).
The fact that m divides 2004 - 5m + 5 tells us that m divides 2009. But
6j2004-5m+5 =} 6lm-1, so m must be of the form 6k+l; the only divisors of
2009 of that form are 1,7 and 49. Then, let a= 6p, c = 6r, b = mq and d =ms
(where (p,r) = (q, s) = 1) and notice that the function f(x,y) = (6x,my) maps
all points from Sn Poo to the set S' of points in the unit square which satisfy
px + qy E N and rx + sy E N. Moreover, the correspondence is a bijection and
so there are a points in 8 1•
But for any a, just pick p = a + 2 and q = r = s = 1 and it is easy to
note that there will be exactly a points in S'. Therefore, going back on our
argument, the numbers a = 6a, c = 6, b = d = m will contain exactly 2004
points in their interior for every one of the values of m we found to work.
Therefore, the possible values of M are 1, 7, 49.
7. E We will prove that (0, 0) is the only solution. Indeed, suppose there were any
other solutions (m, n) and we would need to have m and n of the same sign.
Suppose them positive (as (m, n) solution implies (-m, -n) solution) and we
will have (5 + 3,12:r = (3 + 5v'2)n.
But (5 + 3,12)m = a+ b,12 and (3 + 5,l2)n = c + d,12 for some natural
numbers a, b, c, d. There is equality between the two powers if and only if a = c,
and by expanding the powers with Newton's formula we will have
These must be equal, and it is obvious that 5 does not divide c and 3 does
no divide a. But a = c and therefore 5 cannot divide a, which can only happen
if in the expansion of a will end with the term 5° · 1s'l' (:). This implies that
m is even and similarly n is even. Then, by extracting square root out of the
relation (5 + 3v'2im;,,, (3 + 5v'2)n we get that (m, n) solution implies (~, %)
solution. This means that we can always decrease our solutions by a factor of
80 Solutions
(1)
Obviously, the above sum is finite, because from a point onward the integer
parts will be 0. Now, count a 1 for every multiple of p among the numbers
1, 2, ... ,x; count another 1 for every multiple of rl' among these numbers, count
another 1 for every multiple of p 3 and so on. What we get is exactly the sum
in (1). But the number of times every number in 1, 2, ... , x has been counted is
exactly p's exponent in it. So the sum in (1) will be the sum of the exponents
of pin the representations of 1, 2, ... ,x, which is the exponent of pin x!
Back to the problem. The least common multiple (LCM) of several numbers
is the product after all primes p of the greatest powers of that prime which
divide any of our numbers. So for the two least common multiples to be equal
we would need to have, for all primes p, that the highest power of p (call it
a) which divides the numbers 1, 2, ... , n to equal the highest power of p which
divides (7), . . ,
(:).But for all k we have
(n)
k
nl
= kl(n-k)I
(2)
and let p• be the highest power of p which divides any of the numbers
1, 2, ... , n. According to the lemma, the exponent of p in (2) will be
(3)
We took the sum only up to a because for greater indices the integer parts
above will be 0. But Lb+ cJ- LbJ- LcJ will always be 0 or 1, so the sum in 2
will always be at most a. Therefore, in order for the exponents of p in the two
LCM's to be equal, there must be at least one k such that the sum in (3) is a,
which is saying that all the terms in (3) are 1 for that particular k.
We need to show that this happens for all p if and only if n + 1 is prime.
Suppose that it happens, and let p be a prime divisor of n + 1, so n = pq - 1
for some q. If p is not the only divisor of n + 1, then p :5 n so the exponent
of p in n will be nonzero. Then by assumption, for some k we need to have
l~ J- l~ J- ln; J k = 1. If k = s s<
pq' + with p we would need that
Number Theory 81
l?J-ln+~;p"J-lp~~lJ=l
for all i. The above relation is easy to prove since i :5 a, and we will have
any (i,j), the number p;; will divide both a+ i and b + j. So (a+ i, b + j)
82 Solutions
will be invisible points for all 1 :5 i, j :5 n, and thus the square with vertices
(a+ 1, b + 1), (a+ 1, b + n), (a+ n, b + n), (a+ n, b + 1) will be formed out of
only invisible points.
11. M It is ea.sy to prove by induction that all the terms of the sequence are natural
numbers. But first of all, let us prove that:
LEMMA For any n and k we have an IUnk.
First, let r and s be the roots of the characteristic equation x 2 - ax+ 1 = 0
and we will have r + s = a and rs = 1. The general solution of our second
degree linear recurrence with the initial conditions a 1 = 1 and a 2 = a will be
rn - Sn
an = - - - . But for any n and k we have
r-s
such that 1 =• l'b and sop=• pl'b. If we take l to be the residue of pl' modulo
a we get bl =• p and so p will be of the form ka + lb for some integer k.
Now, assume the lemma true for n - 1 and let us prove it for n. Let d
be the common divisor of ai, ... ,an-1 and because of the induction hypothesis,
any integral number will be uniquely writable as ki ~ + ... + kn-1 and-l, where
k2, ... , kn-1 range over sets of the form {O, 1, ... , ,13-1} (not necessarily the same
,6 for all k;'s). Then, any multiple of d may be written uniquely as k 1a 1 + ... +
kn-1an-l· But d and an are co-prime, and as the base-case n = 2 tells us, every
integer may be written as kd + knan, where kn ranges over {O, 1, ... , d - 1}.
Yet as we said before, any number of the form kd can be written uniquely as
kl al+ ... +kn-lan-1· Therefore, any integer can be written as kl al+ ... +knan,
where k2, ... ,kn range over Sets of the form {O, ... ,a - 1} for some values of a.
Therefore, the lemma is true. But what it tells us is that there to ev-
ery integer kia1 + ... + kndn may be associated the unique point in the lat-
tice (ki, k,, ... ,kn), where ki ranges over the whole of Z and k; ranges over
{O, 1, ... ,a; - 1} for i > 1. So to have a function <P as required by the problem,
it is enough to construct it such that the points associated to <P(k) and ,P(k-1)
are always adjacent in the lattice in IR".
But such a function ma:Y be constructed rather easily. The set of our points
(ki, ... ,kn) with k; E {O, 1, ... ,a; -1} for i > 1 is a box inn dimensions which is
infinite in one direction (the direction of the coordinate k1). It is the analogue
in more dimensions of an infinite strip in IR.2 or of an infinitely long cube in IR.3 .
And it is not hard to place the points of this infinite box in a sequence which
is one-to-one with Z such that two consecutive points are adjacent.
Indeed, it can be easily proven by induction over q that the lattice points
of a box of the form (Ii, ... ,1.) with 4 ranging over some {O, 1, ... ,,13; -1} can be
put in a sequence such that two consecutive poiµ.ts are adjacent in the lattice.
Then, to construct our infinite sequence in our infinite box (k1, ... ,kn) just stick
together the sequences obtained in the finite boxes of the form (x, k,, ... ,kn) for
fixed x. That yields a sequence of the points in our infinite box such that two
consecutive points are adjacent.
13. M First of all, it is easy to prove by induction that every '"" is natural and
square-free. But then, let p 1 < p 2 < ... be the primes, and associate to every
n a number f(n) whose binary representation is this: f(n) has a 1 as its k-th
digit from the right if and only if Pkl'"n· Thus, f(n) completely determines x.,
as it tells us which primes divide it.
But the recurrence is nothing morethanf(n+l) = f(n)+l, and as f(O) = 0
(xo = 1 has no prime divisors) then !(2005) = 2005 = 11111010101. x20os must
be therefore be
84 Solutions
P1PaPsP1PsPoP10P11 = 2 · 5 · 11 · 17 · 19 · 23 · 29 · 31 = 734653810
14. M Let us assume that the numbers ai, ... , ak are non-zero, and then for any
ar
n we will have + ... + ak a perfect square. But take any prime p greater
than all the a3's and let di, ... , dk be the orders of ai, ... , ak modulo p. Taken a
multiple of di, ... , dk and then the perfect square ar + ... +a~ will be congruent
to 1 + 1 + ... + 1 = k modulo p. But this means that k is a quadratic residue
modulo any prime p greater than all the a;'s.
We will show that this implies that k is a perfect square; indeed, assume it
is not, and then it will be writable as l 2 2eq 1 9_2 ..• qm where q1 , ... , qm are distinct
primes different from 2, and e is 0 or 1. But considering the Legendre symbol
gives us:
(~)·(~)···(q;)=l
for all primes p large enough. A pplying quadratic reciprocity yields
But Dirichlet's Theorem tells us that in any arithmetic progression with the
ratio co-prime with the first term we can find an infinity of prime numbers. So
just consider the progression 8q1 q2 .. .qm · n + 2rq2 .. .qm + 1 where r is chosen such
that 2rq2 ... qm+l is a non-residue modulo qi (we can always do that, as 2q2 ... qm
and q1 are co-prime). But in this arithmetic progression we will find an infinity
of primes p, and for those primes all the Legendre symbols in (1) will be 1,
except for that of qi, which will be -1. We thus find a contradiction, so k must
be a perfect square. Therefore, the number of zeroes among a 1 , ... , a 20o5 will be
2005 - k, and since k must be a perfect square, 2005 - k cannot be less than
2005 - 442 = 69. Of course, this minimum number of zeroes can be attained,
for example for ai = a2 = ... = aag = 0 and a10 = an = ... = a20os = 1.
We can also sidestep using Dirichlet's Theorem by obtaining a contradiction
in this way: just pick any b of the form 8q1 q2 ... qm · n + 2rq2 .. .qm + 1 and by
taking its prime divisors to be pi, ... ,pm we will see that it is impossible for
k = l 2 2eq1 ... qm to be a quadratic residue modulo all the p/s. Of course, in
choosing b we must be careful not to have any of the divisors of the ai 's among
its prime divisors, but this can easily be done by using the Chinese Remainder
Lemma.
15. M LEMMA For p prime and for all i < p - 1 we have that o' + 1i + 2' + ... +
(p-1)' =.o.
Number Theory 85
a,
00... 0ak ... 00... 0ak ... a, ... 00 ... 0ak ... a,
where the sequence ak ... a1 is repeated p times, and there are d - k zeroes
between any two such sequences, has the sum of the digits ap. But this number
is equal to
ak ... a1(1...0100... 0l00... 01) = a(l + lOd + 102d + ... + 10<•-l)d) (1)
where the number 1...0100... 0100 ... 01 hasp l's and d - 1 zeroes between
any two l's. But the RHS of (1) is divisible by ap, since lQd 1=}1 + lQd + =•
l0 2d + ... + 10(P-l)d =•
0. So we have found a multiple of ap with the sum of
the digits ap.
But since every numbers which is co-prime with 10 can be written as a
product of primes, the above claim tells us that it is writable in such a way
(proof by induction, because if one of its divisors is writable, it too will be).
But now take an n = 2"5'm with (m, 10) = 1 and consider c > a, b. Then, let
86 Solutions
1 + ... + ~,
but it won't appear in the numerator (it will appear in all terms
n .
in the numerator but one, so it won't divide the numerator). So q will divide
the reduced denominator of the fraction 1 + ... + ~, and thus that reduced
n n
denominator will be greater than 2 . But since the entire expression is > 1, the
numerator (which is a power of p) will be greater than i·
As n grows to infinity,
the exponent of p in the numerator will do the same.
Back to the problem. Assume that there are only finitely many such num-
bers, and therefore for all n ;::: some N we will have that 1 + ~ + ... + ~ has
only one prime factor in the numerator. Then, for any prime p > N we will
Number Theory 87
prove by induction over k that the unique prime factor in the numerator of
1 1
Ek = 1 + 2 + ... + P" _ 1 is p.
