SOLUTIONS TO THE YEAR 2000

INTERNATIONAL MATHEMATICAL OLYMPIAD

Ilan VARDI

H

Institut des Hautes Etudes S ientiques
35, route de Chartres
91440 { Bures-sur-Yvette (Fran e)

De embre 2000

IHES/M/00/80
 Solutions to the year 2000 International Mathemati al Olympiad

Ilan Vardiy

This arti le is meant to omplement my earlier paper on Mekh{mat entran e examinations [9℄. In that
paper I stated that the Mekh{mat problems appeared to be at the level of Olympiad problems whi h were an
appropriate standard for omparison. In order to verify this laim, I de ided to solve under similar onditions
(without use of outside referen es or help) all the problems for a given Olympiad year. Doing all problems
in a given year appears to be the fairest test of problem solving, as one annot simply hoose the problems
one is best at. Finally, sin e many people work on Olympiad problems, one an ompare one's solution with
the best that has been found (this was not always possible with the Mekh{mat problems).
My on lusion is that the IMO problems are generally mu h more hallenging than the Mekh{mat
problems. In parti ular, it took me about 6 weeks to omplete all the Olympiad problems and 6 weeks for
all the Mekh{mat problems. However, there were 6 Olympiad problems versus 25 Mekh{mat problems.
On the other hand, the Olympiad problems were in general less interesting than the Mekh{mat problems.
In fa t, I only found Olympiad problems 3 and 6 to be interesting. Problem 3 required an interesting idea
whi h took me a long time to dis over (3 weeks!), and it also took me a long time to nd an idea for 6,
whi h follows from a result of some independent interest (Lemma 6.3).
The other problems were solved in a relatively short time without too mu h diÆ ulty. If there was an idea
in 2 then I missed it as my solution onsists of a ompletely unmotivated algebrai manipulation. Problem 5
is stated onfusingly and has an alternate, possibly intra table interpretation. The la k of larity might have
been intentional in order not to give away the solution, see Remark 5.1.
The statements of the geometry questions 1 and 6 are of Byzantine omplexity, and this appears to
be typi al of the geometry questions posed in the IMO. Moreover, Problem 1 is in some sense de eptive.
In omparison, the Mekh{mat geometry problems were very elegant and in general interesting. Mekh{mat
problems 1 and 2 were of omparable diÆ ulty to the geometry problems posed here.
Sin e the sour e of the Mekh{mat problems [8℄ only in luded one number theory question and one
ombinatorial question, it is less meaningful to ompare the other problems to the Mekh{mat questions.
I would like to thank K.S. Sarkaria and Harry Tamvakis for helpful omments. Alternate solutions appear
on John S holes' IMO web site [7℄. Further IMO problems are given in [1℄ [3℄ [5℄ [6℄.

 41st IMO, Taejon, South Korea. International Mathemati al Olympiad site: olympiads.win.tue.nl/imo/
y Preliminary version, November 2000. IHES, 35 route de Chartres, 91440 Bures{sur{Yvette, ilanihes.fr

1
1 Problems
Day 1

1. Two ir les 1 and 2 interse t at M and N .

Let ` be the ommon tangent to 1 and 2 so that M is loser to ` than N is. Let ` tou h 1 at A and
2 at B . Let the line through M parallel to ` meet the ir le 1 again at C and the ir le 2 again at

D.
Lines CA and DB meet at E ; lines AN and CD meet at P ; lines BN and CD meet at Q.
Show that EP = EQ.
2. Let a; b; be positive real numbers su h that ab = 1. Prove that
   
1 1 1
a 1+
b
b 1+

1+
a
 1:

3. Let n  2 be a positive integer. Initially, there are n eas on a horizontal line, not all at the same point.
For a positive real number , dene a move as follows:

hoose any two eas, at points A and B , with A to the left of B ;

let the ea at A jump to the point C on the line to the right of B with BC=AB = .

Determine all values of  su h that, for any point M on the line and any initial positions of the n eas,
there is a nite sequen e of moves that will take all the eas to positions to the right of M .

Day 2

4. A magi ian has one hundred ards numbered 1 to 100. He puts them into three boxes, a red one, a
white one and a blue one, so that ea h box ontains at least one ard.
A member of the audien e sele ts two of the three boxes, hooses one ard from ea h and announ es
the sum of the numbers on the hosen ards. Given this sum, the magi ian identies the box from
whi h no ard has been hosen.
How many ways are there to put all the ards into boxes so that this tri k always works? (Two ways
are onsidered dierent if at least one ard is put into a dierent box.)
5. Determine whether or not there exists a positive integer n su h that n is divisible by exa tly 2000
dierent prime numbers, and 2n + 1 is divisible by n.
6. Let AH1 , BH2 , CH3 be the altitudes of an a ute{angled triangle ABC . The in ir le of the triangle ABC
tou hes the sides BC , CA, AB at T1, T2 , T3 , respe tively. Let the lines `1 , `2, `3 be the re e tions of
the lines H2 H3 , H3 H1 , H1 H2 in the lines T2 T3 , T3 T1 , T1T2 , respe tively.
Prove that `1 , `2 , `3 determine a triangle whose verti es lie on the in ir le of the triangle ABC .

