Zumning Feng, Yi Sun - USA and International Mathematical Olympiads 2007-2008 - 76p PDF

USA and International Mathematical
Olympiads
2007-2008
Edited by
Zuming Feng Yi Sun

Contents
Acknowledgments 3
1 USAMO 2007 4
1.1 USAMO 2007 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 USAMO 2007 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Team Selection Test 2007 13

2.1 TST 2007 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 TST 2007 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 USAMO 2008 29
3.1 USAMO 2008 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 USAMO 2008 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4 Team Selection Test 2008 40

4.1 TST 2008 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 TST 2008 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5 IMO 2006 47
5.1 IMO 2006 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.2 IMO 2006 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
6 IMO 2007 57
6.1 IMO 2007 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.2 IMO 2007 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7 Appendix 74
7.1 2006 Olympiad Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.2 2007 Olympiad Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.3 2003-2007 Cumulative IMO Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.4 2008 USAMO and USA IMO Team Selection Results . . . . . . . . . . . . . . . . . . . . . . 76
2
Acknowledgments
Thanks to Adam Hesterberg, Brian Lawrence, Ian Le, Po-Shen Loh, and Alison Miller for proofreading
this book. Thanks to Cecil Rousseau for the USAMO files.
Problems for the 2008 USAMO were chosen by the USAMO Committee [Steve Blasberg, Steven Dunbar,
Zuming Feng, Gregory Galperin, Elgin Johnston, Kiran Kedlaya, Cecil Rousseau (chair), Richard Stong,
Zoran Šuniḱ, and David Wells]. Proposals were made by members of the committee and other highly
experienced individuals [Titu Andreescu, Gabriel Carroll, Gabriel Dospinescu, Răzvan Gelca, Gerald Heuer,
Ian Le, Thomas Mildorf, and Sam Vandervelde]. Many thanks to all of these persons, especially those who
contributed additional solutions that were helpful in grading the USAMO and construction of this booklet.
The 2008 Team Selection Test (TST) was prepared by Zuming Feng, Răzvan Gelca, and Ian Le.
The Mathematics Olympiad Summer Program (MOSP) was held at the University of Nebraska-Lincoln.
Because of a generous gift from the Akamai Foundation, the 2008 MOSP was able to invite 56 students.
Among these students, 12 were female students preparing for the (Chinese) Girls’ Mathematics Olympiad.
Gabriel Carroll, Zuming Feng (Academic Director), Răzvan Gelca, Aliekber Gurel, Chris Jeuell, Ian Le,
Po-Ru Loh, Po-Shen Loh, Andy Niedermaier, and Josh Nichols-Barrer served as instructors. Alison Miller
and Yi Sun were junior instructors/leading graders. Zarathustra (Zeb) Brady, Adam Hesterberg, Brian
Lawrence, Keler Marku, and Maria Monks were graders.
3
1 USAMO 2007
36th United States of America Mathematical Olympiad
April 24 and 25, 2007
1.1 USAMO 2007 Problems

1. Let n be a positive integer. Define a sequence by setting a1 = n and, for each k > 1, letting ak be
the unique integer in the range 0 ≤ ak ≤ k − 1 for which a1 + a2 + · · · + ak is divisible by k. For
instance, when n = 9 the obtained sequence is 9, 1, 2, 0, 3, 3, 3, . . . . Prove that for any n the sequence
a1 , a2 , a3 , . . . eventually becomes constant.
(This problem was suggested by Sam Vandervelde.)
2. A square grid on the Euclidean plane consists of all points (m, n), where m and n are integers. Is it
possible to cover all grid points by an infinite family of discs with non-overlapping interiors if each
disc in the family has radius at least 5?
(This problem was suggested by Gregory Galperin.)
3. Let S be a set containing n2 +n−1 elements, for some positive integer n. Suppose that the n-element
subsets of S are partitioned into two classes. Prove that there are at least n pairwise disjoint sets in
the same class.
(This problem was suggested by Andás Gyárfás.)
4. An animal with n cells is a connected figure consisting of n equal-sized square cells.1 The figure
below shows an 8-cell animal.
A dinosaur is an animal with at least 2007 cells. It is said to be primitive if its cells cannot be
partitioned into two or more dinosaurs. Find with proof the maximum number of cells in a primitive
dinosaur.
(This problem was suggested by Reid Barton.)
1
Animals are also called polyominoes. They can be defined inductively. Two cells are adjacent if they share a complete
edge. A single cell is an animal, and given an animal with n-cells, one with n + 1 cells is obtained by adjoining a new cell by
making it adjacent to one or more existing cells.
4
n
5. Prove that for every nonnegative integer n, the number 77 + 1 is the product of at least 2n + 3 (not
necessarily distinct) primes.
(This problem was suggested by Titu Andreescu.)
6. Let ABC be an acute triangle with ω, Ω, and R being its incircle, circumcircle, and circumradius,
respectively. Circle ωA is tangent internally to Ω at A and tangent externally to ω. Circle ΩA is
tangent internally to Ω at A and tangent internally to ω. Let PA and QA denote the centers of ωA
and ΩA , respectively. Define points PB , QB , PC , QC analogously. Prove that
8PA QA · PB QB · PC QC ≤ R3 ,
with equality if and only if triangle ABC is equilateral.
(This problem was suggested by Kiran Kedlaya and Sungyoon Kim.)
5
1.2 USAMO 2007 Solutions
1. Let n be a positive integer. Define a sequence by setting a1 = n and, for each k > 1, letting ak be
the unique integer in the range 0 ≤ ak ≤ k − 1 for which a1 + a2 + · · · + ak is divisible by k. For
instance, when n = 9 the obtained sequence is 9, 1, 2, 0, 3, 3, 3, . . . . Prove that for any n the sequence
a1 , a2 , a3 , . . . eventually becomes constant.
Solution 1. For k ≥ 1, let

sk = a1 + a2 + · · · + ak .
We have
sk+1 sk+1 sk + ak+1 sk + k sk
< = ≤ = + 1.
k+1 k k k k
On the other hand, for each k, sk /k is a positive integer. Therefore
sk+1 sk
≤ ,
k+1 k
and the sequence of quotients sk /k is eventually constant. If sk+1 /(k + 1) = sk /k, then
(k + 1)sk sk
ak+1 = sk+1 − sk = − sk = ,
k k
showing that the sequence ak is eventually constant as well.