First of all, remark that the sum of rationals whose numerators in reduced
form are divisible by p also has the numerator divisible by p. For k = 1, the
1
expression Ek can be written as ( 1+p~ ) + (2+ p ~ 2 ) + ... ,which is a
sum of numbers whose numerators are divisible by p. The numerator of the
sum will .also be divisible by p, and thus we have proved the base case of our
induction. Now, assume the problem true for k - 1 and let us prove it for k.
First, if !£-
is the reduced form of 1 + ~ + ... +pk-~_ 1 , we can infer that
a> 1 by an argument similar to that in the Lemma (invoking Chebyshev's The-
1 1
orem). But now, note that for all m the sums Sm = mp + 1 + ... +mp + p _ 1 all
have the numerator divisible by p (for the same reason for which the numerator
of E(l) is divisible by p). Moreover, the sum~+~+ ... + p(p•-~ _ l) also has
the numerator divisible by p (because of the induction hypothesis); summing
up this expression with So, S 1 , ... ,S,•-1_ 1 will yield the desired result.
1 1
So now we have that S( k) = 1 + 2 + ... + pk _ 1 has the numerator a power
for some l. It is easy to see why p" cannot divide this for any n and with k
large enough.
19. M First of all, let us show that there are an infinity of n such that S(3") ;:::
S(3"+ 1 ). Assume that it is not so, and that for n ;::: N we have S(3") <
S(3"+ 1 ); but S(3") is always a multiple of 9 (well, for n;::: 2) and therefore
we will have S(3"+1) ;::: S(3") + 9 for n ;::: N and by iterating this we get
S(3") ;::: S(3N) + 9(n- N). But the number 3" has at most 1+log10 3" digits,
so S(3") ~ 9(1 + nlog 10 3). Of course, this would imply that for all n
88 Solutions
1 g<•-l)k
1• +2k + ... + (p-1)• = 1+ g•+ g2k + ... + g<•-2)k = -=---1 k
-g
which is 0 if the denominator is non-zero (meaning if p - 1 does not divide
k, asp-1 is the order of g) and -1 otherwise. Therefore, since in (1) we cannot
have that sum equal to 0, we must have p - 1Ik and then k =•
-1 =? Pl k + 1.
This means that n is square free, since p cannot divide k.
Let n = PlP2···Pm with Pl < p2 < ... < Pm and we will use p - ll k for
every one of its prime divisors. We will have p1 - 1IP2··Pm, but since P2···Pm
has no prime factors less than Pl we must have p 1 - 1 = 1 =? p 1 = 2. Then,
p2 - ll2P3···Pm, and P2 - 1 can ouly have prime factors with 2 and since P2 > 2
we need that p2 = 3. Furthermore p3 - ll6P4···Pm, and p3 - 116 which implies
p3 = 7 and finally p4 - ll42p5 ...pm =? p4 - 1142 =? p4 = 43. We cannot have a
p5, since there are no prime values of p5 > 43 which satisfy p5 - 1142 · 43.
So the above argument tells us that the only possible values of n are
1, 2, 6, 42, 42 · 43. And each of these works, since every one of the primes that
comprise them divide the corresponding expressions.
23. M We will determine all natural k's such that there is a solution (a,b) to
a2 + b2 +ab = k
ab-1
* a2 + b2 - (k - l)ab = -k (1)
Take the solution with the smallest value of a + b, and we may assume
b S a. If (a, b) is a solution, then so is ((k- l)b- a, b), as it is easy to note from
(1). But if (a, b) verifies (1), then b2 = -k + a(b(k - 1) - a) < a(b(k - 1) - a),
so b(k - 1) - a > 0. But then b(k - 1) - a+ b ~ a+ b (from the minimality of
the solution (a, b)) so we will have that b(k - 1) ~ 2a. But note that
90 Solutions
(k - l)a - Ja 2 (k2 - 2k - 3) - 4k 2a
b= 2 ::':k=l=?
k =•
l. This means that k = pat + l, where a is a positive integer and t is not
divisible by p.
So we have, for each n,
CASE 1: p > 2.
We will now show that the RHS has the exponent of p equal to a+ /3, where
j3 is the exponent of p in n.
Indeed, let us look at the exponent of p in the term paitiin-i ( 7). The
~i ·(n -i-
1
1 ) and the number ~i has the exponent at least j3 - i +2 (easy to
see). We must concur that the exp~~ent of the term paitiln-i ( 7) is at least
ai+/3-i+2, which is greater than a+/3 for i <: 3. Also, the termp2at2 l"- 2 ( ; )
has the exponent at least 2a+ j3. We see thus that all the terms in the expansion
(1) have exponents greater than a+/3 except for the first term which has exactly
a+ j3 (remember that t and l are non-divisible by p).
This means that p"lpa+~j, where j is non-divisible by p. This means that
n S a+j3 Sa+ log. n. This is impossible for sufficiently large n. Contradiction.
CASE 2: p= 2
We have that 2" lk" - l" for an infinity of n and with k and l odd. Thus,
we have 2"l(kl- 1 )" - 1(where1- 1 is the reciprocal of l modulo 2") and let e
be the order of (kl- 1 ) modulo 2". Since •11"(2") = 2n-1, e must be of the form
2d. But eln also.
Now, we have 2nlk 2 d - l2 d. Since 4 cannot divide k2 d-i + l2 d-i for d ~ 2
(those two numbers are congruent to 1 modulo 4) we must have 2"l(k 2' - ' -
!2 '-')(k 2' - ' + ! 2'-') =? 2"- 1 1(k 2 ' - ' - ! 2'-'). By continuing this, we would ulti-
mately obtain 2n-d+2 Ik2 - !2 .
Thus, let a be the exponent of 2 in the prime decomposition of k 2 - ! 2
(since k fl, the exponent is finite). We will have n-d+2 Sa, but 2dln (since
s
2d is the order of (kl- 1 ) modulo 2") and this yields d log2 n. Of course this
would produce a <". n - log2 n + 2, which is impossible for sufficiently large n.
Contradiction.
92 Solutions
We have therefore proved the problem in the case when the "associated
set" of a and b has the gcd 1. Now, consider any two different numbers a and b
and consider the gcd of their infinite "associated set" d. Thus, their associated.
set would be {dn1 < dn2 < ... < dnk < ... }.Now take the numbers ad and bd
whose associated set includes {n1 < "2 < ... < nk < ... }.The associated set
of ad and bd is infinite and has the gcd 1. From what has been stated above,
this means that ad and bd a.re integers and from this it follows that a and b a.re
integers.
25. M We will prove the statement by induction; it is trivial to verify it for the first
11 terms of the sequence, so assume that all the terms up to an+ 5 ( n 2: 6) a.re
integers and we will show that an+6 will be.
First of all we shall prove that for any m $ n + 5, am is co-prime with all
of am-i,am-2,am-3,am-4· Indeed, suppose that there existed a primep such
that it divides both am and am-k for k $ 4, and take the smallest m for which
this happens. A look at the first 6 terms shows that we must have m 2: 6. Then,
amam-5 = ~-lam-4 + am-2am-3 and whatever the value of k, p will have to
divide the term in the RIIS which does not contaln am-k· One of the factors of
that term and am-k itself will be both divisible by p, and this contradicts the
minimality of m. So any am and am-k for 1 $ k $ 4 must be co-prime.
What we need to do to prove the induction is to show that
(1)
where from now on all equivalences a.re taken modulo an+l· Taking the
recurrence relations for n - 1, n - 2, n - 3, n - 4, n - 5 modulo an+2 yield to us
an+5an =an+2an+3, an+4an-1 = anan+3, an+3an-2 =. an-1an+2, an+2an-3 =
anan-1 and anan-3 = -an-lan-2· .
By multiplying the first and the fourth of the above relations, and by divid-
ing from them the product of the second, the third and the fifth (where division
by a number means multiplying by its inverse modulo an+b as we have proved
that all those numbers a.re co-prime with an+i) we will obtaln precisely the
desired (1).
26. SOLUTION BY ANDREI STEFANESCU
MForanyk E Zwewillconsiderthenumbersxk = 1+6••+',yk = 1-63•+1
and Zk = -62•+1. Note that the triplet (xk, y., z.) is a solution of the equation,
as
= 2 + 66k+3 - 66k+3 = 2
Since we can define this triplet for all k's, there will be an infinity of solu-
tions.
27. SOLUTION BY ANDREI STEFANESCU
D First of all, we can exclude the cases n = 1, 2, as in these cases we will
have the solutions a = b = c = d = 4 and a = b = c = d = 2. For each n 2 3,
we will write the solutions of our equation as (a, b, c, d) with a 2 b 2 c 2 d.
Assume that we have such a solution for a certain n, and pick that solution
with the minimal a (meaning that any other solution would have a' 2 a). Let
s = b + c + d, and therefore s S 3b S 3a. The hypothesis thus becomes
2 2 s 2 s 2 (n2 - 3)
s =a(n bed-2s-a) 2 3(n s-2s-s)= - - 3- -
by using a S s. This implies that n 2 S 6, but we assumed that n 2 3.
If bed < s, then becanse we have s S 3b, it follows that ed < 3 and therefore
d = 1 and c E {1,2}. If c = 2, then bed< s becomes 2b < 1+2 +band this
implies b < 3. Since b 2 c = 2, we obtain that b = 2 and then going back to
the hypothesis, it would yield n = ~~. Thns, a is a perfect square, a 2 2
and y'af5, leading to a = 25 and n = 3. If c = 1, it follows that Jab E l\l and
b S a S b + 2 (since a S s = b + 1+1). But Ja(a - 1) and Ja(a - 2) can
never be integers, hence a= band n = 2(a + l). Thns a E {1, 2}; for a= 1 we
a
obtain n = 4, and for a = 2 we get n = 3.
Therefore, all the n's for which we have solutions are n E {1, 2, 3, 4}.
Chapter 3
COMBINATORICS
SOLUTIONS
k
1. M Let B = LJ A; and m = #B. Denote B = {bi, ... , bm} and let Xi be the
i=l
number of subsets A; in which b; appears. We obviously have that Xi ::". 1,
because B is the union of all those subsets.
Consider an m x k table, in which the intersection of row i with column
j contai~ 1 if b; E A; and 0 otherwise. The sum of all the numbers in our
m k k
8= Lx' = L#A; ::". ;- (1)
1 I
In our table, let us calculate the number of pairs of l's situated on the same
row. Summing by columns, this is L #A; n
A3, but summing by rows we get
i<j
~(X•)
'7 Xf-x·
~· Therefore,
2' =
2
(
Lx ) s2
But l:xf ;::: ---'- = -;;;;; (where S = Lxi) and thus the above
m
inequality becomes
95
96 Solutions
the ant will step on the n + 1-th diagonal, there will be at most n + 1 - kn
poisoned points on that diagonal (because there are at least kn among the first
n diagonals). There are at least n + 2 - kn points on the n + 1-th diagonal
accessible to the ant from safe points on then-th diagonal (it's easy to see that
p points on a diagonal have at least p + 1 neighbors altogether on the next
diagonal). But from the definition of kn, there are at most kn+l - kn poisoned
points on then+ 1-th diagonal after the ant has reached then+ 1-th diagonal.
Therefore, since we can access n+2-kn points on then+ 1-th diagonal, and at
most kn+l - kn are poisoned, then at least n + 2 - kn+l points on the n + 1-th
diagonal can be reached in safety. This proves our induction.