2
2 Solutions
Problem 1. Two ir les 1 and 2 interse t at M and N .
Let ` be the ommon tangent to 1 and 2 so that M is loser to ` than N is. Let ` tou h 1 at A and
2 at B . Let the line through M parallel to ` meet the ir le 1 again at C and the ir le 2 again at D .

Lines CA and DB meet at E ; lines AN and CD meet at P ; lines BN and CD meet at Q.

Show that EP = EQ.

P Q

Solution: The somewhat ompli ated formulation of this problem is de eptive, and it is easily seen to be a
ombination of two simple results.
One starts by looking at a subproblem whi h an be analyzed by onsideration of similar triangles, that
is, the points A; B; C; D; E; M . In order to do this, one introdu es 3 new points U; V; W , whi h are the
midpoints of CM , MD and AB , respe tively.
C U M V D

A W B

Sin e CM is a hord of the ir le 1 and A lies on 1 , it follows that AC = AM . Thus, 6 MCA = 6 CMA.
Sin e CM is parallel to AB , 6 CMA = 6 BAM .
It follows CU = UM = AW . Similarly, one has DV = V M = BW . One therefore sees that CD =
2CU + 2DV = 2AB .
But, sin e CD and AB are parallel, the triangle CDA is similar to the triangle ABE , so it follows
that CE = 2 AE , in other words, that CA = AE . Finally, sin e 6 BAE = 6 BAM , one sees that EM is
perpendi ular to AB . The rst result is therefore
Lemma 1.1. Under the assumptions of Problem 1, EM is perpendi ular to AB .
This result laries the situation onsiderably sin e it is now lear that EP = EQ if and only if MP =
MQ, sin e CD is also perpendi ular to EM .

3
 P Q

A X B

One therefore turns to the proof of MP = MQ. Extend the line MN to interse tion AB at X . Sin e
CD is parallel to AB , it follows that triangle NP M is similar to triangle NAX and that triangle NQM is
similar to triangle NBX . The result therefore follows from the following interesting fa t.
Lemma 1.2. Let a line be tangent to two distin t ir les at two dierent points A and B . Then the line
joining the interse tions of the two ir les (or the tangent line if they tou h on e) bise ts the line segment
AB .

A B

Proof: The proof uses some elementary properties of inversion, see [2℄. Let the notation be as in Problem 1,
i.e., let 1 ; 2 be the ir les, M and N be the interse tion points and X the midpoint of AB .
Consider the ir le with enter at X and radius AX . One does an inversion with respe t to this
ir le. Sin e 1 and 2 are orthogonal to (this is obvious at A and B ), they are preserved. Sin e the line
ontaining MN passes through innity, it is transformed into a line or ir le passing through X . However,
sin e 1 and 2 are preserved, M is mapped to N and N is mapped to M . Finally, let the line MN interse t
at Y , then Y is xed, sin e it lies on . Thus, the line MN goes to a line or ir le ontaining X , M , N ,
and Y . However, M , N , and Y lie on a line, therefore MN goes to a straight line, in parti ular to itself.
One on ludes that the line MN passes through X , whi h is the midpoint of AB and the result follows. The
ase in whi h the ir le are tangent is left as an exer ise.
Remark 1.1. Harry Tamvakis has noted that Lemma 3.2 follows easily using the power of a point with
respe t to a ir le. In other words, it is easily shown that if the point X is external to a ir le , then for
any line XMN tou hing the ir le at M and N , then XM  MN is a onstant (in the ase in whi h the
line XA is tangent to the ir le, one takes XA  XA).

4
Problem 2. Let a; b; be positive real numbers su h that ab = 1. Prove that
   
1 1 1
a 1+
b
b 1+

1+
a
 1:

Proof: The proof relies on the following identity.

Lemma 2.1. Let a; b; be positive real numbers su h that ab = 1, then
   
1 1 1
1 a 1+ b 1+ 1+
b a
(1)     
1 1 1
= + 2 a+ 1 + 1 (1 b) :
b a

This identity easily implies the result. To see this, note that if ab = 1, then at least one of a; b; is  1
and at least one is  1. It follows that at least one of the ordered pairs (a; b), (b; ), or ( ; a) has its rst
element  1 and the se ond  1. One an, without loss of generality, take this ordered pair to be (a; b). The
reason is that Lemma 2.1 will be applied, and this result also holds with the variables on the right hand side
of (1) repla ed by (a; b; ) 7! ( ; a; b) and (a; b; ) 7! (b; ; a), sin e the left hand side of (1) is invariant under
these y li shifts.
One now applies Lemma 2.1. First one uses the arithmeti {geometri inequality whi h gives
  r
1 1
2
+