sk
sk = a1 + a2 + · · · + ak and = qk .
k
Since ak ≤ k − 1, for k ≥ 2, we have
k(k − 1)
sk = a1 + a2 + a3 + · · · + ak ≤ n + 1 + 2 + · · · + (k − 1) = n + .
2
m(m+1)
Let m be a positive integer such that n ≤ 2 (such an integer clearly exists). Then
sm n m−1 m+1 m−1

qm = ≤ + ≤ + = m.
m m 2 2 2
We claim that
qm = am+1 = am+2 = am+3 = am+4 = . . . .
This follows from the fact that the sequence a1 , a2 , a3 , . . . is uniquely determined and choosing am+i =
qm , for i ≥ 1, satisfies the range condition
0 ≤ am+i = qm ≤ m ≤ m + i − 1,
and yields
sm+i = sm + iqm = mqm + iqm = (m + i)qm .

sk = a1 + a2 + · · · + ak .
We claim that for some m we have sm = m(m − 1). To this end, consider the sequence which
computes the differences between sk and k(k − 1), i.e., whose k-th term is sk − k(k − 1). Note that
6
the first term of this sequence is positive (it is equal to n) and that its terms are strictly decreasing
since
(sk − k(k − 1)) − (sk+1 − (k + 1)k) = 2k − ak+1 ≥ 2k − k = k ≥ 1.
Further, a negative term cannot immediately follow a positive term. Suppose otherwise, namely that
sk > k(k − 1) and sk+1 < (k + 1)k. Since sk and sk+1 are divisible by k and k + 1, respectively, we
can tighten the above inequalities to sk ≥ k 2 and sk+1 ≤ (k + 1)(k − 1) = k 2 − 1. But this would
imply that sk > sk+1 , a contradiction. We conclude that the sequence of differences must eventually
include a term equal to zero.
Let m be a positive integer such that sm = m(m − 1). We claim that
m − 1 = am+1 = am+2 = am+3 = am+4 = . . . .
This follows from the fact that the sequence a1 , a2 , a3 , . . . is uniquely determined and choosing am+i =
m − 1, for i ≥ 1, satisfies the range condition
0 ≤ am+i = m − 1 ≤ m + i − 1,
and yields
sm+i = sm + i(m − 1) = m(m − 1) + i(m − 1) = (m + i)(m − 1).
2. A square grid on the Euclidean plane consists of all points (m, n), where m and n are integers. Is it
possible to cover all grid points by an infinite family of discs with non-overlapping interiors if each
disc in the family has radius at least 5?
Solution. It is not possible. The proof is by contradiction. Suppose that such a covering family F
exists. Let D(P, ρ) denote the disc with center P and radius ρ. Start with an arbitrary disc D(O, r)
that does not overlap any member of F. Then D(O, r) covers no grid point. Take the disc D(O, r)
to be maximal in the sense that any further enlargement would cause it to violate the non-overlap
condition. Then D(O, r) is tangent to at least three discs in F. Observe that there must be two of
the three tangent discs, say D(A, a) and D(B, b), such that ∠AOB ≤ 120◦ . By the Law of Cosines
applied to triangle ABO,
(a + b)2 ≤ (a + r)2 + (b + r)2 + (a + r)(b + r),
which yields
ab ≤ 3(a + b)r + 3r2 , and thus 12r2 ≥ (a − 3r)(b − 3r).
√ 2
Note that r < 1/ 2 because D(O, r) covers
√ no grid point, and (a−3r)(b−3r) √≥ (5−3r) because
√ √ each
disc in√
F has radius
√ at least 5. Hence 2 3r ≥ (5−3r), which gives 5 ≤ (3+2 3)r
√ < (3+2 3)/ 2 and
thus 5 2 < 3 + 2 3. Squaring both sides of this inequality yields 50 < 21 + 12 3 < 21 + 12 · 2 = 45.
This contradiction completes the proof.
Remark: The above √ argument

√ shows that no covering family exists where each disc has radius
greater than (3 + 2 3)/
√ 2 ≈ 4.571. In the other direction, there exists a covering family in which
each disc has radius 13/2 ≈ 1.802. Take discs with this radius centered at points√of the form
(2m + 4n + 12 , 3m + 12 ), where m and n are integers. Then any grid point is within 13/2 of one
√
of the centers and the distance between any two centers is at least 13. The extremal radius of a
covering family is unknown.
7
3. Let S be a set containing n2 +n−1 elements, for some positive integer n. Suppose that the n-element
subsets of S are partitioned into two classes. Prove that there are at least n pairwise disjoint sets in
the same class.
Solution. In order to apply induction, we generalize the result to be proved so that it reads as
follows:
Proposition. If the n-element subsets of a set S with (n + 1)m − 1 elements are partitioned into
two classes, then there are at least m pairwise disjoint sets in the same class.
Proof. Fix n and proceed by induction on m. The case of m = 1 is trivial. Assume m > 1 and
that the proposition is true for m − 1. Let P be the partition of the n-element subsets into two
classes. If all the n-element subsets belong to the same class, the result is obvious. Otherwise select
two n-element subsets A and B from different classes so that their intersection has maximal size. It
is easy to see that |A ∩ B| = n − 1. (If |A ∩ B| = k < n − 1, then build C from B by replacing
some element not in A ∩ B with an element of A not already in B. Then |A ∩ C| = k + 1 and
|B ∩ C| = n − 1 and either A and C or B and C are in different classes.) Removing A ∪ B from S,
there are (n+1)(m−1)−1 elements left. On this set the partition induced by P has, by the inductive
hypothesis, m − 1 pairwise disjoint sets in the same class. Adding either A or B as appropriate gives
m pairwise disjoint sets in the same class.
Remark: The value n2 + n − 1 is sharp. A set S with n2 + n − 2 elements can be split into a set
A with n2 − 1 elements and a set B of n − 1 elements. Let one class consist of all n-element subsets
of A and the other consist of all n-element subsets that intersect B. Then neither class contains n
pairwise disjoint sets.
4. An animal with n cells is a connected figure consisting of n equal-sized square cells.2 The figure
below shows an 8-cell animal.
A dinosaur is an animal with at least 2007 cells. It is said to be primitive if its cells cannot be
partitioned into two or more dinosaurs. Find with proof the maximum number of cells in a primitive
dinosaur.
Solution 1. Let s denote the minimum number of cells in a dinosaur; the number this year is
s = 2007.
Claim: The maximum number of cells in a primitive dinosaur is 4(s − 1) + 1 = 8025.
2
Animals are also called polyominoes. They can be defined inductively. Two cells are adjacent if they share a complete
edge. A single cell is an animal, and given an animal with n-cells, one with n + 1 cells is obtained by adjoining a new cell by
making it adjacent to one or more existing cells.
8
First, a primitive dinosaur can contain up to 4(s − 1) + 1 cells. To see this, consider a dinosaur in
the form of a cross consisting of a central cell and four arms with s − 1 cells apiece. No connected
figure with at least s cells can be removed without disconnecting the dinosaur.
The proof that no dinosaur with at least 4(s − 1) + 2 cells is primitive relies on the following result.
Lemma. Let D be a dinosaur having at least 4(s − 1) + 2 cells, and let R (red) and B (black) be
two complementary animals in D, i.e., R ∩ B = ∅ and R ∪ B = D. Suppose |R| ≤ s − 1. Then R
can be augmented to produce animals R̃ ⊃ R and B̃ = D \ R̃ such that at least one of the following
holds:
(i) |R̃| ≥ s and |B̃| ≥ s,