So we just showed that the ant can reach any diagonal (since n+ 1-kn > 0
always), but we didn't yet prove the existence of an infinite path. Let An be
the set of all the points on the n-th diagonal on which the ant can get to, and
connect a point X E An to a point Y E An-1 by a line if the ant can safely get
from Y to X. We'll construct an infinite path (meaning a sequence k; of points
such that k; E A; and k; is connected to k,_ 1 for all i) which will be the correct
path for the ant, as follows:
For any m, we will try to find a km E Am from which we can construct
arbitrarily long paths. We'll assume that ko, ... , kn-1 have been built with this
property and we will show that there exists such a kn exists (ko = (0, 0) has
this property, as as proved above). By assumption, we can build arbitrarily long
paths from kn-1, and these paths pass through a finite set (An)· Therefore, a
sequence of arbitrarily long paths from kn-1 will pass through some element of
An, which we will pick as our kn. By following the path ki, ... , kn, . .. the ant
will always survive.
4. D Assume that there are no such three fields. First of all, note that no color
can appear in more than 16 fields (otherwise, by the pigeonhole principle, one
of the other colors would appear in less than 6 fields). Now, let us prove the
following fact:
LEMMA No row (or column) can contain more than 4 colors. Moreover, if
it contains 4 colors it must be of the form ABBCBBCBBD, where A,B,C,D
are 4 different colors.
PROOF Consider a row, and take its smallest segment which contains at
least 4 colors. Take its endpoints to be of colors A and D, and because that is
the smallest segment to contain at least 4 colors, then A cf D. Moreover, the
colors A and D do not appear again in the interior of the segment, which must
be paved by at least two other colors. The segment must have an even length (or
otherwise the endpoints A and D would have a different color midway between
them).
98 Solutions
Now, if the interior of the segment has exactly 2 other colors B and C, it is
easy to see (through a brief analysis of the cases) that the only possibility which
doesn't permit three equidistant different fields is when the segment is the whole
row, and of the form AB BCBBCBB D. If the interior of the segment contained
r > 2 colors, then let one of them be X and color the other r - 1 2'. 2 colors in
another color (say Y). The only possibility for A, B, X, Y to coexist peacefully is
the above-mentioned pattern (AXXYXXYXXB or AYYXYYXYYB), but
when we split Y back into its original colors, we will always obtain a triplet of
equidistant different fields.
Back to the problem. Since the number of times each color appears is at
least 6, we notice that the number of rows which contain a color A + the number
of columns which contain it can be no less than 5. Thus, if we sum this after all
15 colors, we get that total number of times when a color appears on a row +
the total number of times when a color appears on a column is at least 75. On
the other hand, this number is the sum after all the rows of how many colors
are on that row + the sum after all the columns of how many colors are on
that column. Because we have in total 20 rows and columns and because none
of them can contain more than 4 colors, by the pigeonhole principle we have 15
rows or columns which contain 4 colors each. Let's assume we have at least 8
such rows (it's exactly the same if we have at least 8 such columns).
Call the color that appears 6 times on a certain 4-color row (or column) the
"big color" of that row (or column). Call the other 3 the "little colors" of that
row. No color can be a "big color'' for more than two rows (otherwise we would
have 18 instances of that color, and we said that each color appears at most 16
times). So among the "big colors" of the eight 4-color rows there are at least 4
colors. So the columns 2,3,5,6,8,9 each contain all these four "big colors", and
thus these columns will be 4-color columns. But one of the eight 4-color rows
will be one of the rows 2,3,5,6,8,9 and therefore its "big color'' will be the "big
color" of the 4-color columns 2,3,5,6,8,9. Therefore, that color will appear at
least 6 x 6 = 36 times! Contradiction.
5. E The following configuration of 16 knights works, and we will show that we
have no such configuration with more than 16. Now, we will split the squares of
the 5 x 5 board into a number of subsets, depending on how many jump options
a knight on that field has. So, we have squares from which we can make 2,3,4,6
or 8 possible jumps, and note that our configuration is that of the 3-squares
and 4-squares (an x-square is a square of x possible jumps). So assume that
there exists a configuration of at least 17 knights.
Combinatorics 99
THE CONFIGURATION OF 16
KNIGHTS THAT WORKS IS THAT
WITH KNIGHTS ON THE SHADED
SQUARES
We will call two squares neighbors if a knight can jump between them .. First
of all, if we had a knight in one of the corners (the corners are 2-squares), then
both the corner's neighbors have to contain knights. But those neighbors have
in total 10 neighbors of their own, and at most 3 of these can be occupied by
knights (the corner field being one of them). So we have at least 7 unoccupied
squares. Now, we can have at most 8 squares blank (as we have 25 - 8 = 17
knights), so at least 2 of the other corners must have knights. But then the
neighbors of the 3 corners would just have too many neighbors which would
have to be left blank, as it is easy to see.
So none of the corners are taken. Then, suppose the center has a knight,
and because it has 8 neighbors, 6 of them would have to be left blank. 6 + 4 >
25 - 17 = 8, which is the maximum number of blank squares that we can have.
So the center is empty too, and therefore at most 3 of the 3,4 and 6-squares can
still be blank. But any 6-square has four 4-square neighbors, so if a 6-square were
in the configuration, at least two 4-squares would be missing. But any 4-square
is missing, then its at least two 3-square neighbors must also be blank (because
a 3-square only has three neighbors: the blank center and two 4-squares, and if
one of the 4-squares does not have a knight, our 3-square can't have either). So
in this case, we would also have under 17 total occupied squares.
Therefore, all the 6-squares must be blank and we are left with the 16-
configuration we showed in the beginning of the solution. This is therefore the
maximum number.
6. D Let a be the total number of the 4-pieces, aoo the number of tromiones with
their vertex (the field which is adjacent to the two other fields in the tromino)
at the intersection of an even column with an even line, a 10 the number of
tromnioes with the vertex on an odd line and an even column, ao1 those on an
even line and an odd column and a 11 the number of those on an odd line and
an odd column.
First of all, color all the odd lines white, and all the even lines black, such
that there will be 2n+ 1 more white fields than black ones. A 4-piece will always
have as many whites as blacks, and the only pieces which have more white fields
than black fields are the trominoes with their vertex on an odd line (and in the
100 Solutions
case of these tromnioes, the difference between whites and blacks is 1). So the
number of tromnios with their vertex on an odd line is at least 2n + 1 (because
the 2n + 1 difference between whites and blacks must be accounted for), and
thus au + a10 ::". 2n + 1.
Now consider the following coloring: color black all the fields with both
coordinates even and white all the rest. We will have n 2 black fields and 3n2 +
4n + 1 whites. Any 4-piece will cover three whites and 1 black, and any tromino
with one of the coordinates of the vertex even will cover 2 whites and 1 black.
All other tromnios will cover 3 whites. Now, the number of whites is
n2 = a + an + a10 + a01
If we eliminate a from these we get
3(aoo +an+ a10 + ao1) ::". 12n + 5 =? aoo +an+ a10 + a01 ::". 4n + 2
So the number of trominoes is at least 4n + 2.
7. M We shall prove a lemma related to the inclusion-exclusion principle.
LEMMA For any sets A 1 ,A2 , ... ,An we have
· PROOF For any element x E #{A1 LJA2 LJ ... LJAn}, suppose it is part of
exactly k of the sets A;. Then x's contribution to the LHS of (1) is 1, while its
contribution to its RHS is k - k(k; l). But 1 ::". k - k(k; l) (with equality
iff k E {1, 2}), and by summing up all these inequalities for all x we get exactly
(1). Moreover, we only have equality if k E {1, 2} for all x's, so each x must be
part of 1 or 2 of the sets in order to have equality.
Back to the problem. There can be no more than n pairs of P;'s that
intersect, as each such intersection between two Pi's uniquely determines a set
Pk. And since we have only n Pk's, we have at most n pairs of P;.'s that intersect.
Obviously, since all sets are distinct, two such sets intersect in only one element.
So we will have
Combinatorics 101
L #{A;nAj}sn
1:5i<j:5n
But then
our vertices (we are working with a connected component, so ever,y vertex will
appear eventually).
Now consider all the odd levels as a set, and all the even levels as the other.
The graph would be bipartite if two vertices in any one of the sets wouldn't be
connected. And if a point A in level x is connected to a point B in level x + 2k,
then they determine a cycle of odd length (because there would be a point Y
in a level above A and B such that from Y you can go down to both A and B,
and going from A up to Y, down to B and from there to A via the edge (A, B)
determines a cycle of odd length). Our lemma is thus proven.
Back to the problem. Take an odd cycle C of minimum length (suppose
that length is c), and note that it must be of length at most 5 (no triangles). If
any two non-adjacent vertices of C would be connected, that that line breaks C
up into an odd cycle and an even cycle, thus contradicting C's minimality. Now 1
any point x of C is connected to at most 2 points in C. If it were connected to
three of them (any two adjacent points of C cannot be both connected to x,
as we have no triangles), then at least two of these (say A and B) would be a
distance k of 2 apart from each other. So these A and B would be linked by a
path of length 2 (through x), which contradicts the minimality of C.
But that means that the number of edges between C and the points outside
C is at most 2(n- c) (each vertex not in C is connected to at most two of those
in C). So there is a vertex in C which has at most 2 (n - c) connections to
c
the points outside c. But as it was said before, it is connected to only 2 of the
.
pomts . C ("its imme
m . diate ne1"ghbors ) 1 and t hus its
. degree lS
. at most ~ 2n <
_ 2n
5
(because c 2: 5).
10. M Let us assume that there is no such infinite monochrome sequence 1 and
take a maximal white sequence 2k1 < kl + k2 < ... < 2kn (by maximal I
mean which cannot be extended any further) and a maximal black sequence
2!1 < !1 + l2 < ... < 2lm.
Assume that kn < lm.
Let us now consider all the white even natural numbers between 2kn + 1
and an arbitrary 2x, and suppose their number is W. If one of these even white
numbers (say 2k) had k + kn white, then we could extend the maximal white
sequence, which we assumed impossible. Thus, if 2kn < 2k S 2x is white, then
2k < k + kn S x + k is black. And therefore, the number b of black numbers
between 2kn + 1 and x + kn is at least W.
In the same way, if B is the number "of black even numbers between 21,,, + 1
and 2x, the number w of whites between 2lm + 1 and x + lm is at least B. But
B + W 2: x - lm (the total number of even numbers between 2lm + 1 and 2x)
and b+w S x-kn (the total number of numbers between 2kn + 1 and x+ kn)·
So 0 S (b - W) + (w - B) S lm - kn for all x. This means that b - W (for
Combinatorics 103
instance) is bounded for all x, and therefore there can be only a finite number
of black squarE<! which can't be written as kn + k for some white 2k. So from
a point onward, all black squares are of the form k + kn where 2k is white and
similarly from a point onward all white squarE<! will be of the form Im + I for
some black 21.
So, we can say that for some N, then k 2'. N is black iff2k-2kn is white and
I 2'. N is white iff 21 - 21m 2'. N is black. But for any k, 2k - 2kn and 2k - 21m
must be of the same color, or otherwise k wonld have to be both white and
black. So if Im - kn = a > 0, and that means that the numbers 2x and 2x + 2a
have the same color from a certain point onward. So the infinite arithmetic
sequence 2x, 2x + 2a, ... , 2x + 2na + ... has the same color throughout. But it is
easy to see that this progression also verifies the conditions in the hypothesis,
so we proved our initial assumption to be wrong.
11. E Let the sides of the scale be A and B and the weights be in the order
c1 ;::: C2 ;::: ... ;::: Cn+i, and we have c1 + ... + Cn+l = 2n. We will prove the
following two lemmas, which will be sufficient for our proof.