1 = 1 ;
so that
1
+ 2  0:

Sin e a  1 and b  1, one has
1 1
a+ 1  0; 1  0; 1 b  0 :
b a
It follows that ea h term on the right hand side of (1) is non{negative whi h proves the result.
Proof of Lemma 2.1: One expands
   
1 1 1
a 1+ b 1+ 1+
b a
using the transformation rules ab 7! 1= , a 7! 1=b, b 7! 1=a. Thus
  
1 1 1 b 1 1
b 1+ 1+ = b+ +1 +1 +b
a a a a
b 1
= +2 :
a
One ontinues with
  
1 b 1 1 a b 1 1 2
a 1+ +2 = b + 2a 2+ + + + a
b a b a a b b
1 1 1 a b
=a+ +b+ + + 2:
a b b a

5
Thus, the left hand side of (1) equals
a b 1 1 1
(2) 3+ + + a b :
b a a b
One now omputes
  
1 1 1 a 1 2
+ 2 a+ 1 = + + +a 2a +2
b b b b
(3)
a 1 1
=2+ + a ;
b b
and  
1 1 b
(4) 1 (1 b) = 1 b + :
a a a
It is seen that the right hand sides of (3) and (4) add up to the right hand side of (3), whi h proves Lemma 2.1.
Remark 2.1. This proof appears to be ompletely unmotivated. In parti ular, there does not seem to be a
on eptual explanation for the identity of Lemma 2.1. However, this identity ts into the general prin iple
that inequalities should follow from identities, e.g., from the positivity of perfe t squares [10℄.
Remark 2.2. A similar argument was found by Robin Chapman [7℄ using the identity
    2 !
1 1 1
a 1+ b 1+ = b a2 1 :
b b
This shows that the left hand side is  b a2 , so the produ t of the three su h identities is  b a2 b2 a 2 = 1,
whi h yields the square of the desired inequality. This works if all terms in the produ t are positive, but it
is easily shown that at most one su h term an be negative, in whi h ase the inequality also holds.
Remark 2.3. One an try the general method of Lagrange multipliers to solve this problem. However, this
method is usually onsidered beyond the s ope of the High S hool urri ulum.

Problem 3. Let n  2 be a positive integer. Initially, there are n eas on a horizontal line, not all at the
same point. For a positive real number , dene a move as follows:

hoose any two eas, at points A and B , with A to the left of B ;

let the ea at A jump to the point C on the line to the right of B with BC=AB = .

Determine all values of  su h that, for any point M on the line and any initial positions of the n eas, there
is a nite sequen e of moves that will take all the eas to positions to the right of M .
Solution: All values   1=(n 1).
Proof: (a) First assume that   1=(n 1). Call a jump of (any) leftmost ea over the rightmost ea a big
move. Then the eas will all go to +1 using a su ession of big moves.
First, assume that the eas all at distin t positions, let xn < xn 1 <


< x1 be the position of the eas,
and let m be the minimum distan e between any two onse utive eas.
One now performs a big move whi h puts the eas at positions x0n < x0n 1 <


< x01 , where xn = 0

xn 1 ,. . . ,x02 = x1 , and x01 = x1 + (x1 xn ).

6
 Let m0 be the minimum of the dieren es between onse utive eas in the new position. Sin e x0n ; : : : ; x02
are the same as in the previous position, it follows that the new minimum satises m0  min(m; x01 x02 ).
However, x01 x02 =  (x1 xn ) and x1 xm  (n 1) m. It follows that m0  min(m; (n 1)m)  m,
sin e (n 1)  1, by assumption.
It follows that the minimum distan e between onse utive eas is nonde reasing. Thus, let M > 0 be a
lower bound for all these minimum distan e. It follows that x0n  xn + M , i.e., the rightmost ea moves at
least distan e M to the right at every iteration. This shows that all the eas go to +1.
If the eas are not all distin t, then, sin e at least two eas are distin t, it is lear that after at most
n 1 iterations of big moves, all eas will be ome distin t, whi h redu es to the previous ase.
(b) Assume that  < 1=(n 1). This seems to require some kind of physi ial reasoning, as dire t methods
su h as those in part (a) do not seem to work, see Remark 3.1.
Thus, one sees that there is a tradeo between the position of the rightmost ea and the distan e between
the eas. In other words, a greater distan e to the right in urs a penalty in the distan e between the eas.
Thus, one is led to the onsideration of an energy to ea h pla ement of the eas, where the position of the
eas is the kineti energy, and the distan es between the eas is the potential energy.
By areful examination of the ase of two eas, one is led to the following denition.
Denition. Let the position of the eas be x = (x1 ; x2 ; : : : ; xn ), where x1  x2 