(ii) |R̃| = |R| + 1,
(iii) |R| < |R̃| ≤ s − 1.
Proof. If there is a black cell adjacent to R that can be made red without disconnecting B, then (ii)
holds. Otherwise, there is a black cell c adjacent to R whose removal disconnects B. Of the squares
adjacent to c, at least one is red, and at least one is black, otherwise B would be disconnected. Then
there are at most three resulting components C1 , C2 , C3 of B after the removal of c. Without loss of
generality, C3 is the largest of the remaining components. (Note that C1 or C2 may be empty.) Now
C3 has at least d(3s − 2)/3e = s cells. Let B̃ = C3 . Then |R̃| = |R| + |C1 | + |C2 | + 1. If |B̃| ≤ 3s − 2,
then |R̃| ≥ s and (i) holds. If |B̃| ≥ 3s − 1 then either (ii) or (iii) holds, depending on whether |R̃| ≥ s
or not.
Starting with |R| = 1, repeatedly apply the Lemma. Because in alternatives (ii) and (iii) |R| increases
but remains less than s, alternative (i) eventually must occur. This shows that no dinosaur with at
least 4(s − 1) + 2 cells is primitive.
Solution 2. (Based on Andrew Geng’s solution) Let s = 2007. We claim that the answer is 4s − 3 =
8025.
Consider a graph with the cells as the vertices and whose edges connect adjacent cells. Let T be
a spanning tree in this graph. By removing any vertex of T , we obtain at most four connected
components, which we call the limbs of the vertex. Limbs with at least s vertices are called big.
Suppose that every vertex of T contains a big limb, then consider a walk on T starting from an
arbitrary vertex and always moving along the edge towards a big limb. Since T is a finite tree, this
walk must traverse back on some edge at some point. Then the two connected components of T made
by deleting this edge are both big, so they both contain at least s vertices, which means that the
dinosaur is not primitive. It follows that a primitive dinosaur contains some vertex with no big limbs.
By removing this vertex, we get at most four connected components with at most s − 1 vertices each.
This not only shows that a primitive dinosaur has at most 4s − 3 cells, but also shows that any such
dinosaur consists of four limbs of s − 1 cells each connected to a central cell. It is easy to see that
such a dinosaur indeed exists.
n
5. Prove that for every nonnegative integer n, the number 77 + 1 is the product of at least 2n + 3 (not
necessarily distinct) primes.
9
0
Solution. The proof is by induction. The base is provided by the n = 0 case, where 77 + 1 =
71 + 1 = 23 . To prove the inductive step, it suffices to show that if x = 72m−1 for some positive
integer m then (x7 + 1)/(x + 1) is composite. As a consequence, x7 + 1 has at least two more prime
factors than does x + 1. To confirm that (x7 + 1)/(x + 1) is composite, observe that
x7 + 1 (x + 1)7 − ((x + 1)7 − (x7 + 1))

=
x+1 x+1
7x(x5 + 3x4 + 5x3 + 5x2 + 3x + 1)
= (x + 1)6 −
x+1
= (x + 1)6 − 7x(x4 + 2x3 + 3x2 + 2x + 1)
= (x + 1)6 − 72m (x2 + x + 1)2
= {(x + 1)3 − 7m (x2 + x + 1)}{(x + 1)3 + 7m (x2 + x + 1)}

√
Also each factor exceeds 1. It suffices to check the smaller one; 7x ≤ x gives
√
(x + 1)3 − 7m (x2 + x + 1) = (x + 1)3 − 7x(x2 + x + 1)
≥ x3 + 3x2 + 3x + 1 − x(x2 + x + 1)
= 2x2 + 2x + 1 ≥ 113 > 1.
Hence (x7 + 1)/(x + 1) is composite and the proof is complete.
6. Let ABC be an acute triangle with ω, Ω, and R being its incircle, circumcircle, and circumradius,
respectively. Circle ωA is tangent internally to Ω at A and tangent externally to ω. Circle ΩA is
tangent internally to Ω at A and tangent internally to ω. Let PA and QA denote the centers of ωA
and ΩA , respectively. Define points PB , QB , PC , QC analogously. Prove that
8PA QA · PB QB · PC QC ≤ R3 ,
Solution. (By Zuming Feng) Let the incircle touch the sides AB, BC, and CA at C1 , A1 , and B1 ,
respectively. Set AB = c, BC = a, CA = b. By equal tangents, we may assume that AB1 = AC1 = x,
BC1 = BA1 = y, and CA1 = CB1 = z. Then a = y + z, b = z + x, c = x + y. By the AM-GM
√ √ √
inequality, we have a ≥ 2 yz, b ≥ 2 zx, and c ≥ 2 xy. Multiplying the last three inequalities
yields
abc ≥ 8xyz, (†),
with equality if and only if x = y = z; that is, triangle ABC is equilateral.
Let k denote the area of triangle ABC. By the Extended Law of Sines, c = 2R sin ∠C. Hence
ab sin ∠C abc abc
k= = or R = . (‡)
2 4R 4k
We are going to show that
xa2
PA QA = . (∗)
4k
10
In exactly the same way, we can also establish its cyclic analogous forms
yb2 zc2
PB QB = and PC QC = .
4k 4k
Multiplying the last three equations together gives
xyza2 b2 c2
PA QA · PB QB · PC QC = .
64k 3
Further considering (†) and (‡), we have
8xyza2 b2 c2 a3 b3 c3
8PA QA · PB QB · PC QC = ≤ = R3 ,
64k 3 64k 3
0 denote the radii of ω, ω , Ω , respectively. We consider
Hence it suffices to show (∗). Let r, rA , rA A A
the inversion I with center A and radius x. Clearly, I(B1 ) = B1 , I(C1 ) = C1 , and I(ω) = ω. Let ray
AO intersect ωA and ΩA at S and T , respectively. It is not difficult to see that AT > AS, because ω
is tangent to ωA and ΩA externally and internally, respectively. Set S1 = I(S) and T1 = I(T ). Let `
denote the line tangent to Ω at A. Then the image of ωA (under the inversion) is the line (denoted
by `1 ) passing through S1 and parallel to `, and the image of ΩA is the line (denoted by `2 ) passing
through T1 and parallel to `. Furthermore, since ω is tangent to both ωA and ΩA , `1 and `2 are also
tangent to the image of ω, which is ω itself. Thus the distance between these two lines is 2r; that is,
S1 T1 = 2r. Hence we can consider the following configuration. (The darkened circle is ωA , and its
image is the darkened line `1 .)
PA
T1
S
C1
l
QA
B1
O
I
S1
l2
T A1
B C
HA
l1
11
By the definition of inversion, we have AS1 · AS = AT1 · AT = x2 . Note that AS = 2rA , AT = 2rA
0 ,
and S1 T1 = 2r. We have
x2 0 x2 x2
rA = . and rA = = .
2AS1 2AT1 2(AS1 − 2r)
Hence µ ¶
0 x2 1 1
PA QA = AQA − APA = rA − rA = + .
2 AS1 − 2r AS1
Let HA be the foot of the perpendicular from A to side BC. It is well known that ∠BAS1 = ∠BAO =
90◦ − ∠C = ∠CAHA . Since ray AI bisects ∠BAC, it follows that rays AS1 and AHA are symmetric
with respect to ray AI. Further note that both line `1 (passing through S1 ) and line BC (passing
through HA ) are tangent to ω. We conclude that AS1 = AHA . In light of this observation and using
the fact 2k = AHA · BC = (AB + BC + CA)r, we can compute PA QA as follows:
µ ¶ µ ¶
x2 1 1 x2 2k 2k
PA QA = − = −
2 AHA − 2r AHA 4k AHA − 2r AHA
Ã ! Ã !
x2 1 x2 1
= 1 2 − BC = 1 1 − (y + z)
4k BC − AB+BC+CA 4k y+z − x+y+z
µ ¶
x2 (y + z)(x + y + z)
= − (y + z)
4k x
x(y + z)2 xa2