LEMMA 1 No weight is greater than the sum of all the weights that come
after it. Indeed, if this were the case, we would have Ck > ck+l + ... + Cn+l =>
Ck 2'. n - k + 2. But c1 2'. eo 2'. ... 2'. Ck 2'. n - k + 2, so we would have
says that the area of such a polygon (not necessarily convex) will be q + ~ - 1,
where q is the number of lattice points in the polygon's interior, and p is the
number of lattice points on its sides.
Thus, let our polygon be the broken line from A to C, the segment from
C to (n,n+l), the segment from (n,n+l) to (-1,n+l), the segment from
(-1,n + 1) to (-1,0) and the segment from (-1,0) to (0,0) (we a.re choosing
the polygon like this to have no self-intersections, even though we can apply
Pick even with self-intersections). The area of this polygon is the area we a.re
looking for plus 2n + 1.
But the number of interior points is 0 (since the broken line passes through
every point in the square) and the number of points on the boundary is (n + 1) 2
(the number of points of the square and its interior) plus 2n + 3 (the number
of the points on the segments we a.re adding). Therefore,
S+2n+l= (n+1)2+2n+3
2
13. E We will prove the problem by induction over n. For n = 1, 2 the problem is
trivial. Now, assume it true for n - 1 and let us prove it for n. If n will ever be
first in the sequence, it will be moved to the last place by the next move, and
nothing will ever remove it from there (the only move that could invert the last
number would be the inversion of all the numbers, but this can only happen
when n is first in line). In this case, the other n - 1 numbers would behave
independently of n and according to the induction hypothesis, 1 will reach the
top of the sequence sometime.
If n never reaches the top, then the number k which is initially last will
never be moved away from there (as we explained in the previous paragraph).
Thus, we will give the number n the "role" of k, and note that the first n - 1
numbers (meaning except for the number k) behave exactly like a sequence of
n - 1 numbers. Indeed, it does not matter that we have n instead of k, because
we assumed anyway that n never reaches the top, so its exact value does not
matter. So if k f 1, because of the induction hypothesis, 1 will eventually reach
the top.
But if we had k = 1 then n has the role of 1, and according to the induction
hypothesis, n will initially reach the top, which we assumed not to be true.
14. D Draw an arrow on every edge from the smallest to the largest number, and
note that every triangle XYZ will have only one vertex (call it the 'apex' of
that triangle, and say it is X) for which one of XY and XZ points toward X
and the other points away from it. It is easy to check that a triangle is positive
if the arrow that points toward X is before the arrow that points away from
Combinatorics 105
v
\7~---- 'apex
But it is easy to check that this happens for all boundary points iff the edge
between each of them and the next boundary point in clockwise order points
away from it. This means that the boundary points and the edges between them
form a cycle, which is obviously impossible, as we cannot have each number
greater than the next. We must therefore have at least 7 positive triangles.
15. E Let us consider, from all the ways to split the set of our vertioes into k - 1
disjoint sets A,, .. ., Ak-1' that way which yields the smallest number of edges
inside one of the sets. Now take a vertex from A;; if it would have less neighbors
in some A1 that it has in A;, then we could move it to A1 !llld contradict the
minimality of our arrangement. So the number of neighbors it has in A1 is 2':
the number of neighbors it has in A; for all j. Summing this up for all j, we
have that the degree of that vertex is 2': (k - 1) times the number of neighbors
that vertex has in its own set. Summing this up for all vertices and dividing by
2 yields that the total number of edges is 2': (k - 1) times the number of edges
106 Solutions
inside the same set. So we can erase those in-set edges, and the remaining graph
will be multipartite, and thus it will have no k-clique.
16. M In a complete oriented graph, call well-ordered triangles those whose edges
form a cycle, and call the other triangles badly-ordered. The minimum number
of well-ordered triangles is 0, as we can construct such a graph as follows: place
the numbers 1, 2, ... , n in the vertices of our n-graph and draw the edge between
i and j from vertex: i to vertex j. It's easy to see that in this case we have no
well-ordered triangles.
Since the total number of triangles is always (;) , instead of finding the
maximum number of well-ordered triangles, we will find the minimum number
of badly-ordered triangles. Note that every badly-ordered triangle contains one
and only one vertex from which two of its edges leave. Therefore, the number
of badly-ordered vertices is equal to the number of pairs of edges that leave
from the same vertex. If k; is the out-degree of vertex i then the number we are
looking for is t, (i). But keeping in mind that ki + ... +kn equals the total
L(k;-k3)2
(k1 + ··· + kn) 2 + _i<~;_ _ __
n n n0-D
T= 2 ---4-
L(k;-k;)2
n(n-1) 2 _ n(n-1) + _i<J~·~--
=
8 4 2n
So to find the minimum value of T is equivalent to determining the mini-
mum of the non-negative sum S = L(k; - k;) 2 • For odd n, we can make it 0
i<j
by taking k; = n;
1 for all i (this can be accomplished by drawing an arrow
n-1
out ofi into i + 1, i + 2, ... , i + - 2 - for all i).
For even n = 2a it is more complicated, but we can show that the sum is
always greater than or equal to a 2 . Indeed, consider the numbers l; = 2k;-2a+ 1
Combinatorics 107
and l; are odd integers that sum up to O; then Swill equal~ L(l;-l;) 2. Assume
i<j
that li, ... ,lk < 0 < lk+i,···,l2a and that k ~a (k <a is perfectly similar); we
will have lk+l + ... + l2a = -li - ... - lk = M and obviously M ~ a. Therefore,
a2M2
~~=M2~a2
Now, the last thing w~ need to prove is that there actually exists such a
graph for which S = a 2 • For any 1 :s; i < j :s; a draw an arrow from i to j, from
j to 2a + 1 - i, from 2a + 1 - i to 2a + 1 - j and from 2a + 1 - j to i, and
then draw arrows from i to 2a + 1 - i for all 1 :s; i :s; a. It's easy to check that
we have S = a 2 • So the number of well-ordered triangles will be, for odd n, at
most
A and B iff they are friends with X. Finally, give A and B and extra color
only for themselves, and we have added in total only 2m + 1 beads. So we
have accomplished our coloring with just (m + 1) 2 beads, which proves our
hypothesis.
The case n = 2m+ 1 is proved in the exact same way, just that the equality
case is when we split up the natives into a group of m and a group of m + 1
and make everyone be friends only with those of the other group.
18. M Let us assume that n is not a power of 2, and consider the polynomials
f(X) = X"' + X"• + ... + X"• and g(X) = x» + X" + ... +x••. We have, by
assumption, that h = f - g of 0 and that
i<J i<j
h(•l(l)p(l) = 2•h(k)(l)
Since p(l) = 2n and we assumed n a non-power of 2, then this can only
happen if h(•l(l) = 0. Thus, the induction is complete, and we will have all the
derivatives of h at the point 1 equal to 0. Of course, for polynomials this can
only happen if h is zero throughout, which is impossible because ff g.
19. D Initially, on the i-th row and j-th column, we will find the number lO(i-l)+j.
After k transformations, suppose that a•(i,j) is the number situated at the
intersection of the i-th row and j-th column. It is easy to see that the quantity
L
s. = a•(i,j)(lO(i - 1) + j) is left invariant by all transformations. But the
i,j
problem tells us that at some point, the numbers a•(i,j) are just a permutation
of the numbers lO(i - 1) + j, and so by the rearrangement inequality we have
s. ~ L(lO(i - 1) + j) 2 for all k. But s. is conserved, and its initial value is
iJ
L(lO(i-1) + j) 2 , so the previous inequality is an equality, and therefore the
i,j
numbers are in the initial order at that point.
20. M We shall exclude the cases when all the a;'s are greater than all the b;'s or
that when all the b;'s are greater than all the a;'s, since they are trivial. That
means that ai < bi and an > bn 1 so by continuity there must be a certain k
Combinatorics 109
such that ak < bk and ak+l > bk+l · Therefore, a; < b, for all i from 1 to k, and
ai > bi for all i from k to n. Therefore, our sum is
00
i +1
4-8<2=-· -4
2' i=O
Contradiction, and thus we will always have two congruent triangles.
22. SOLUTION BY ADRlAN ZABARlUC
D Let us first inspect the case IA,I - m for all i E {1, 2,. .. , n }.
LEMMA Let m,n,k E N be such that n > 2m and Ai,A2,.-.,Ak C
{1,2,. ..,n} so that IA.I - m and A,nA; of 0, \11 < i :5 j < n. Then we
have
k :5 (~~ ~)
Consider any permutation" E Sn of the elements of the set {1, 2, .. .,n}.
Let us write the numbers of this permutation on a circle which has a red dot,
starting the writing of the numbers from the red dot, counterclockwise. Call an
110 Solutions
Our problem is now proven, as the lemma tells us that there are at most
( :-::. ~) sets with m elements and at most ( :-::. ~) with m - 1 elements, so
we have at most ( :-::. ~) + ( :-::. ~) = (:) sets in total. This upper bound
can be reached by considering all the sets of m and m-1 elements which contain
the element 1.
23. M The total number of colorings of our 2005 points with 2 colors is 22005 .
Similarly, the number of colorings which leave a certain 18-number arithmetic
progression of the same color is 22005 - 17 (all the points outside the progression
can be colored in any way, and the progression can be colored either all white
or all black). If m is the total number of 18-progressions, let A, be the set of
those colorings which preserve progression i monochrome. But we just showed
before that #A;. = 22005 - 17 and we will have that
But what ism? A progression a,a+r, ... ,a+ 17r is completely defined by a
l
2005 - aj
and r such that a+ 17r ~ 2005. For any a we have thus - -1-7-
2005- a
~ -- 1-7-
progressions which start with a. So the total number of progressions will be
# Al uu...
A
,_::;
22005-17 . 2005 . 2004
34 < 22005
as an easy computation reveals. So not all colorings axe in Al LJ ... LJ Am
and thus at least one coloring will leave no 18-progression monochrome, as we
needed to prove.
24. D Let us assume that a checker will eventually have the y-coordinate 5, and
choose the origin such that the point where that checker reaches the 5-th level
is (0, 5). Let a =../52+ 1 , and a will satisfy the equation a 2 = a + 1. At
any moment, assign to our distribution of checkers the swn Laik-likl, where
k
(i.,j.) axe the coordinates of checker k. It is easy to see that any upward jump
in the lattice plane preserves this sum, whereas all the downward, leftward and
rightward jumps decrease it. Therefore, in the end our sum will be at least a 5
(because we have a checker on (0, 5)), but it will be at most equal to the initial
value of the sum (when the entire lower half-plane was covered by checkers).
Let us calculate the initial value of the sum, which we will call S. The sum
after all the checkers on the line y = j ::; 0 will be
f
i=-oo
ai-lil = -ai + 2fi=O
ai-i = -ai + 2aiE ~ =
=a; (~-1)
a-1
=a;~
a-1
The initial value of Swill be the sum after all non-positive j's of the above
expression, so because his sum is non-increasing we will have
get from vertex: ai:::a;;::;- to ~ by the path (a, ... an_i,a2 ... an-1b1),
( a2···an-1b1, ag ... an-1 bib2) 1 ••• 1 (an-l bi ... bn-2 1 b1 1 ••• , bn-1)
An arrow between a and b will therefore represent an n-digit number, ob-
tained by taking a's first digit, a and b's n-2 common digits, and b's last digit.
Therefore, our problem is equivalent with finding an Euler cycle in G. We will
therefore prove that every m-vertex connected graph with all out-degrees and
all in-degrees 2 has an Euler cycle.
We will first prove the following lemma.