 xn , then the kineti
energy K (x) and the potential energy P (x) are
n
X
K (x) = x1 ; P (x) =  (x1 xi ) ;
i=2
where  is a onstant. The total energy is E (x) = K (x) + P (x).
This leads to the following.
Lemma 3.1. If  = =(1 (n 1)), then the energy is never in reased by a move.
The main result follows dire tly from Lemma 3.1. Indeed, let P (x) = P=(1 (n 1)) (x). Sin e  < 1=(n 1),
one has =(1 (n 1)) > 0 so that the potential energy is always positive. The Lemma shows that
K (x) + P (x)  E for some E independent of x, thus K (x)  E P (x)  E for all x, and the result follows.
Proof of Lemma 3.1: First onsider the ase of a big move. Thus, the eas are initially at xn 


 x1
and xn jumps to x01 = x1 + (x1 xn ), while the other eas remain at the same positions, i.e., x0n =
xn 1 ,. . . ,x02 = x1 .
The hange in potential energy is
Xn Xn
P (x0 ) P (x) =  (x01 x0i )  (x1 xi )
i=2 i=2
nX1 n
X
= [(x01 x1 ) + (x1 xi )℄  (x1 xi ) = (n 1) (x01 x1 ) (x1 xn )
i=1 i=2

= [(n 1) 1℄(x1 xn ) :

Moreover, the hange in kineti energy is
K (x0 ) K (x) = x01 x1 = (x1 xn ) :

7
Thus, the dieren e in total energy is
E (x0 ) E (x) = K (x0 ) K (x) + P (x0 ) P (x) = (x1 xn ) + [(n 1) 1℄(x1 xn )
= (x1 xn )( + [(n 1) 1℄) ;
whi h is learly zero when  = =(1 (n 1)). One has therefore proved
Lemma 3.2. If  = =(1 (n 1)), then the total energy is preserved by a big move.
One now onsiders other types of moves. The rst possibility is if x01 = x1 , i.e., the position of the
rightmost ea does not hange. In this ase, the kineti energy stays the same, while the potential energy
de reases, for any  > 0. This is obvious, sin e

P (x0 ) P (x) = (xi x0j ) < 0 ;

where the move onsisted of ea xi jumping to x0j , and x0j > xi .

Finally, one onsiders the possibility that x01 > x1 , but that this is not the result of a big move, i.e., the
ea x01 was previously at xj , where xj > xn . One performs the same omputation as with a big move. The
hange in potential energy is
X n n
X
P (x0 ) P (x) =  (x01 x0i )  (x1 xi )
i=2 i=2
X n
X
= [(x01 x1 ) + (x1 xi )℄  (x1 xi ) = (n 1) (x01 x1 ) (x1 xj )
1in i=2
6 j
i=

 [(n 1) 1℄(x 1 xj ) ;

sin e x01 x1  (x1 xj ), as xj ould at most jump over x1 . Similarly, the hange in kineti energy is

K (x0 ) K (x) = x01 x1  (x1 xj ) ;

so that the total hange in energy is

E (x0 ) E (x)  ( + [(n 1) 1℄)(x1 xj ) = 0 ;

if  = =(1 (n 1)). This shows that the total energy is nonin reasing, whi h nishes the proof of
Lemma 3.1.
Remark 3.1. This type of solution was previously dis overed by Gerhard Woeginger [7℄.
Remark 3.2. It an be shown dire tly that the strategy used in part (a), i.e, using only big moves, does
not take the eas to +1. In fa t, a big move is a linear transformation x0 = Ax, where A is the matrix
0 1
1+ 0


0 
B 1 0


0 0 C
B
B C
C
A=B
B 0 1


0 0 C
.. C
:
B .. .. .. C
 . . . . A
0 0


1 0

8
This matrix has hara teristi polynomial (x 1)(xn 1 (xn 2 + xn 3 +


+ 1)), whi h has one eigenvalue
1 while all other eigenvalues have absolute value stri tly less than 1, when  < 1=(n 1).
Using this method, or equivalent methods, it is simple to show that x1 xn ! 0 exponentially, i.e.,
the distan e between the rightmost ea and the leftmost ea goes to zero exponentially. Sin e, in a simple
iteration of big moves, the position of the rightmost ea is essentially the sum of these distan es, one gets
that the position of the rightmost ea is bounded above.
However, this argument seems to break down if other types of moves are allowed. In parti ular, the
distan e between the rightmost and lefmost eas no longer has to go to zero, e.g., if the leftmost ea never
moves. Moreover, for moves other than big moves, the relabelling of the eas depends is no longer unique,
as a ea may land between any two others, to that the hoi e of matrix is no longer unique. Finally, even
without this problem, the asymptoti s of random matrix produ ts is a nontrivial question.