= = ,
4k 4k
establishing (∗). Our proof is complete.
Remark: Trigonometric solutions of (∗) are also possible.

For a given triangle, how can one construct ωA and ΩA by ruler and compass?
12
2 Team Selection Test 2007
48th IMO Team Selection Test
Washington, D.C.
May 22 and 23, 2007
2.1 TST 2007 Problems

1. Circles ω1 and ω2 meet at P and Q. Segments AC and BD are chords of ω1 and ω2 respectively,
such that segment AB and ray CD meet at P . Ray BD and segment AC meet at X. Point Y lies
on ω1 such that P Y k BD. Point Z lies on ω2 such that P Z k AC. Prove that points Q, X, Y, Z are
collinear.
(This problem was suggested by Zuming Feng and Zhonghao Ye.)
2. Let n be a positive integer and let a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn be two nondecreasing

sequences of real numbers such that
a1 + · · · + ai ≤ b1 + · · · + bi for every i = 1, . . . , n − 1
and
a1 + · · · + an = b1 + · · · + bn .
Suppose that for any real number m, the number of pairs (i, j) with ai − aj = m equals the number
of pairs (k, `) with bk − b` = m. Prove that ai = bi for i = 1, . . . , n.
(This problem was suggested by Kiran Kedlaya.)
3. Let θ be an angle in the interval (0, π/2). Given that cos θ is irrational, and that cos kθ and cos[(k+1)θ]
are both rational for some positive integer k, show that θ = π/6.
(This problem was suggested by Zhigang Feng, Zuming Feng, and Weigu Li.)
4. Determine whether or not there exist positive integers a and b such that a does not divide bn − n for
all positive integers n.
(This problem was suggested by Thomas Mildorf.)
5. Triangle ABC is inscribed in circle ω. The tangent lines to ω at B and C meet at T . Point S lies on
ray BC such that AS ⊥ AT . Points B1 and C1 lies on ray ST (with C1 in between B1 and S) such
that B1 T = BT = C1 T . Prove that triangles ABC and AB1 C1 are similar to each other.
(This problem was suggested by Zuming Feng and Zhonghao Ye.)
13
6. For a polynomial P (x) with integer coefficients, r(2i − 1) (for i = 1, 2, 3, . . . , 512) is the remainder
obtained when P (2i − 1) is divided by 1024. The sequence
(r(1), r(3), . . . , r(1023))
is called the remainder sequence of P (x). A remainder sequence is called complete if it is a permutation
of (1, 3, 5, . . . , 1023). Prove that there are no more than 235 different complete remainder sequences.
(This problem was suggested by Danilo Gligoroski, Smile Markovski, and Zoran Šunić.)
14
2.2 TST 2007 Solutions
1. Circles ω1 and ω2 meet at P and Q. Segments AC and BD are chords of ω1 and ω2 respectively,
such that segment AB and ray CD meet at P . Ray BD and segment AC meet at X. Point Y lies
on ω1 such that P Y k BD. Point Z lies on ω2 such that P Z k AC. Prove that points Q, X, Y, Z are
collinear.
B
P
A
D
Y
X/X 1
Q
C
Z
Solution 1. We consider the above configuration. (Our proof can be modified for other configura-
tions.) Let segment AC meet the circumcircle of triangle CQD again (other than C) at X1 .
First, we show that Z, Q, X1 are collinear. Since CQDX1 is cyclic, ∠X1 CD = ∠DQX1 . Since
AC k P Z, ∠X1 CD = ∠ACP = ∠CP Z = ∠DP Z. Since P DQZ is cyclic, ∠DP Z + ∠DQZ = 180◦ .
Combining the last three equations, we obtain that
∠DQX1 + ∠DQZ = ∠X1 CD + ∠DQZ = ∠DP Z + ∠DQZ = 180◦ ;
that is, X1 , Q, Z are collinear.
B
P
A
D
Y
X/X 1
Q
C
Z
Second, we show that B, D, X1 are collinear; that is, X = X1 . Since AC k P Z, ∠CAP = ∠ZP B.
Since BP QZ is cyclic, ∠BP Z = ∠BQZ. It follows that ∠X1 AB = ∠CAP = ∠BQZ, implying that
15
ABQX1 is cyclic. Hence ∠X1 AQ = ∠X1 BQ. On the other hand, since BP DQ and AP QC are
cyclic,
∠QBD = ∠QP D = ∠QP C = ∠QAC = ∠QAX1 .
Combining the last two equations, we conclude that ∠X1 BQ = ∠X1 AQ = ∠QBD, implying that
X1 , D, B are collinear. Since X1 lies on segment AC, it follows that X = X1 . Therefore, we
established the fact that Z, Q, X are collinear.
B
P
A
D
Y
X
Q
C Z
To finish our proof, we show that Y, X, Q are collinear. Since ABQX is cyclic, ∠BAQ = ∠BXQ.
Since AP QY is cyclic, ∠BAQ = ∠P AQ = ∠P Y Q. Hence ∠P Y Q = ∠BAQ = ∠BXQ. Since
BX k P Y and ∠BXQ = ∠P Y Q, we must have Y, X, Q collinear.
P B
A
Y D
Q
C
Z
Solution 2. We claim that AXQB is cyclic. Because ACQP and P DQB are cyclic, we have
∠XAQ = ∠CAQ = ∠CP Q = ∠DP Q = ∠DBQ = ∠XBQ,
establishing the claim.
16
B
P
A
Y D
Q
C Z
Since AXQB and BP DQ are cyclic, we have
∠QSC = ∠ABQ = ∠P BQ = ∠CDQ,
implying that XDQC is cyclic.