LEMMA Any graph with out(v) = in(v) for all vertices v can be broken up
into disjoint cycles (disjoint as in having no common edge, but allowing common
vertices).
We will prove it by induction after the number of edges e. For e = 0 we have
nothing to prove; assume it true for all nwnbers smaller that e and let's prove
it for e. Take a vertex and create a path out of it. That path will intersect itself
at some point, and thus we have a cycle. If we remove that cycle, we are still
left with a graph that has out(v) = in(v) for all vertices v and the induction is
proven.
Now, consider our graph as broken up into k disjoint cycles, and we will
rearrange that cycles such that the graph is broken up into k - 1 disjoint cycles
(where we don't restrict ourselves just to simple cycles; a vertex may appear
may appear more than once in a cycle, but an edge may not). Take a cycle C. If
all the edges are in C, it is our Euler cycle; if not, there is some cycle D which
touches C (if not, then C would be isolated from the edges that are outside of
it, and connectedness is violated). We can therefore unite C and D into one big
graph in the obvious way: start from one of their common points, walk on all
of C and return to that point, then walk on all of D and return to that point.
What you have is just a bigger cycle.
So we will be able to put all our edges into one big cycle, which will be our
desired Euler cycle.
26. M Call a set even if the sum of its elements is even, and odd otherwise.
Let d(n,k) be the difference between the number of even and odd subsets of
{1,2, ... ,n} with k elements. We will compute a recurrence for d(n, k).
First of all, d(2a + 1, 2b + 1) = -d(2a, 2b), because we can split up the
number of 2b + 1-element sets of {1,2, ... , 2a + 1} into two groups: those which
contain 2a+ 1 and those which do not. In the group of those which do not contain
2a + 1, there is a bijection between the even subsets and the odd subsets (that
bijection is f(xi, ... , x2Hi) = (2a + 1 - xi, ... , 2a + 1 - x 2o+i)), and so in that
group the difference between the number of even subsets and odd subsets is
0. So the only difference between even sets and odd sets arises from the sets
Combinatorics 113
which contain 2a+ 1 and 2b other numbers from {1, 2, ... , 2a }. Hence the formula
d(2a+ 1, 2b+ 1) = -d(2a, 2b) (the minus sign arises because even-summed sets
in {1, 2, .. ., 2a + 1} become odd-summed sets in {1, 2, .. .,2a} when we remove
the element 2a + 1).
Similarly, using the same argument and bijection, we come up with the re-
currences d(2a+l, 2b) = d(2a, 2b) and d(2a, 2b+l) = 0. Computing a recurrence
for d(2a,2b) is straightforward, as we again split the difference in "the differ-
ence in number between even and odd sets that contaffi 2a" plus "the difference
between even and odd sets that do not contain 2a''. We thus get d(2a, 2b) =
d(2a-1,2b-1) +d(2a-1, 2b), which in light of our other recurrences becomes
d(2a, 2b) = -d(2a - 2,2b- 2) + d(2a - 2, 2b). If we let f(a, b) = (-l)•d(2a, 2b)
then the recurrence becomes f(a, b) = f(a-1,b) + f(a -1,b- 1). It is easy to
see that(;) verifies both this relation and the initial values f(k,1) (which are
easy to check). Therefore, we will have
smallest number that was not in Xn· It is thus obvious that Xo C X1 C X2 C ...
and Xn+1-Xn is an (a, b)-adapted set. Therefore, we will be able to write N as
a union of (a,b)-adapted sets Un;:,o(Xn+l -Xn), because the sets are disjoint
and every time we add a new set, we also add the smallest element that had
been left out.
But let us show how to pass from Xn to Xn+I· For every n, let kn be the
smallest number not in Xn. Our algorithm is to add to Xn the set A + km or
if that isn't possible, the set B +kn. Assume that We cannot perform this.
First of all, because Xm C Xn for m :'> n, we will also have km < kn
(strictly). But from the description of the algorithm, it follows that in Xn there
are no sets of the form A+ x or B + x, with x <". kn (if there were, they would
have to be added at a certain point, so x would be an km for m < n; but this
is impossible, as x <". kn)· Therefore, the number kn +a+ b is not in Xn and
the only reason why neither of the sets A+ kn and B + kn can be added is that
the numbers kn + a and kn + b are already added.
So the number kn + b has already been added, and note that it is only part
of the sets A+(kn+b), B+(kn +b), A+(kn+b-a), A+ (kn-a), B+(kn-a).
The first three sets could not have been added before (as it was said in the
previous paragraph), so one of the last two must have been added sometime. If
A+ (kn - a) was added before, it would mean that kn is already added, which
is not by its definition. So the one which was added before is B + (kn - a). But
why would we have ever chosen to add B +(kn - a) over A+ (kn - a) (which,
as was said, was never added) and from its description, our algorithm always
prefers A+ km over B + km? We thus obtain a contradiction, so the algorithm
must function indefinitely.
31. M Note that if the main diagonal is initially infected, the whole board will
become so in time. Therefore, n cells are enough, and we will show that they
are the minimum number. Indeed, suppose that we initially have k infected
cells, and we will call an edge in our board infected if it is the boundary of an
infected square (we consider as edges only segments of unit length)
Since we initially have k infected squares, we have at most 4k infected edges.
With every new infection, we get at most two new infected edges (as the square
that is to be infected has at least two already infected edges), and since we
will make n 2 - k new infections until we cover the board, the total number of
infected edges will be at most 4k+2(n2 -k). But in theend all theedges must be
infected, and there are 2n(n+l) of them, so 2n(n+l) :'> 4k+2(n2 -k) =? n :'> k,
which is exactly what we needed to prove.
116 Solutions
32. D Three triangles with a common vertex determine a 7-graph that works. We
will now show that 7 is the maximum number, and that there are no such 8-
vertex graphs. Suppose for the purpose of contradiction that there were one,
G.
LEMMA Any 6-graph has either a triangle or 3 independent vertices (3
vertices with no edges hetween them). This can be proven easily, by taking a
vertex and looking at how many neighhors it has. If it has at least 3 neighhors,
then if any of them are connected we get a triangle, otherwise the neighbors
are 3 independent vertices. If our vertex has at most 2 neighbors, then if any of
its 3 non-neighbors are not connected we get 3 independent vertices, otherwise
the 3 non-neighbors form a triangle.
The lemma is thus proved and we shall return to the problem. Every vertex
V E G has at least 2 neighhors, because otherwise we could apply the lemma
and get a triangle or 3 independent vertices among V's non-neighbors. It is
easy to see that those 3 points and V contradict the hypothesis. Moreover, a
vertex A cannot be connected to all other vertices; if it were and if any B were
connected to two other vertices besides A, these four vertices would fonn an
impossible 4-cycle. Otherwise, the degree of all vertices in the 7-vertex subgraph
obtained by removing A is at most 1, and we can easily see that we will have 4
independent vertices.
Now, it is easy to see that in every graph with the degree of every vertex
at least 2 we will have a cycle. Let us consider a minimal cycle with n vertices
and note that n f 4 (4-cycles contradict our hypothesis). If n = 3 then let
our triangle be ABC and note that any other vertex V must be connected
to exactly one of A, B and C (otherwise V ABC would he a contradictory 4-
suhgraph) Thus, the other 5 vertices except for those of our triangle will be split
into 3 groups: A's neighbors, B's neighbors and C's neighbors. Moreover, note
that two vertices from two of these groups cannot be connected, otherwise those
vertices and two of A, B, C would form an impossihle 4-cycle. If two vertices
in the same groups (say A's group) were connected, then these two vertices, A
and a non-neighhor of A (we said that no vertex can be connected to all other
vertices) would yield a contradiction. But then, all 5 vertices except for A, B,C
are independent, which is impossible.
So the cycle of minimum length is not a triangle, and then n ~ 5. Then
any two non-consecutive vertices in the cycle cannot be connected (otherwise,
we contradict the minimality of our cycle). Note that a vertex that is not in
our cycle cannot be connected to two vertices in our cycle; indeed, if that
were the case we would ohtain a triangle, a 4-cycle or a cycle smaller than the
minimal one. Therefore, if n = 5, there will be two of the vertices outside our
Combinatorics 117
cycle which are not connected (because they can't form a triangle). So those
two vertices (call them U and V) will be connected to at most two vertices
in our pentagon; thus take any A and B of the other three vertices of the
pentagon such that A and B a.re non-connected, and therefore A, B, U, V a.re
independent. If n E {6, 7}, take any vertex V outside our cycle and take the
at least 5 vertices of the cycle to which it is not connected. There will be 3
independent vertices among them, and these 3 together with V will form a
contradictory 4-subgraph. Finally, if n = 8 there will be 4 independent vertices
in this octogon. The problem is therefore proved.
33. D Suppose that the dimensions of the initial rectangle are both non-integers.
Then, call the direction of one of its sides the horizontal and the other direction
the vertical; let b be the length of the big rectangle's horizontal side. Now, take
all the rectangles of the subdivision and break their integral dimension into
several rectangles with that dimension 1. Thus, all our small rectangles will
have a dimension 1, and call a rectangle horizontal if the dimension of size 1 is
horizontal, or vertical otherwise.
Let all the vertical rectangles whose lowest point is at a distance x from
the bottom of the big rectangle form "level x". Obviously, there will be a finite
number of such levels xi, ... 1 Xk where we consider levels with integer x, even if
there are no rectangles in those levels. Let S(x•) be the sum of the horizontal
dimensions of the rectangle in level Xk.
We will prove by induction after i the following statement: we have { S( i)} =
{b} and for every one of finitely many levels i + h with 0 < h < 1 we will have
{ S(i + h)} = 0. ({y} means the fractional part of y). It is easy to note that the
base case i = 0 is proven in the same way as the inductive step, so we will just
show that if the theorem is true for i - 1 it is true for i.
Indeed, consider all the levels i - l+j• (0 < jk < 1) and all the levels i+h•
(0 < h• < 1), and we know that {S(i - 1)} = {b} and {S(i - 1 + jk)} = 0.
Take the horizontal line situated at height i and look at the rectangles which
are not below it and which it touches. That line will cut the entire rectangle
over a length b, but it will cut horizontal rectangles over integer lengths, the
levels i-1 + jk over integer lengths (by the induction hypothesis) and the level
i over the length S(i). So S(i) plus a number of integers yields b, and therefore
{S(i)} = {b}. A similar argument will prove that {S(i + hk)} = 0.
But this brings us to our final argument; if the vertical dimension of the
big rectangle were not an integer, then the only level of vertical rectangles that
can touch the top side will be a level of the form i + h, where 0 < h < 1.
Looking at the top part of the big rectangle, it cuts the whole thing over a
length b, it cuts the horizontal rectangles over an integer length and the level
118 Solutions
i +hover an integer length (since {S(i + h)} = 0). Therefore bis an integer,
which contradicts our initial assumption. We must have at least one dimension
of our big rectangle integral.
34. E Let us call the coordinates of a certain field of our matrix the pair (i, j), where
i is its row and j its column. An independent set of 1's will be a collection of
l's all situated on different rows and columns. Let k be the maximum size of
an independent set of l's, and aasume WLOG that we have an independent set
of 1's on the first k rows and columns, i.e. there is a 1 at the intersection of
i
row with column i (for 1 sis k). Let m be the smallest number of rows and
columns that cover our 1's.