Problem 4. A magi ian has one hundred ards numbered 1 to 100. He puts them into three boxes, a red
one, a white one and a blue one, so that ea h box ontains at least one ard.
A member of the audien e sele ts two of the three boxes, hooses one ard from ea h and announ es the
sum of the numbers on the hosen ards. Given this sum, the magi ian identies the box from whi h no
ard has been hosen.
How many ways are there to put all the ards into boxes so that this tri k always works? (Two ways are
onsidered dierent if at least one ard is put into a dierent box.)
Solution: There are 12 ways. If the boxes are denoted R; W; B , then the 2 basi ongurations are
R = f1; 4; 7; : : : ; 100g; W = f2; 5; 8; : : : ; 98g; B = f3; 6; 9; : : : ; 99g;

and
R = f1g; W = f2; 3; 4; : : : ; 99g; B = f100g ;
and the 12 solutions onsists of the 6 permutations of ea h basi onguration.
Proof: Let the boxes be A; B; C . One rst notes that n 7! 101 n preserves f1; 2; : : : ; 100g, and in the
following, the expression \by symmetry" will in lude this map. One begins the proof with the following
useful result.
Lemma 4.1. If A, B , C satisfy the onditions of the problem, and a and a + k are in A, then it is not
possible to have b in B su h that b + k is in C .
Proof: In fa t, it is easily seen that this is equivalent to the onditions of the problem. Thus, if b 2 B and
+ k 2 C , then a + b + k is in A + B and in A + C , whi h is not allowed. 2
Lemma 4.2. If A, B , C satisfy the onditions of the problem, and A ontains a and a + k and box B
ontains b and b + k , then if is in C and 1  + k  100, then + k is in C .
Proof: One applies Lemma 4.1 twi e. Thus, applying it with respe t to A shows that + k annot be in B ,
and applying it with respe t to B shows that + k annot be in A. The result follows. 2
Lemma 4.3. Let A, B , C satisfy the onditions of the problem , then it is not possible to have a; a + k 2 A
and b; b + k 2 B , where k = 1 or k = 2.

9
Proof: Assume that su h a and b existed. Then C would have an element su h that 1  + k  100
or 1  k  100, so there would exist a 0 su h that 0 + k 2 C . It would follow from Lemma 4.2 that
1  a  rk  100, r 2 Z, are all in A and similarly, 1  b  rk  100, r 2 Z, are all in B , 1   rk  100,
r 2 Z, are all in C . If k = 1, then ea h set has 100 elements while if k = 2, ea h set has 50 elements, so A,
B , C do not form a partition of f1; 2; 3; : : : ; 100g. This ontradi tion proves the result. 2
The proof will now be split up a ording to the minimum of all the dieren es of two elements whi h are
in a same box.
(a) The minimum is 1. Without loss of generality, say that a; a + 1 2 A. Now either B and C are of the
form f1g and f100g, or there exists (without loss of generality) b 2 B su h that 1 < b < 100 and (by the
symmetry n 7! 101 n) a 2 C su h that < 100.
By Lemma 4.3, b + 1 and + 1 are in A. It follows that b + + 1 is in A + B and in A + C , whi h is not
allowed. This possibility is therefore ex luded unless B and C are of the form f1g and f100g.
(b) The minimum is 2. Without loss of generality, let a; a + 2 2 A. Sin e the minimum is 2, a + 1 is not
in A and, without loss of generality, let a + 1 2 B . Now either a + 3  100 or a 1  1. By symmetry
(n 7! 101 n), assume that a + 3  100. By Lemma 4.1, one annot have a + 3 2 C . By Lemma 4.2, one
annot have a + 3 in B , and by part (a), one annot have a + 3 2 A, sin e a + 2 2 A. This ase is therefore
ex luded.
( ) The minimum is 3. Sin e A, B , C form a partition of f1; 2; : : : ; 100g, one of these sets must have at least
34 members, sin e otherwise, one would have jAj + jB j + jC j  33 + 33 + 33 = 99. Without loss of generality,
let this set be A. Now split up A into 33 bins f1; 2; 3g, f4; 5; 6g,. . . ,f97; 98; 99g. Sin e the minimum dieren e
of elements of A is 3, ea h bin an have at most one element. But sin e there are more than 33 elements,
and only 100 is left over, ea h bin has one element, thus exa tly one element. Sin e there are at least 34
elements, it follows that 100 is in A. Sin e 100 2 A, one annot have 98 or 99 in A, and therefore 97 2 A.
Similarly, one has 94; : : : ; 7; 4; 1 2 A. It follows that A = f1; 4; 7; : : : ; 100g.
A similar argument shows that B and C are f2; 5; : : : ; 98g and f3; 6; : : : ; 99g in some order.
(d) The minimum is > 3. This is not possible, as one of the boxes has at least 34 elements, and thus
minimum dieren e  3, by the argument used in ( ).
Remark 4.1. Harry Tamvakis has found the following simpler argument. His argument begins by onsid-
ering the general problem with ards numbered 1 to n. One also shows that there are only two types of
solutions, i.e., one with f3a + kg, k = 0; 1; 2, and f1g, f2; : : : ; n 1g, fng, but the main point is that the rst
type of solution is stable under n 7! n + 1,(by adding the ard numbered n + 1 to the solution with n ards),
while the se ond is not. This observation yields a simple indu tion argument: Assume that the result holds
for n, then given a legal onguration with n + 1 ards, remove ard n + 1 from the boxes. This is a solution
for n ards and therefore of rst type, unless removing ard n + 1 left one box empty. In the rst ase, one
easily shows that ard n + 1 must t into the orre t arithmeti progression modulo 3, yielding the rst type
of solution. In the se ond ase, one similarly removes ard 1 from the boxes whi h shows that its box only
ontained this one ard, and one gets the se ond solution.