Because XDQC is cyclic, ∠DQX = ∠DCX = ∠P CA. Since P Z k AC, ∠P CA = ∠CP Z = ∠DP Z.
Hence ∠DQX = ∠DP Z. Since P DQZ is cyclic, ∠DP Z + ∠DQZ = 180◦ . Combing the last two
equations yields ∠DQX + ∠DQZ = ∠DP Z + ∠DQZ = 180◦ ; that is, X, Q, Z are collinear.
Likewise, we can show that Y, X, Q are collinear.
2. Let n be a positive integer and let a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn be two nondecreasing

sequences of real numbers such that
a1 + · · · + ai ≤ b1 + · · · + bi for every i = 1, . . . , n − 1
and
a1 + · · · + an = b1 + · · · + bn .
Suppose that for any real number m, the number of pairs (i, j) with ai − aj = m equals the number
of pairs (k, `) with bk − b` = m. Prove that ai = bi for i = 1, . . . , n.
Remark: It is important to interpret the condition that for any real number m, the number of
pairs (i, j) with ai − aj = m equals the number of pairs (k, `) with bk − b` = m. It means that we
have two identical multi-sets (a multi-set is a set that allows repeated elements)
{ai − aj | 1 ≤ i < j ≤ n} and {bk − b` | 1 ≤ k < ` ≤ n}.
In particular, it gives us that

X X
(ai − aj ) = (bk − b` ), (∗)
1≤i<j≤n 1≤k<`≤n
X X
(ai − aj )2 = (bk − b` )2 (∗∗)
1≤i<j≤n 1≤k<`≤n
17
and
n
X n
X
|ai − aj | = |bi − bj |. (∗ ∗ ∗)
i,j=1 i,j=1
We present three solutions. The first solution is based on (∗), the second is based on (∗∗), and the
third is based on (∗ ∗ ∗).
Solution 1. Put sn = a1 + · · · + an = b1 + · · · + bn . Then

n−1
X
2 (a1 + · · · + ai ) = 2(n − 1)a1 + 2(n − 2)a2 + · · · + 2(1)an−1
i=1
= (n − 1)a1 + (n − 3)a2 + · · · + (1 − n)an + (n − 1)sn
X
= (n − 1)sn + (ai − aj )
1≤i<j≤n
and similarly
n−1
X X
2 (b1 + · · · + bi ) = (n − 1)sn + (bk − b` ).
i=1 1≤k<`≤n
By (∗), these two quantities are equal, so

n−1
X n−1
X
2 (a1 + · · · + ai ) = 2 (b1 + · · · + bi ).
i=1 i=1
Consequently, each of the inequalities a1 + · · · + ai ≤ b1 + · · · + bi for i = 1, . . . , n − 1 must be an

equality. Since we also have equality for i = n by assumption, we deduce that ai = bi for i = 1, . . . , n,
as desired.
Solution 2. Expanding both sides of (∗∗) yields

n
X X n
X X
(n − 1) a2i + 2 ai aj = (n − 1) b2i + 2 bk b` .
i=1 1≤i<j≤n i=1 1≤k<`≤n
Squaring both sides of the given equation a1 + · · · + an = b1 + · · · + an gives

n
X X n
X X
a2i + 2 ai aj = b2i + 2 bk b` .
i=1 1≤i<j≤n i=1 1≤k<`≤n
From the above relations we easily deduce that

n
X n
X
a2i = b2i .
i=1 i=1
By the Cauchy-Schwartz inequality, we obtain that

Ã n
!2 Ã n
!Ã n ! Ã n !2
X X X X
b2i = a2i b2i ≥ ai bi
i=1 i=1 i=1 i=1
18
or ¯ n ¯
n
X ¯X ¯ X n
¯ ¯
b2i ≥¯ ai bi ¯ ≥ ai bi . (†)
¯ ¯
i=1 i=1 i=1
We set si = a1 + · · · + ai and ti = b1 + · · · + bi for every 1 ≤ i ≤ n. By Abel’s summation formula,
we have
n
X
ai bi = s1 b1 + [s2 − s1 ]b2 + [s3 − s2 ]b3 + · · · + [sn − sn−1 ]bn
i=1
= s1 (b1 − b2 ) + s2 (b2 − b3 ) + · · · + sn−1 (bn−1 − bn ) + sn bn .
By the given conditions, si ≤ ti and bi − bi+1 ≤ 0 for every 1 ≤ i ≤ n − 1 and sn = tn . It follows that
n
X
ai bi ≥ t1 (b1 − b2 ) + t2 (b2 − b3 ) + · · · + tn−1 (bn−1 − bn ) + tn bn
i=1
n
X
= t1 b1 + [t2 − t1 ]b2 + [t3 − t2 ]b3 + · · · + [tn − tn−1 ]bn = b2i .
i=1
Combining the last inequality and (†), we conclude that the equality case holds for every inequality
we discussed above. In particular, si = ti for i = 1, . . . , n. These inequalities immediately give us
an = bn , an−1 = bn−1 , . . . , a1 = b1 and the problem is solved.
Solution 3. If u = (u1 , u2 , . . . , un ) and v = (v1 , v2 , . . . , vn ) are two nonincreasing sequences, we say
that u majorizes v if u1 + · · · + un = v1 + · · · + vn and u1 + · · · + ui ≥ v1 + · · · + vi for i = 1, 2, . . . , n − 1.
It is not difficult to see that (an , · · · , a1 ) majorizes (bn , . . . , b1 ). By a theorem of Birkhoff, it follows
that there are constants cσ ∈ (0, 1], where σ runs over some set S of permutations of {1, . . . , n}, with
P
σ∈S cσ = 1 and X
cσ aσ(i) = bi for i = 1, 2, . . . , n.
σ∈S
We will prove the inequality