Because that set is maximal, there are no l's with coordinates (i,j) with
i > k and j > k. Call one of the rows 1, 2, ... , k or one of the columns 1, 2, ... , k
necessary iff there is a 1 on it outside the square {l i s s k, 1 js s k}. We
can see that row i and colwnn i cannot both be necessary, because if they were
we could replace (i, i) with the two elements on column i and row i that are
outside the square {l s is s s
k, 1 j k} and get a greater independent set of
1's. So there are at most k necessary rows and columns, and its easy to see llow
that all the l's must be on them. Thus, we proved that ms k and showing that
m 2'. k is simple (we cannot even cover the independent set of size k with less
thank rows and columns, much less the whole table). So we must have k = m.
35. D We will prove the statement by induction over n. It is true for 1, as we have at
least ten 1-digit sequences (since each every digit must appear once). So there
are no such numbers for n = 1 1 but the problem remains logically true, because
any such number will be rational.
Now, asswne it true for n and let us prove it for n+ 11 where we have at most
n+9 sequences of n+ 1 successive digits. Ifwe had two n+l-sequences with the
same first n digits, then the nwnber of n-sequences of consecutive digits must
be at most n + 8 and we can apply induction to discover that x is rational. But
if the opposite holds, i.e. to every n-sequence Bk there corresponded a unique
digit right after it (call it a.) then we can easily prove that xis rational.
Indeed, as the decimal representation of our number is infinite, we will find
a sequence a, ... an which appears twice. But let b1 ••• bm be the digits between
the two apparitions of this sequence, and that means that the digit that comes
after ai ... an is bi; so bi must appear after the second apparition of ai ... an as
well. Then, the digit that comes after a2···anb1 is b., so it must come after the
second apparition of~ as well ... Thus, our number will be periodic, as
the sequence a, ... anb, ... bm repeats to infinity and x will therefore be rational.
36. SOLUTION BY ANDREI STEFANESCU
Combinatorics 119
M Since the game cannot end in a draw, one of the players must have a
winning strategy (because if one of them does not have a winning strategy,
it means that the other player can always stop him from winning by winning
himself). We will show that the first player always has a winning strategy, and
to do that assume for the purpose of contradiction that the second player has
a winning strategy.
Then, it means that whatever k the first player (call him A) initially writes
down, the second player (B) can win. But note that then A can just write down
1 and fro!Jl this point the game is just as if B had started (since even in the
case when B starts, the number 1 could never be written except on the first
tum). Therefore, A has a winning strategy, and this yields a contradiction to
our initial assumption. Thus, A always has a winning strategy.
37. SOLUTION BY ANDREI STEFANESCU
We will prove the statement by induction over n, where the case n = 2 is
trivial to check. Assume it true for all k < n and we will prove it for n; now,
Turan's theorem tells us that our graph must have a triangle, whose vertices
we will call A,B,C. Let kA,kB and kc be the number of neighbors each of
these points has outside the set {A,B,C}. If kA +kB+ kc:,,; 3n - 5 then
(kA + kB) + (kB + kc) + (kc + kA) :,,; 6n - 10; by the pigeonhole principle,
one of the brackets will be at most 2n - 4, and we will assume k A + kB to be
that bracket. Then, by removing A and B, the resulting graph will have 2n - 2
vertices and at least (n- 1) 2 + 1 edges, so by the induction hypothesis it will
have at least n- 1 triangles. Together with the triangle ABC, our initial graph
will have at least n triangles.
If kA +kB +kc ;::: 3n - 4, then let n; for 0 :,,; i :,,; 3 be the number of
vertices outside {A, B, C} which have i neighbors in {A, B, C}. We will have
n1+2n2 + 3n3 = kA +kB+ kc ;::: 3n-4, and the number of triangles formed by
vertices outside {A, B,C} with the vertices {A, B, C} will be at least n 2 +3n3.
This happens because any vertex X which is connected to two of A, B, C will
form a triangle with those two, while any X which is connected to all of A, B, C
will form three triangles together with pairs of them. But n 1 +n2 :,,; 2n- 3 (since
we have just 2n vertices in total), so n2 + 3ns ;::: 3n - 4 - (n1 + n2) ;::: n- 1.
Together with the triangle ABC, this yields in total at least n triangles.
Chapter 4
ALGEBRA
SOLUTIONS
1. E Let P(X) = xn+xn- 1an-l + ... +ao = (X -x1)(X -x2) ... (X -xn), where
xi, ... , Xn are the roots of the polynomial, all in the interval (0, 1). The formal
derivative of P will be
n-2
P'(X) = nxn-l + (n- l)an-lxn- 2 + L ia,x•- 1 (1)
i=O
But the derivative is also equal to
(X - x1) ... (X - Xn-1)+ (X - x1) ... (X - Xn-2)(X - Xn)+
+... + (X - x2) ... (X - Xn) (2)
Therefore, evaluating (1) and (2) at X = 1 and equating them will yield
n-(n- l)(x1 + ... +xn) +8 > n- (n- l)(x1 + ... +xn) =? S?. 0
exactly what we needed to prove.
121
122 Solutions
10a2 ::; a 2 (2002b2 - a 2) < 2a 2A+ >-2 =? 2a 2(5 - -") ::; >-2 (2)
But this takes place for an infinity of fractions ~, and since ~ is very close
to v'2002 an infinity of values of a must be numerators for this infinite number
of fractions. So if-"< 5, then the LHS of (2) can be arbitrarily large, while the
RHS is finite. So -" 2': 5.
3. M Let us take two cases. First of all, suppose that there is a number i such
that a,+ a;+ 2 > ~· Then our sum will be
1
ai+3ai+4 + ... + ai-2ai-l ~ 2(ai+4 + ... + ai-1) ~
::; (a;+ a;+2)(a;+4 + ... +a,_,) ::; a;(ai+3 + ... + a,_2) + ai+2(ai+4 + ... + a1-1)
By applying this into (1), we get
S::; (a;+ ai+2)(a;+l + a,+3 + ... + a1-1) = (a;+ a,+2)(1- a, - ai+2) ::; 41
1 .
because x(l - x) ::; 4 for all x between 0 and 1. Now, let us examme the
second case, when for all i we have a, + a,+ 2 ::; ~· Multiplying this by a;+1
gives us
ai+l
aiai+l + ai+lai+2.::::; - 2-
=L (a 3 + a 2 c- a 2 b- 2a 2 b- 2abc + 2ab2 + ab 2 + b2 c- b3 ) =
= L 4a2 c- L2a2 b-6abc 2: 0
By dividing this by 2abc we get
2 (~ + ~ + > ~ +~ + +3 -=) =
b c a -c a b
exactly what we had to prove.
5. D Since the sequences ai and bi are ordered in the same way, Chebyshev's
inequality tells us that
(a1 +an) 2 •
If (k +l+ 1) - 4aian ::; 0, then (2) holds because of the obvious
b
2a 2 + "2 2'. 2av'b
2b 2 + ~ 2'. 2bvfa
b
2ab+ "2 2'. 2bv/a
2ab + ~ 2'. 2av'b
By summing all these up, we get
1
J(f(x) - f(y)) = f(f(x)) + f(x) · f(y) + f(f(y)) - 1 =a - 2(f(x) - f(y)) 2 (1)
for all real x,y. Now we will try to prove that f(x)- f(y) can take on any
real value. Consider in the initial relation y = y0 with f(y 0 ) = bf 0 (if such a
number y 0 does not exist then f is identical zero, which is a contradiction) and
we get
a-1-b2
f(xo - b) - f(xo) = bxo + - -2- -
a - 1- b2
But by varying x 0 , the number m = bx0 + - -2- - will take on any real
m2
value, so letting x = xo - bandy= xo in (1) we get f(m) =a - 2 , for any
realm.
Algebra 125
a1 1 1
(af+x) 2 ::; 2a,(a,-l)+x -2a,+1(a;+1-l)+x
for all i's. These inequalities are true, because
(af-a;+x)(af+a,+x)::; (at+x) 2 *
=>~< 2ai
(at+x) 2 - (af-a;+x)(af+a,+x)
1 1 1
= af - ai +x af + ai + x ::; at - ai +x al+i - ai+l +x
since ai(ai+ 1) ::; ai+l (ai+l -1) (because the ai's are an increasing sequence
of natural numbers).
126 Solutions
10. D Let Y• = - -1- - and let us assume for the purpose of contradiction that
n-1 +xk
LY• > 1. But then
>La• (n-1)2
- ai + ... +ak-1 +ak+l +···+an
= (n- 1) 2 _!!1:__ > (n- 1) 2 _!!1:__
S-a• 1-a•
This contradicts (2), so we cannot have 2::---1- - > 1. Our problem
n- 1 +xk
is thus proven.
11. SOLUTION BY GABRIEL DosPINESCU
D The Cauchy-Schwartz inequality gives us
Lb':c~)Lx2 L(i:J
By adding 3 to both sides of this we get
=(x+y+z)JL(b:S +~ (2)
3 " (a+c)(b+c)
ab '""'
# 4 ~L., #3(a+b)(b+c)(c+a) ~4L.,ab(a+b) #
But let Q = anXn+ ... +ao and asswne that it has other non-zero coefficients
than the leading one. Let x• be the greatest power k < n and with a non-zero
coefficient and the coefficients of n+k on both sides of (3) will be equal. But they
will be 2nkanak = 2n2anak =? n = k (since am ak # 0) which is impossible.
Therefore, Q = anXn, and in this case (3) immediately yields n = 1 and a 1 = 1.
So the only solution for Q is Q(X) = X. But then R = ±1 and so the only
solutions for P are P(X) = ±1.
13. M Letting x = 0 yields J(f(y)) = y + j2(0), and therefore we will have
Since this holds for all x's, we must necesaarily have f(O) = 0 and therefore
f(f(y)) = y and thus f is injective. But by making y = 0 in the hypothesis,
we will have f(x 2 ) = J2(x). First of all, this. tells us that f is positive for all
positive values of x. Second of all, it tells us that f(-x) = ±f(x), and since f is
injective we must have f(-x) = -f(x). So by making z = y'X in the hypothesis
and taking f (y) instead of y, we will have
f(p) = f ( ~ + ··· + ~) = qf ( ~) =? f ( ~) = a~
So we proved that f(x) =ax holds for any rational x. But we also know
that the function is increasing, and if we take two sequences of rational numbers
kn /' x (the rationals kn are smaller than x and approach it) and ln ',,. x (the
rationals ln are greater than x and approach it) we will have
Algebra 129
and since kn and ln tend to x we will have, for all x, ax :'> f (x) :'> ax =?
f(x) =ax. But we also have f(f(x)) = x, so a2 = 1, and therefore the only
two functions that can work are f(x) = x and f(x) = -x. We can easily check
that only f(x) = x satisfies.
14. M We will show that the largest such k is 5. Indeed, any k > 5 is not good,
because we can choose a= b = 1 and c = 2 (which are not sides of a triangle)
and we will have k · 2 · 1·1>23 +1 3 + 13 ¢} 2k > 10.
Now, let us prove that any k :,,; 5 works for us. Take 0 < a :,,; b :'> c which
satisfy kabc > a3 + b3 + c3 and assume that they are not side-lengths of a
triangle. This could only happen if c > a + b and then we would have
a 3 +b 3 (a+b) 3
5ab2: kab> c2 +--c- 2: c2+ ~ (1)
We used that
5
5ab > ;;;(a+ b) 2 =? 4ab > (a+ b)2
which is obviously impossible. We have thus proved that the greatest such
k is indeed 5.
15. D LEMMA There exists a sequence on positive integers 1 = k 1 < k2 < ... <
kn < ... such that
(1)
is unbounded.
We can construct this sequence inductively such that for all i we have
ak - ak 2
~ 2: 3. Indeed, suppose that we have constructed k 1 < ... < kn- 1
and since the sequence a• is unbounded, we will always find a kn such that
ak. 2: 3ak._, =? ak. - ak._,
a•. -> ~3
This sequence will yield all the terms in the series (1) greater than ~, and
therefore the sum will be unbounded.