Problem 5. Determine whether or not there exists a positive integer n divisible by exa tly 2000 dierent
prime numbers, and 2n + 1 is divisible by n.

10
Solution: Yes, there exists su h a number. In fa t, for any positive integer k, there exists a positive integer
nk divisible by exa tly k dierent prime numbers su h that nk divides 2nk + 1.
Proof: The proof is by indu tion. The indu tion starts with k = 1, and with n1 = 3, sin e 3 divides
9 = 23 + 1.
Now assume that nk has been onstru ted with nk dividing 2nk + 1, and nk having exa tly k dierent
prime fa tors. To onstru t of nk+1 , one uses the following steps (for ease of notation, let N = nk ):

(a) If p is the largest power of p dividing N and p is the largest power of p dividing 2N + 1, then p+ is
2
the largest power of p dividing 2N + 1.
2
(b) N 2 divides 2N + 1.
2
( ) There is a prime q not dividing N but dividing 2N + 1.
(d) Let nk+1 = N 2 q, then nk+1 satises the above onditions.
Proof of (a): Let p be any prime dividing N with p and p as dened in (a). One an therefore write
2N = A p 1, for some integer A not divisible by p. One now uses the binomial theorem to ompute
(2N )p = (A p 1)p = A0 p+2 + A p+1 1 = A1 p+1 1;

for some integer A1 = A0 p + A not divisible by p. Iterating this gives (2N )p = A p+ 1 ; for some


integer A not divisible by p. Similarly,

 
N 2 N p  N=p  + N=p  00  + +1 N
2 = ((2 ) ) = (A p 1) =A p + A  p+ 1:
p
2
Sin e A is not divisible by p and, by denition, N=p is not divisible by p, it follows that 2N +1 is divisible
by p+ , but by no higher power, as laimed.
2
Proof of (b): Note that, by assumption,   , so one has that p2  divides 2N + 1. Sin e this holds for
2
every p dividing N , it follows that N 2 divides 2N + 1.
Proof of ( ): Let M be the largest divisor of 2N +1 su h that M is divisible only by primes whi h divide N .
2
Similarly, let M 0 be the largest divisor of 2N + 1, su h that M 0 is divisible only by primes whi h divide N .
It follows from part (a) that M 0 = MN , sin e part (a) proves this for ea h separate prime power dividing N .
Sin e M and N both divide 2N + 1 it follows that both M and N are smaller or equal to 2N + 1. Therefore

M 0 = MN  (2N + 1)2  (2N +1 )2 = 22N +2 :

On the other hand, using N  3,
2
2N = (22N )N=2 = 22N (22N (N=2 1)
2 N ;
2 +3

2 2
sin e N=2 1  3=2 and thus 2N (N=2 1)  3. It follows that M 0 < 2N + 1, and thus 2N + 1 must be
divisible by a prime q whi h does not divide N .
2
Proof of (d): Let nk+1 = N 2 q. Then by (b) one has 2N  1 (mod N 2 ) so that
2nk+1  ( 1)q = 1 (mod N 2 ) ;

11
re alling that q must be odd sin e it divides a power of 2 plus 1. By the hoi e of q, one also has that
2
2N  1 (mod q) so that
2nk+1  ( 1)q = 1 (mod q) :
The result follows from the Chinese Remainder Theorem.

Remark 5.1. The statement of the problem an be interpreted to mean: \n is divisible by exa tly 2000
dierent prime numbers, ea h with multipli ity one," In other words, n is assumed to be squarefree. It is
on eivable that a more pre ise form of the question was avoided in order not to give away the idea of the
solution, whi h depends on having primes with high multipli ities.
The alternate question appears to be intra table, in the nature of resolving whether there are an innite
number of primes of the form 2p 1. See [4℄ for examples of these types of unsolved problems.