n
X n
X
|ai − aj | ≥ |bi − bj |,
i,j=1 i,j=1
with equality if and only if ai = bi for i = 1, . . . , n. With this result, we complete our proof by noting
(∗ ∗ ∗).
We have
n
X X n
X X n
X
|ai − aj | = cσ |ai − aj | = cσ |aσ(i) − aσ(j) |
i,j=1 σ∈S i,j=1 σ∈S i,j=1
¯ ¯
Xn X n ¯X
X ¯ n
X
¯ ¯
= cσ |aσ(i) − aσ(j) | ≥ ¯ cσ (aσ(i) − aσ(j) )¯ = |bi − bj |,
¯ ¯
i,j=1 σ∈S i,j=1 σ∈S i,j=1
using the fact that |x1 | + · · · + |xm | ≥ |x1 + · · · + xm | for all real numbers x1 , . . . , xm .
This establishes the desired inequality; it remains to check the equality condition. For this, we must
have ¯ ¯
X ¯X ¯
¯ ¯
cσ |aσ(i) − aσ(j) | = ¯ cσ (aσ(i) − aσ(j) )¯
¯ ¯
σ∈S σ∈S
19
for each pair i and j; in particular, for each pair i and j, the sign of aσ(i) − aσ(j) must be the same
for all σ ∈ S for which aσ(i) 6= aσ(j) . It follows by the lemma below that the sequence aσ(1) , . . . , aσ(n)
itself must be the same for all σ ∈ S, yielding ai = bi .
Lemma. Let σ be a permutation of {1, . . . , n}, and let (a1 , . . . , an ) be an n-tuple of real numbers.
If
(aσ(1) , . . . , aσ(n) ) 6= (a1 , . . . , an ),
then there exist i, j ∈ {1, . . . , n} such that ai < aj but aσ(i) > aσ(j) .
Proof. We proceed by induction on n. There is no harm in assuming that a1 ≤ · · · ≤ an . Let m

be the least integer for which am = an . If {m, . . . , n} = {σ(m), . . . , σ(n)}, then σ also permutes
{1, . . . , m − 1} and we can reduce to that case. Otherwise, there is some i ≥ m such that σ(i) < m,
and there is some j < m such that σ(j) ≥ m. This pair i, j has the desired property.
Remark: The problem can be made slightly simpler by requiring a1 < · · · < an and b1 < · · · < bn , as
this avoids the lemma at the end of the third solution. This solution also reveals the relation between
this problem and Muirhead’s inequality. For more details of majorization, Muirhead’s inequality,
and Birkhoff’s theorem, one may visit the site en.wikipedia.org/wiki/Muirhead’s inequality.
Note also that it is an easy exercise with generating functions to construct counterexamples if the
majorization condition is dropped, even ignoring cases where the two sets differ by a translation plus
a reflection.
3. Let θ be an angle in the interval (0, π/2). Given that cos θ is irrational, and that cos kθ and cos[(k+1)θ]
are both rational for some positive integer k, show that θ = π/6.
Note: We present two solutions. Both solutions are based on the following facts.
Lemma 1. For every positive integer n, there is a monic polynomial (that is, a polynomial with
leading coefficient 1) Sn (x) with integer coefficients such that Sn (2 cos α) = 2 cos nα.
Proof. We induct on n. The base cases n = 1 and n = 2 are trivial by taking S1 (x) = x and
S2 (x) = x2 − 2. Assume the statement is true for n ≤ m. Note that by the addition-to-product
formulas, 2 cos[(m + 1)α] + 2 cos[(m − 1)α] = 4 cos mα cos α. Thus Sm+1 (x) = xSm (x) − Sm−1 (x)
satisfies the conditions of the problem, completing the induction.
Lemma 2. If cos α is rational and α = rπ for some rational number r, then the possible values of
cos α are 0, ±1, ± 12 .
Proof. Since r is rational, there exists positive integer n such that rn is an even integer. By lemma
1, Sn (2 cos α) = 2 cos(nα) = 2 cos(rnπ) = 1; that is, 2 cos α is a ration root of the monic polynomial
Sn (x) with integer coefficients. By Gauss’ lemma, 2 cos α must take integer values. Since −1 ≤
cos α ≤ 1, the possible values of 2 cos α are 0, ±1, ±2.
Solution 1. We note that if cos x is rational, then cos nx is rational for every positive integer n.
Indeed, this fact follows from a easy induction on n by noting the product-to-sum formula
2 cos nθ cos θ = cos[(n + 1)θ] + cos[(n − 1)θ].
20
Thus both cos(k 2 θ) = cos[k(kθ)] and cos[(k 2 −1)θ] = cos[(k−1)(k+1)θ] are rational. By the Addition
and subtraction formulas, we have
cos[(k + 1)θ] = cos kθ cos θ − sin kθ sin θ and cos(k 2 θ) = cos[(k 2 − 1)θ] cos θ − sin[(k 2 − 1)θ] sin θ.
Setting r1 = cos kθ, r2 = cos[(k + 1)θ], r3 = cos[(k 2 − 1)θ], r4 = cos(k 2 θ), , and x = cos θ in the above
equations yields
q q
r2 = r1 x ± (1 − r12 )(1 − x2 ) and r4 = r3 x ± (1 − r32 )(1 − x2 ),
or q q
± (1 − r12 )(1 − x2 ) = r2 − r1 x and ± (1 − r32 )(1 − x2 ) = r4 − r3 x.
Squaring these two equations and subtracting the resulting equations gives
2(r1 r2 − r3 r4 )x = r12 + r22 − (r32 + r42 ).
Since r1 , r2 , r3 , r4 are rational and x is irrational, we must have r1 r2 − r3 r4 = 0 or
cos kθ cos[(k + 1)θ] = cos(k 2 θ) cos[(k 2 − 1)θ].
By the product-to-sum formulas, we derive
cos[(2k + 1)θ] − cos θ cos[(2k 2 − 1)θ] − cos θ

=
2 2
or cos[(2k + 1)θ] − cos[(2k 2 − 1)θ] = 0. By the sum-to-product formulas, we obtain
2 sin[(k − k 2 + 1)θ] sin[(k 2 + k)θ] = 0,
implying that either (k − k 2 + 1)θ or (k 2 + k)θ is a integral multiple of π. Since k is an integer, we