But, back to the problem, for all i we will alwa;ys have
130 Solutions
ak,+1 - ak~ + ak~+2 - ak,+1 + ... + ak,+i - akt+i-1 ;:::: ak,+i - ak,
akt+i ak,+i ako+i ako+i
But by summing this up or all i, our series will be greater than series (1),
which is unbounded. Our series is therefore unbounded.
16. M Note that any function of the form f(x) = x + ': + b satisfies, for all real a
x
and b. We will prove that these are the only solutions, and to do that we will
consider the function g( x) = f (x) - x. This function satisfies
xg (x + ~) + yg(y) = yg (v + ~) +xg(x)
xg(x + y) + !g
y
(!)
y
= !g (! + !) + xg(x)
y x y
=?
Ifwe now let g(l) = a+b and g(-1) = -a+b, writing (2) for y = -1 gives
us
=? g(x) = ': + b
x
Therefore, we will have f(x) = x + ': + b, as desired.
x
Algebra. 131
(1)
But
2=~+vibc~I>=1
and therefore ( 1) is true; because we came to this point by equivalences,
the problem is proved.
19. D No, there doesn't exist such a set. Indeed, suppose there was an M =
{a1, ... ,a•} with this property and let F = {f E M[X]}, the set of polyno-
mials with coefficients in M and with all the roots in M. According to our
assumption, we have such polynomials of arbitrarily large degree.
Consider an arbitrary f in F, of the form
132 Solutions
= (X" - t,x"-'a, + C~)x"- 2 a~ + ... )... (X'< - t.x«-'a• + C; )x«- 2a~ + ...)
Let n = t 1 + ... + tk be the degree of the polynomial, and it is easy to
see that the coefficient or xn- 1 is = - s L
tiai, and the coefficient xn- 2 is
i
n
Therefore, we have M• 2'. IT(l - lz,I•), but•'~ Mk= 0 and
i=l
Jim
k-'>oo
IT (1- lz,I•) = IT (1- Jim lz,1•) = 1
i=l i=l
k-oo
4-S 4S-S2
8+ l+ p 2'. P+ 1+"P' <*4+ PS 2: P+ P 2 +4S-S2 '*
# (S-2) 2 2'.P(P+1-S) (1)
s
If P(P + 1- S) 0 the above inequality holds and we are done. So let us
. pos1t1ve;
asswne it . . b ut p ~ S2 so
4
and let these numbers be x., xi, ... (where our xis among these) and let all the
other numbers in the sequence be xe, Xfi ....
It is ea.sy to see that if we eliminate these numbers and replace them all
with a, the property is preserved (no number is the beginning sequence of any
other). According to the induction hypothesis, we will have
2-+2-+
Xe Xf
... +~<3
a
(1)
But
1 1 1 1 1 1
;;: > lOa + lOa + 1 + ... + lOa + 9 > "X;;- + X, + .. .
since the numbers xk,xz, ... are among lOa, 10a+ 1, ... ,10a+9. But then,
from (1) we get
and
1 1 1 11 1 m
am+n =an- - - - - - ... - - - - <an- - - - - ... --=an- -
an an+l am+n-1 an an an an
But if a• < v'28, then by applying the above relation we would have ak+2B <
a• - ~ = a~ - 28 < 0, and if k s 1598 this would contradict the assumption.
ak ak
Therefore, a• <". v'28 for all2 1 s k 2
s 1598.
1 21 255
But we know that an+l = an +~- 2 S an + 2s - 2 S an - 28" By
successively applying this inequality for n = 1, ... , 1598 and summing all the
inequalities up, we have that:
Algebra. 135
2 2 55 2
ai 5 ga <a, - 1597 · 28 =? a 1598 s 56 2 - 3136.96 <0
which is obviously impossible. We thus have a contradiction, and there will
s s
exist a k with 1 k 1626 such that < 0. a•
25. M Let us assume that f = ax2 + bx + c and g = a' x 2 + b' x + c! and that
the leading coefficients a and a' are positive (without loss of generality). Take
k = l~ j and then whenever f(x) E Z, we will also have h = g - kf E Z. Let
h = ax2 + {Jx + 'Y and note that 0 Sa< a.
-b+ yb2 -4ac+ 4an
Now, for n large enough, the number Xn =
2a will be
real. Note that
1< 1 .~.y'b2-4ac+4an=?a<a
- yb2-4ac+4an a -
This contradicts our assumption that a: < a, so we can only have a: = 0.
But then the above argument also shows that the second coefficient of h is 0,
so h must be an integer constant. And thus, we proved that g = kf + c, where
k and c are integers.
26. M First of all, by taking y = -1 we get f(-1) = f(x) + f(-x) + f(-1) =?
f(-x) = -f(x), so f is an odd function. But the hypothesis is
have that f(a) + f(b) = 2f (a; b). It is easy to note that this holds even if
a + b + 2 =f 0, because then we just switch a and b in the previous argument
and get a valid result.
But we can apply this back into the hypothesis, and we have that
Now, replace in the hypothesis y by f ( ~) and (1) tells us that we will have
(2)
2ya~<>_-_4_
> __
- al+ ... +an-1
n-1
+4I>-I:,
(
1) n-1
=
.
fn,a,.(a1, ... ,an) 2':. fn-l,a,._ 1(ai, ... ,an-1) 2':. ... 2':. /3,a 8 (a1ia2,a3)
Yet we can prove by induction that <>n-i :::; ( n - i - 1)2 for all 0 :::; i :::; n - 3.
Indeed, this is verified for i = 0, and if we have that <>n-i+l:::; (n - i) 2 then
(T2 - Tl + T• - Ta+ ... + T2• - T,._1) 2 :::; L(T; - r,_ 1)2 '*
i$j
<* (T2 + ... + T2•) 2 +(Ti+ ... + T2k-1) 2 - 2(T2 + ... + T,.)(T1 + ... + T2•-1) :::;
:::; 2k(T{ + ... + T2k') - 2 LT,T; #
i<j
Algebra. 139
i<j
i<j i<j
which is true, since 4LT2,T2; ::; 2(k - l)(T? + ... + ~.) and
i<j
4 LT,._1T2;-1 ::; 2(k - l)(Tf + ... + T~L 1 ). Now, we need to get from this
i<j
special case to the general case. We will assume that S is composed of blocks of
consecutive integers which we will denote by B,, B 4 , ••• , B 2•. But between blocks
B2i and B2i+2 there will exist consecutive indices not in S, and let B2i+1 be that
block. I.et Bi and B2•+• be the blocks before B 2 and after B 2k respectively.
If b, will be the sum of the numbers in block B;, then the sum in LHS will
be (b2 + b4 + ... + b2k) 2. From the particular case, we know that this is at most
the sum of squares of the form (b; + ... + b; ) 2. But it is easy to see that all these
terms are among the terms in the RHS of what we have to prove (alongside
other non-negative squares). Therefore, the inequality is proven.
31. E Let the roots of the polynomial be -a1, ... ,-an. Since all the coefficients
are non-negative, f will always be positive for positive values of x, so it does
not change sign on [O, oo). Therefore, all its roots will be negative, and f(2) =
(2 + ai)(2 + a2) ... (2 +an), where the a,'s are positive numbers whose product
is 1. But
f(2) = t
k=O
(2n-k . L.
l$i1< ... <ik:5:n
a;,a;, ... a;.)
By applying the AM - GM inequality to the numbers a;, a;, ... a;, we get
that L a;, a;, ... a;, 2: (~) and therefore we will have
l$i1< ... <ik$n
J(2) 2: t
k=O
2n-k (~) = 3n
32. M We will prove by induction over k the following statement: for any such
sequence given by ao = 0, ai = 2a + 1 and an= (4x + 2)an-1 + (2y + l)an-2
(where a, x, y E Z), and for all n between 1 and 2• we have that ~ is an
n
irreducible fraction with both numerator and denominator odd. It is trivial for
k = 1 and now, let us assume it true for k - 1 and prove it for k.
First of all, an =2 an-2 so it is easy to prove by induction that for all
even n, an is even and for all odd n, an is odd. Therefore, we have solved our
problem for odd values of n and now we must prove it for even values. Since
140 Solutions
for all 1 ::; m ~ 2k-l the number a2m is even, let a2m = 2bm. Now, we want
to find a recurrence for the sequence bm; if r, s are the roots of the equation
z2 - (4a: + 2)z - (2y + 1) = 0, then an is of the form c1rr + c2r!,'. Therefore,
bm -_a2m_.=!..2m ~2m
2 - 2 r1 + 2 r2
and bm will be given by a second-degree linear recurrence, the roots of
whose characteristic equation are rf and r~. That recurrence will be
a 3 + b3 + c3 - i= (1 - b- c) 3 + b3 + c3 =
= -3(b- l)(c- l)(b+ c) = 3(a - l)(b - l)(c- 1) (1)
Now, to do the same thing for g(X, Y) = (1- X - Y) 5 + X 5 + Y 5 - 1. Fbr
the exact same reason, the polynomials X - 1, Y - 1 and X + Y will divide g,
and because g has degree 5 it will be equal to (X - l)(Y - l)(X + Y) times
some polynomial of degree two. So we will have
(1 - X - Y) 5 + X 5 + Y5 - 1=
= (X - l)(Y - l)(X + Y)(-5X2 - 5Y 2 -5XY + 5X + 5Y - 5) =
= -5(X -l)(Y- l)(X + Y)(X 2 + Y 2 +XY-X-Y+1)
Therefore, letting X = b, Y = c and X + Y = 1 - a we will have a 5 + b5 +
c5 - 1 = 5(a - l)(b- l)(c - l)(b2 + c2 +be - b - c + 1), and what we need to
prove is that
30l(a - l)(b- l)(c- 1)1 ~ 45l(a - l)(b- l)(c- l)llb2 + c2 +be- b- c + ll (3)
s2
But for all b, c, by denoting b + c = S and be = P we will have IPI ~ 4
and therefore
b2 + c2 + be-b- c+ 1 = s2 - S + 1- P 2:
> 3S2 _ 8 + ! + ~ = ( sva ___!__) \ ~>~
- 4 3 3 2 y'3 3-3
By using this inequality, (3) is obviously true, and our problem is thus
solved.
34. SOLUTION BY ANDREI UNGUREANU
D We will show that the answer is no, so assume for the purpose of contra-
diction that there is such an f. Then, f(2) = 2 as it can be seen from letting
x = 1, and let 2 +a =sup lf(x)I, where a;::: 0 (we know that sup;::: 2, since
f(2) = 2 and that it is finite since the function is bounded) 2Assume that a > 0,
and then pick x such that lf(x)I = 2+a-g, where e < 2 (2a+ a) (this is possible,
1G);:::1G)+42:2
142 Solutions
and since f (~) :::; 2 the above can only happen when f (~) -2.
Letting a: = 2 gives us
f (~) = 6
which is impossible. Therefore, there are no such functions.
35. M First of all, note that f is one-to-one, because f(n) = f(p) obviously implies
n = p (from the hypothesis, setting m = 1). Now, we will have
them, we must have f(l) - 2005 = 1 aud then our function will be f(m) =
m+2005.