Problem 6. Let AH1 , BH2 , CH3 be the altitudes of an a ute{angled triangle ABC . The in ir le of the
triangle ABC tou hes the sides BC , CA, AB at T1 , T2 , T3 , respe tively. Let the lines `1 , `2 , `3 be the
re e tions of the lines H2 H3 , H3 H1 , H1 H2 in the lines T2 T3 , T3 T1 , T1 T2 , respe tively.
Prove that `1 , `2 , `3 determine a triangle whose verti es lie on the in ir le of the triangle ABC .
C

L1 H1
T1
L3
T2
H2

A H3 T3 B

L2

Solution: The solution requires some basi results from triangle geometry whi h are easy to prove, but
whi h will only be quoted here.
Theorem 6.1. Let the notation be as in the statement of Problem 6. Then
(a) The enter of the in ir le is the interse tion of the bise tors of angles A, B , C .
(b) The enter of the ir um ir le is the interse tion of the right bise tors of the sides AB , BC , CA.
( ) The triangle AH2 H3 is similar to the triangle ABC , i.e., orresponding angles are equal. Similar state-
ments hold for BH3 H1 , and CH1 H2 .
These results immediately allow some progress.
Lemma 6.1. Let the notation be as in the statement of Problem 6, then `1 , `2 , `3 determine a triangle
(possibly a point) similar to triangle ABC .

12
 C

T2
H2 I

A H3 T3 B

Proof: Let I be the enter of the in ir le. Sin e AH2 H3 is similar to ABC , and AI bise ts the angle
BAC , re e ting H2 H3 about AI makes it parallel to BC . Sin e T2 T3 is perpendi ular to AI , it follows that
re e ting H2 H3 about T2 T3 also makes it parallel to BC , sin e two re e tions along two perpendi ular axes
is equivalent to one rotation of 180 degrees. Thus, `1 is parallel to BC . Similarly, `2 is parallel to AC and
`3 is parallel to AB . Note also that `1 opposite angle A, `2 opposite B , and `3 opposite B . 2
In order to pro eed, one introdu es oordinates for points inside a triangle by their distan e from the
sides. Thus, a point P inside the triangle ABC will be denoted by (x; y; z ), where x; y; z are the distan es
from P to BC , AC , and AB , respe tively (sin e the triangle is assumed to be a ute, the question of signs
does not arise). Note that P is uniquely determined by any two of x; y; z .
Remark 6.1. This parametrization orresponds to the usual bary entri oordinates sin e it is easy to show
that, with h1 the length of AH1 , et ., one has
x y z x y z
P = A + B + C ; where + + = 1:
h1 h2 h3 h1 h2 h3
The advantage of this parametrization is that it hara terizes the in enter and ir um enter from knowledge
of the sides alone.
Lemma 6.2. Let the notation be as in the statement of Problem 6, and P a point in the triangle with
oordinates (x; y; z ).
(a) If (x; y; z ) = (r; r; r), then P is enter of the in ir le of the triangle and r the radius of the in ir le.
(b) If (x; y; z ) = (R os A; R os B; R sin C ), where A = 6 BAC et ., then P is the enter of the ir um ir le
of the triangle and R the radius of the ir um ir le.
C

A B

13
Proof: Part (a) is trivial. To prove part (b), let O be the enter of the ir um ir le. Sin e triangle ABC is
a ute, O lies inside the triangle, and R = jOB j is the radius of the ir um ir le. Let  be 6 OBC = 6 OBC , so
that x = R sin . Furthermore, let  = 6 OCA = 6 OAC and = 6 OAB = 6 OBA. Sin e 2 (+ + ) = 180o
it follows that  = 90o ( + ). But also, one has  + = 6 A so that sin  = os A. The result follows. 2
The nal result follows dire tly from
Lemma 6.3. Let the notation be as in the statement of Problem 6, and let I 0 be the re e tion of I , the
enter of the in ir le, through the line T2 T3 . Then I 0 is the enter of the in ir le of AH2 H3 .
Indeed, assume that Lemma 6.3 holds. Then the distan e between I 0 and the line H2 H3 is the radius of
the in ir le of triangle AH2 H3 . By Theorem 6.1 ( ), this triangle is similar to triangle ABC , so its inradius
is the inradius of ABC multiplied by the fa tor of similarity. By Theorem 6.1 ( ), the fa tor of similarity
is jAH3 j=jAC j = os A, sin e CH3 is an altitude. It follows that the distan e from I 0 to the line H2 H3 is
r os A.
One now notes that, the distan e from I 0 , the re e tion of I through T2 T3, to the line H2 H3 is exa tly
the same as the distan e between I and the line `1, the re e tion of H2 H3 through T2 T3. One on ludes
that the distan e from I to `1 is r os A. Similarly, the distan e from I to `2 is r os B and the distan e
from I to `3 is r os C .
Lemma 6.1 and Lemma 6.2 (b) show that I is the enter of the ir um ir le of the triangle with sides
`1 ; `2 ; `3, and that this ir le has radius r. The result is therefore proved, given that Lemma 6.3 holds.
Proof of Lemma 6.3: In order to prove this, one uses the fa t just proved above that the triangle
AH2 H3 has ratio of similarity os A with respe t to triangle ABC , so has inradius r os A, and then apply
Lemma 6.2 (A).
C