conclude that θ = rπ for some rational number r.
Considering Lemma 2 for α = kθ and α = (k + 1)θ, the possible values of cos kθ and cos[(k + 1)θ]
are 0, ±1, ± 21 . Consequently, both kθ and (k + 1)θ is a integral multiple of π6 . Since 0 < θ =
kθ − (k − 1)θ < π2 , the only possible values of θ are π3 and π6 . Since cos θ is irrational, θ = π6 .
Solution 2. (Based on the work by Kiran Kedlaya) We maintain the notations used in the first
proof. Then s = 2 cos θ is a root of Sk (x) − 2r1 and Sk+1 (x) − 2r2 by the definition of Sn . Define
Q(x) = gcd(Sk (x) − 2r1 , Sk+1 (x) − 2r2 )
where the gcd is taken over the field of rational numbers. Then Q(x) is a polynomial with rational
coefficients, so the sum of its roots (with multiplicities) is rational. Since s is assumed not to be
rational, there must be at least one other distinct root t of Q(x).
Note that the k distinct reals 2 cos(θ + 2πa/k) for a = 0, 1, . . . , k − 1 form k roots of the degree k
polynomial Sk (x) − 2r1 , so they compose all of its roots. Similarly, all of the roots of Sk+1 (x) − 2r2
have the form 2 cos(θ + 2πb/(k + 1)) for b = 0, 1, . . . , k. Note that s and t are roots of Q(x). Therefore
roots of both Sk (x) − 2r1 and Sk (x) − 2r2 , and so they must have at least two distinct common roots.
Each root r of Q(x) must thus satisfy
r = 2 cos(θ + 2πa/k) = 2 cos(θ + 2πb/(k + 1))
21
for some a and b. We either have θ + 2πa/k = θ + 2πb/(k + 1) and thus r = 2 cos θ or θ + 2πa/k =
−θ − 2πb/(k + 1) and thus
π[(a + b)k + a]
θ=− .
k(k + 1)
In the first case, we obtain s, so t must lead to the second value of θ, as s 6= t.
πc
Therefore, we can write θ = k(k+1) for some integer c. By Lemma 2, c/k and c/(k + 1) must both be
cπ
multiples of 1/6, since cos kθ = cos k+1 and cos(k + 1)θ = cos cπ cπ cπ
k are rational. Therefore, θ = k − k+1
is a multiple of π/6. Since t is not rational, θ can only be π/6.
4. Determine whether or not there exist positive integers a and b such that a does not divide bn − n for
all positive integers n.
Note: The answer is no. We present two solutions, based on the following fact.
Lemma 1. Given positive integers a and b, for sufficiently large n we have that
bn+ϕ(a) ≡ bn (mod a).
(The function ϕ is the Euler’s totient function; for any positive integer m, we define to be ϕ(m) the
number of positive integers n less than m that are relatively prime to m.)
Proof. Let a = pα1 1 pα2 2 · · · pαk k , where p1 , . . . , pk are distinct primes. We know that ϕ is a multiplicative
functions, that is,
µ ¶ µ ¶
¡ ¢ ³ ´ ³ ´ 1 1
ϕ(a) = ϕ (pα1 1 ) ϕ (pα2 2 ) · · · ϕ pαk k = pα1 1 − pα1 1 −1 · · · pαk k − pα1 k −1 = a 1 − ··· 1 − .
p1 pk
In particular, ϕ(pαi i ) | ϕ(a) for each 1 ≤ i ≤ k and ϕ(a) < a.

For each pi , if pi divides b, then bn ≡ 0 (mod pαi i ) for n ≥ αi + 1. Hence bn+ϕ(a) ≡ bn bϕ(a) ≡ bn ≡ 0
(mod pαi i ) for n ≥ αi + 1; if pi does not divide b, then gcd(pαi i , b) = 1. By Euler’s theorem, we have
αi
bϕ(pi ) ≡ 1 (mod pαi i ). Since ϕ(pαi i ) | ϕ(a), we have bn+ϕ(a) ≡ bn (mod pαi i ). Therefore, for each pi ,
we have some ni such that for all n > ni , bn+ϕ(a) ≡ bn (mod pαi i ). Take N = max{ni } and note that
for all n > N , we have bn+ϕ(a) ≡ bn (mod pαi i ) for all i ≤ i ≤ k. Since pi are distinct, bn+ϕ(a) ≡ bn
(mod a), as desired.
Solution 1. For all pairs of positive integers a and b, we claim that there exist infinitely many n
such that a divides bn − n.
We establish our claim by strong induction on a. The base case of a = 1 holds trivially. Now, suppose
that the claim holds for all a < a0 . Since ϕ(a) < a, by the induction hypothesis and by Lemma 1,
there are infinitely many n such that
ϕ(a) | (bn − n) and bn+ϕ(a) ≡ bn (mod a).
For each such n, set

bn − n
t= and n1 = bn = n + tϕ(a).
ϕ(a)
22
It follows that
bn1 − n1 ≡ bn+tϕ(a) − (n + tϕ(a)) ≡ bn − n − tϕ(a) ≡ 0 (mod a).
Then, we see that n1 satisfies the desired property. By the induction hypothesis, there are infinitely
many n1 = bn satisfying the conditions of the claim for a, completing the induction.
Solution 2. We prove that no such a, b exist by proving the following: for any a, b, there is an
arithmetic progression n ≡ h (mod m), with m divisible only by primes less than or equal to the
greatest prime factor of a, such that bn ≡ n (mod a) for all sufficiently large n satisfying n ≡ h
(mod m).
Let us induct on highest prime divisor of a. The result is trivial for a = 1. Let p be a prime, and
suppose that the result is true whenever all the prime divisors of a are less than p. Now, suppose
that p is the greatest prime divisor of some a, and write a = pe a1 , where a1 has all prime factors
less than p. By the induction hypothesis, there is an arithmetic progression n ≡ h1 (mod m1 ),
with m1 divisible only by primes strictly less than p, such that for n ≡ h1 (mod m1 ) sufficiently
large, bn ≡ n (mod a1 ). There is no harm in assuming that p − 1 divides m1 . In this case, in this
arithmetic progression, bn is eventually constant modulo p due to the lemma. We can thus choose
a congruence modulo p so that for n an appropriate residue class modulo m1 p, bn ≡ n (mod p). In
this progression, bn is constant modulo p2 , so we can refine our choice of n modulo m1 p to a choice
of n modulo m1 p2 to force bn ≡ n (mod p2 ). We can then repeat the above process until we obtain
bn ≡ n (mod pe ). Since we originally had bn ≡ n (mod a1 ), combining the two congruences using the
Chinese Remainder Theorem gives us bn ≡ n (mod a) for all sufficiently large n in the congruence
class generated at the last step. This completes the induction.
Note: The key idea in both solutions is to reduce a, and the two solutions differ by how fast the
reduction takes place. While the second solution removes the prime divisors of a one by one starting
from the greatest, the first solution reduces a to φ(a).
These solutions remind us of problem 3 of USAMO 1991:
Show that, for any fixed integer n ≥ 1, the sequence

2 22
2, 22 , 22 , 22 , . . . (mod n)
is eventually constant. (The tower of exponents is defined by a1 = 2, ai+1 = 2ai .)
5. Triangle ABC is inscribed in circle ω. The tangent lines to ω at B and C meet at T . Point S lies on
ray BC such that AS ⊥ AT . Points B1 and C1 lies on ray ST (with C1 in between B1 and S) such
that B1 T = BT = C1 T . Prove that triangles ABC and AB1 C1 are similar to each other.
Solution 1. (Based on work by Oleg Golberg) We start with an important geometric observation.
23
A B
M T
Lemma. Triangle ABC inscribed in circle ω Lines BT and CT are tangent to ω. Let M be the
midpoint of side BC. Then ∠BAT = ∠CAM . (Line AT is a symmedian of triangle.)
Proof. We consider the above configuration. If ∠BAC is obtuse, our proof can be modified slightly.)
Let D denote the second intersection (other than A) of line AT and circle ω. Because BT is tangent
to ω at B, ∠T BD = ∠T AB. Hence triangles T BD and T AB are similar, implying that BD/AB =
T B/T A. Likewise, triangles T CD are T AC are similar and CD/AC = T C/T A. By equal tangents,
T B = T C. Consequently, we have BD/AB = T B/T A = T C/T A = CD/AC, implying that
BD · AC = CD · AB.
By the Ptolemy’s theorem to cyclic quadrilateral ABDC, we have BD · AC + AB · CD = AD · BC.