36. E What we need to prove is equivalent to
~f(ab) 1 1) =f (b-a)
(;;--i; - - f (a-b)
2-- 2- ~
f(ab)b-a = 2 f(b-a)
ab 2
~ f(ab)
ab
= 1 C;a)
b- a
(1)
-2-
for all 0 f a f b f 0 (because we divided by a, b and b - a). But for what
positive a: and y can we find such a and b with ab = a: and b; a = y? We need
to have b = :'. aud
a
:'. - a = 2y ~ a 2 + 2ay - a: = 0
a
and this always has a real solution. Therefore, for all positive a: and y we
will have f(x) = f(y). So this expression is constant, aud therefore for all
a: y
144 Solutions
positive x we will have f(x) =ax. Because the function is odd, this will hold
for all x and this will be our solution (which obviously satisfies).
38. SOLUTION BY ANDREI STEFANESCU
D We call any polynomial P(X) = anXn+ ... +ao symmetrical i:ff a;= an-i
for all i. We will use the following lemma:
LEMMA If P is a polynomial which can be written as P = Q · R, where Q
and R are symmetrical polynomials, then P is symmetrical too.
Let Q(X) = b,X• t ... + b0 and R(X) = c,X' + ... + c0 • We have that
i i
ai = E bjCi-j and an-i = E bs-jct-(i-j)· Since Q and Rare symmetrical, it
j=O j=O
follows that b; = bs-j and Ci-j = Ct-(i-j)· Hence ai = an-i for any i, and thus
P is symmetrical.
Let f(X) = xn
+ xn- 1 ••• + 1 and assume that f = g · h, where g and h
are nonconstant polynomials with nonnegative coefficients. Let g(X) = b,X' +
... + bo and h(X) = c,X' + ... + co. Since f = g · h, all the complex roots
of g and h have absolute value 1. Also, if g(<) = 0, then g(<) = g (;) = 0.
It follows that both g and h are a product of symmetric terms of the form
(X - <) ( X - ; ) = X 2 - 2Re(<) + 1orX+1, hence by lemma g and hare
symmetrical.
If all the coefficients of g and h are in {O, 1} we are done. Otherwise, let k
be the least number such that {h, ck} g; {O, 1}. Since b, = c, = 1, it follows
that bo = CQ = 1, thus k 2: 1. As g is symmetrical, bk = b,_k· Computing the
coefficient of x·
gives us 1 = b,co + ... + bs-kCk + ... + boc,. Since b, = C(j = 1
and all the terms of the sum are nonnegative, we obtain bs-kCk = b1;ck = 0,
thus one of h and ck (say ck) is 0. Computing the coefficient of X\ we ha,ve
1 = hco + ... +bock. As {b;, c;} ~ {O, 1} for i < k and Ck= 0, all the terms
but bkCo must be in {O, 1}. It follows that bkCQ =bk must be in {O, 1}, which
leads to a oontradiction with the definition of k.
APPENDIX: Useful facts 145
1. APOLLONIAN CIRCLES Given a segment AB, and a positive number k, the locus
of the points M such that Z
~ = k is a circle with the center on AB, called
one of the Apollonian circles of AB. More precisely, let E and F be the points
inside and respectively outside the segment AB such that ;~ = ;~ = k. We
will show that our Apollonian circle is the circle of diameter EF.
EA MA FA
To do that, note that because EB = M J3 = F B, then ME and MF must
be the two bisectors of the angle A in the triangle ABC. But the two bisectors
are always perpendicular, so LEMF = ~ and therefore M lies on the circle
of diameter E F. Proving the converse, that every point on that circle satisfies
M ~ = k, is just a computation with angles and the Law of Sines.
2. ~RYCENTRlC COORDINATES Given a triangle ABC, we can define the barycen-
tric coordinates of a point X inside it as the triplet (xa,Xb,Xc) with Xa = ~
and similar definitions for b and c. Here, d0 is the distance from X to BC
and h 0 is the altitude from A. Note that '"• = \~;~ and because the
areas [XAB], [XBC] and [XAC] sum up to [ABC], we will always have
Xa+Xb+Xc= 1.
So it turns out that every point inside the triangle only depends on two
of these coordinates, the other being implied by the fact that the sum of all
three is 1. An important fact is that given X, Y, Z inside the triangle ABC
(determined by coordinates (x 0 , Xb, x.), (y0 , Yb, Ye) and (z0 , Zb, zc) respectively),
X, Y, Z are collinear only if the coordinates of one of them is a convex sum of
those of the other two. The fact that the coordinates of Y to be a convex sum
of those of X and Z means that there exists an a between 0 and 1 such that
Yb = O<Xb + (1 - a)zb
The proofs of both these theorems and the proof for the formula of Euler's
function can be found in any elementary number theory book or course.
11. FORMAL DERIVATIVE OF A POLYNOMIAL Though this notion comes from anal-
ysis, one can understand it even within the range of elementary mathematics.
Indeed, given any polynomial
By using this property, one can prove by induction over i that a certain x 1
is a root of P of multiplicity i (meaning that the term (X - xi) appears i times
in the factorization of P) iff P(x 1 ) = 0, pCl)(x 1) = 0, ... , p(i-l)(x 1) = 0 and
pCi)(x1) fO.
APPENDIX: Useful facts 149
S abc ~-~~~~---,-
r = p,S= 4R,S= y'p(p-a)(p-b)(p-c)
BMBN AB2
CMCN AC2
19. LHS "Left Hand Side" of an equation, inequality, identity etc.
20. LINEAR RECURRENCE RELATIONS Consider a sequence (xn)nENo which satisfies
the recurrence Xn+k = ak-lXn+k-1 + ak-2Xn+k-2 + ... + aoxn (called a k-th
order linear recurrence). We can generally determine the form of Xn in terms of
n, the first k terms x 0 , ••• ,x1;_ 1 and the numbers a 0 , ... , a1;_ 1 . More precisely, if
r1, ... , Tk are the roots of the polynomial xk - a1;-1Xk-l - ak-2xk- 2 - ... - ao
APPENDIX: Useful facts 151
(called the characteristic polynomial of the recurrence) and if all the roots are
distinct, then
where the constants c 1 and co are determined by the terms xo and x1.
21. ORDERS AND PRIMITIVE RoOTS IN Zn Given an element a E Zn with (a, n) =
1, call the order of a the smallest non-zero integer d such that a• =n 1. Such a
number always exists, as Euler's Theorem guarantees that a'l'(n) =n 1. Probably
the most important thing about orders is the following fact:
LEMMA If dis the order of a modulo n and e is such that a• =n 1, then die.
Indeed, assume that d does not divide e, and therefore e = kd + r for 0 <
r < d-1. Then, from the definition of the order we will have a•• =n (a•)• =n 1,
but also ae =n adk+r =n 1. So we will have ar =n 1, but this would mean that
r is smaller than the order d (which contradicts the minimality of d). Therefore
r = 0 and die.
Euler's Theorem tells us that a corollary of this lemma is that dl'f'(n),
because alP(n) is always =n 1. Here arises the concept of a primitive root: g is
called a primitive root of n if the order of 9 modulo n is 'f'(n) (so the largest
possible). There is a theorem that says that all numbers of the form pa and 2pa
for prime p have primitive roots.
In the case when n = p is prime, a primitive root 9 has the order 'f'(p) =
p- 1. Therefore, it is easy to note that the numbers 1,9,92 , •.. ,g"- 2 are all
different modulo p (as p-1 is the smallest exponent to which we can raise 9 to
reach 1). But then, the residues 1,9,92 , ••• ,g"- 2 will be the same as the residues
1, 2, ... ,p - 1 (in some order).
22. PICK'S THEOREM Given any polygon (not necessarily concave) with vertices in
lattice points, its area is p + ~ - 1, where pis the number of interior points and
q is the number of points on the boundary of the polygon (including vertices).
This can be proven by triangulating the polygon with respect to all its
inside lattice points, and noting that the triangles which appear have no lattice
points inside or on their edges. The basic exercise concerning lattices is to show
152 APPENDIX: Useful facts
that any such triangle (with vertices in lattice points and no lattice points
on the boundary or inside) has area exactly ~· So if we show that there are
2p + q - 2 triangles in our triangulation, each of them will have area i and
Pick's Theorem will be proven.
But we can show that the total number of triangles in a polygon with
q vertices (we can consider the lattice points on the edges as 'vertices') and
triangulated with respect to p interior lattice points is 2p + q - 2 by induction
over 2p + q. This is rather simple to do, because note that upon removing a
triangle from the polygon, the quantity 2p + q in the remaining polygon drops
by 1.
23. PLANAR GRAPH A planar graph is a graph that can be drawn in the plane
without any intersections between its edges (except in vertices, of course). It is
now well-known that the planar graphs are just the graphs that do not contain
any sub-graph of the form K 5 or K.,a, but the proof of this fact goes is both
very difficult and beyond elementary graph theory.
A planar graph is like a map in the plane, and therefore the notion of
"region" is well-defined, as the vertices and edges of the graph break the plane
up into a number of regions. If v is the number of vertices, e that of edges and r
that of the regions of a planar graph (including the infinite region), the Euler's
formula v + r = e + 2 holds.
The chromatic number of a planar graph is just the minimum number of
colors needed to color its regions such that no two adjacent regions (regions
that have a common side) are colored in the same way. The four-color problem,
solved only recently, tells us that every planar graph has chromatic number at
most 4, meaning that 4 colors are always enough to color any plane map (such
that any two adjacent regions have different colors).
A very important concept in planar graphs is that of the dual graph. Take
a planar graph G, place a vertex V inside every region R of G (including the
infinite one) and through each edge E of the graph connect the two vertices V
and V' which you have constructed in the regions that touch the edge E (thus,
since two regions in the graph can have more common edges, two vertices in the
dual graph can have more edges between them). The resulting graph G' will
also be planar, and every region of G corresponds to a vertex of G' (since there
lies one vertex of G' in every region of G) and every vertex of G corresponds to a
region of G' (since every region of G' contains only one vertex of G). MoreoVer,
the degree of every vertex of G equals the number of edges of the corresponding
region on G' and vice-versa. Note that the planar graph of the planar graph of
G is G itself.
APPENDIX: Useful facts 153
(p-1) =(-1)-.-->='
(p2) (-l)'r
= ''
( ~) = (~) G) for any integers a,b
(1)
APPENDIX: Useful facts 155
A"O A'A"
sinOA'M = sinA'OA"
C"O C'C"
sinOC'M = sinC'0C"
Dividing the two of them gives us (LA'OA" = rr - LC'OC")
That lsst ratio, is by the Law of Sines in triangle A' MC', equal to MC'.
MA'
It is essy to see that we have LAA" A' = LCC"C = B and so the triangles
156 APPENDIX: Useful facts
AA'A" and CC'C" are similar. Moreover, the triangles AMA' and CMG' are
similar (because they are right-angled and LA'AM = rr-LBAM = LC'CM).
Thus, because of the above similarities, ( 1) becomes
n~ 1 (~).
Another interesting thing is triangulating a certain polygon whose set of
vertices is V with respect to a certain finite set A of interior points. This means
that we must break up the polygon into triangles with disjoint interiors which
only have vertices among the points of V LJ A, such that the points of VU A
are located only in the vertices of triangles of the triangulation (so not on the
edges or in the interiors). It can be easily proven (by induction on #VLJA)
that any polygon can be triangulated with respect to any finite set of interior
points.
31. TuRAN's THEOREM This theorem tells us that the maximum number of edges
in a graph with n vertices with no p-clique is obtained when our graph is
the complete multipartite graph K1,1, .. .,!+1,. . .,!+1 where n = (p - l)! + r with
0 :::; r < p - 1. That multipartite graph is a graph whose vertices are broken
up into p - 1 disjoint groups, each with l or l + 1 vertices, such that a vertex
is connected only to vertices from the other groups. This graph is practically
APPENDIX: Useful facts 157
Radu Gologan
l.S.B.N. 973-9417-52-3