T2

I’

A W T3 B

One therefore needs to ompute the distan e from I 0 to AB . Thus, let W be the point on AB su h that
AW I 0 is a right angle. The axis of re e tion T2T3 uts the line AI perpendi ularly, so that r = jT3 I j = jT3 I 0 j.
It follows that jI 0 W j = r os 6 T3 I 0 W .
Let Æ = 6 BAI . Sin e AW I 0 is a right angle, 6 AI 0 W and 6 AIT3 are both equal to 90o Æ. Sin e
jT3 I j = jT3I 0 j, one also has 6 T3 I 0 I = 90o Æ. It follows that 6 T3 I 0 W = 2Æ and thus jI 0 W j = r os(2Æ).
By Theorem 6.1 (a), the line AI bise ts 6 A, so it follows that jI 0 W j = r os A. Moreover, the fa t that
this line bise ts 6 BAC implies that the distan e from I 0 to AC is also r os A. Sin e two distan es to the
sides determine a point uniquely and the inradius is r os A, it follows that I 0 is the enter of the in ir le,
as laimed. 2

14
Remark 6.2. The above proof relies on two results: Lemma 6.1 and Lemma 6.3. The rst is \soft" in the
sense that it follows from known results without any omputation. The se ond is the heart of the te hni al
argument and appears to be of independent interest.
Remark 6.3. The above argument was dis overed only after a using a brute for e algebrai method to
prove the result. The following is a brief outline of this method.
Without loss of generality, one an assume that A = (0; 0), B = (1; 0), and C = (m; n), where 0 < m < 1
p p
and n > 0. Furthermore, let a = (1 m)2 + n2 be the length of the side opposite A, and b = m2 + n2
the length of the side opposite B , and p = 1 + b + the perimeter of the triangle.
On e again, let the enter of the in ir le be I , then it is easily omputed that I has oordinates ((m +
b)=p; n=p), and so r = n=p. If V is on e again the interse tion of T2 T3 with AI, then V = m2+b b I .
The re e tion of a point U about T2 T3 is given by U 7! U + 2 1 kUV
Vk2 V . Thus, if H20 and H30 are
the images of H2 and H3 under a re e tion through T2 T3, then H20 = bm B + 2V and H30 = bm C + 2V .
A omputation shows that line `1 has equation
n n
y= x+ (ma b(b + 1)) :
m 1 bp(m 1)
One an now derive the equation of `2 by using the symmetry x 7! 1 x whi h takes a 7! b and m 7! 1 m.
It follows that `2 has equation
n n n
y= x ((1 m)b a(a + 1)) :
m m apm
Fairly straightforward algebrai omputation eventually yields the oordinates (x0 ; y0) of the interse tion of
`1 and `2 , whi h are
b2 (m 1)2 a2 m2 b + m n (b2 (m 1) a2 m) n
x0 = + ; y0 = + :
abp p abp p
Sin e I = ((b + m)=p; n=p), it is then easy to he k that (x0 ; y0 ) is at distan e r = n=p from I . Sin e this
holds for any two verti es, the result is proved.

Referen es
[1℄ J.{P. Boudine, F. Lo Ja omo, R. Cu uliere, Olympiades Internationales de Mathematiques,

Enon  
es et solutions detailles, Annees 1988 a 1997, Edition du Choix, Marseille, 1998.
[2℄ H.S.M. Coxeter and S.L. Greitzer, Geometry Revisited, Mathemati al Asso iation of Ameri a,
1967.
[3℄ S.L. Greitzer, editor, International Mathemati al Olympiads, 1959{1977, Mathemati al Asso i-
ation of Ameri a, 1979.
[4℄ R.K. Guy, \Unsolved problems in number theory, 2nd Edition." Unsolved Problems in Intuitive
Mathemati s, I. Springer{Verlag, New York, 1994.
[5℄ M.S. Klamkin, editor, International Mathemati al Olympiads. 1978{1985, and Forty Supple-
mentary problems, Mathemati al Asso iation of Ameri a, 1999.

15
 [6℄ International Mathemati s Olympiad, web site: olympiads.win.tue.nl/imo/
[7℄ J. S holes, 41st IMO 2000, www.kalva.demon. o.uk/imo/imo00.html.
[8℄ A. Shen, Entran e Examinations to the Mekh{mat, Mathemati al Intelligen er 16 (1994), 6{10.
[9℄ I. Vardi, Mekh{mat entran e examination problems, IHES preprint IHES/M/00/06, January
2000.
[10℄ D. Zeilberger, Identities in sear h of identity, J. Theoreti al Computer S ien e 117 (1993), 23{38.

16