Combining the last two equations, we obtain that 2BD · AC = AD · BC or
AC BC MC
= = .
AD 2BD BD
Further considering that ∠ACM = ∠ACB = ∠ADB (since ABDC is cyclic), we conclude that
triangle ABD is similar to triangle AM C, implying that ∠BAT = ∠BAD = ∠CAM .
M
B C S
C1
T
B1
24
Because BT is tangent to ω, ∠CBT = ∠CAB, and so
∠T BA = ∠ABC + ∠CBT = ∠ABC + ∠CAB = 180◦ − ∠BCA.
By the lemma, we have ∠BAT = ∠CAM . Applying the Law of Sines to triangles BAT and CAM ,
we obtain
BT sin ∠BAT sin ∠CAM MC
= = = .
AT sin ∠T BA sin ∠BCA AM
Note that T B = T C1 . Thus, T C1 /T A = M C/M A. By equal tangents, T B = T C. In isosceles
triangle BT C, M is the midpoint of base BC. Consequently, ∠T M S = ∠T AC = ∠T AS = 90◦ ,
implying that T M AP is cyclic. Hence ∠AM C = ∠AT C1 . Because
AM MC
= (∗)
AT T C1
and ∠AM C = ∠AT C1 , triangles M AC and T AC1 are similar. Because BC/BM = B1 C1 /T C1 = 2,
triangles ABC and AB1 C1 are similar.
Solution 2. (By Alex Zhai) We maintain the notations in the first proof. As shown at the end of
the first proof, it suffices to show that (∗).
A
O
M
B C S
C1
T
B1
Let O be the circumcenter of ABC. Note that triangles OM C, OCT are similar to each other,
implying that OM/OC = OC/OT or OM · OT = OC 2 = OA2 . Thus triangles OAM and OT A are
also similar to each other. Further note that triangles OM C and CM T are also similar to each other.
These similarities (which amount to the circumcircle of ABC being a circle of Apollonius) give
AM OM OM MC MC MC
= = = = = ,
AT OA OC CT TB T C1
which is (∗).
25
Solution 3. (Based on work by Sherry Gong) We maintain the notations of the previous solutions.
Let ω intersect lines AT and AS again at X and Y (other than A), respectively. Let lines Y B and
CX meet at B2 , and let Y C and BX meet at C2 . Applying Pascal’s theorem to cyclic (degenerated)
hexagon BBY AXC shows that intersections of three pairs of lines BB and AX, BY and XC,
and Y A and CB are collinear; that is, B2 , C2 , S are collinear. Likewise, applying Pascal’s theorem
to cyclic (degenerated) hexagon CCY AXB shows that B2 , C2 , T are collinear. We conclude that
B2 , C2 , S, T are collinear.
M
S
B X C
C2
B2 T
Since ACXY is cyclic, ∠Y CX = ∠Y AX = 180◦ − ∠XAB = 90◦ . Thus ∠B2 CC2 = ∠XCC2 =
180◦ − ∠Y CX = 90◦ . Likewise, ∠C2 BB2 = 90◦ . It follows that BCC2 B2 is inscribed in a circle
with B2 C2 as its diameter. Thus the circumcenter of this circle is the intersection of lines ST and
the perpendicular of segment BC. This circumcenter must be T , and consequently, B2 = B1 and
C2 = C1 .
B X S
C
C1
B1 T
Because AY BC and B1 C1 CB are cyclic, by Miquel’s theorem, SACC1 . (Indeed, ∠ACB = 180◦ −
∠B1 Y S and ∠BCC1 = ∠180◦ − ∠Y B1 S lead to ∠ACC1 = 360◦ − ∠ACB − ∠BCC1 = 180◦ − Y SB1 .)
Also, by Miquel’s theorem, Y AC1 B1 is cyclic. (Indeed, ∠C1 AX = ∠C1 CS = ∠SB1 Y .) By these
26
cyclic quadrilaterals, it is not difficult to obtain ∠ACS = ∠AC1 S (or ∠ACB = ∠AC1 B1 ) and
∠ABC = ∠AY C = ∠AY C1 = ∠AB1 C1 . Consequently, triangles ABC and AB1 C1 are similar to
each other.
Remark: The last approach reveals the problem posers’ motivation. We can view the BCC1 B1 -
{S, Y } as a complete quadrilateral. Then A is is its Miquel’s point. This problem combines two
properties of complete quadrilateral and its Miquel’s points: (1) A lies on SY if and only if BCC1 B1
is cyclic; (2) the line through A perpendicular to SY passes through the circumcenter of B1 BC1 .
6. For a polynomial P (x) with integer coefficients, r(2i − 1) (for i = 1, 2, 3, . . . , 512) is the remainder
obtained when P (2i − 1) is divided by 1024. The sequence
(r(1), r(3), . . . , r(1023))
is called the remainder sequence of P (x). A remainder sequence is called complete if it is a permutation
of (1, 3, 5, . . . , 1023). Prove that there are no more than 235 different complete remainder sequences.
Solution. Define the polynomials
Q0 (x) = b0 ,
Q1 (x) = b1 (x + 1),
Q2 (x) = b2 (x + 1)(x + 3),
Q3 (x) = b3 (x + 1)(x + 3)(x + 5),
Q4 (x) = b4 (x + 1)(x + 3)(x + 5)(x + 7),
Q5 (x) = b5 (x + 1)(x + 3)(x + 5)(x + 7)(x + 9),
Q6 (x) = b6 (x + 1)(x + 3)(x + 5)(x + 7)(x + 9)(x + 11),
where
b0 = 210 , b1 = 29 , , b2 = 27 , b3 = 26 b4 = 23 , b5 = 22 , b6 = 20 .
The product of i consecutive even integers is divisible by 2i · i!. Therefore, for i = 0, 1, 2, 3, 4, 5, 6, we
obtain that the product of i consecutive even integers is divisible by 20 , 21 , 23 , 24 , 27 , 28 , 210 , respec-
tively. This implies that, for any odd integer x and i = 0, . . . , 6, Qi (x) is divisible by 210 .
A polynomial P (x) = a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + a5 x5 with integer coefficients is called reduced
if, for i = 0, . . . , 5,
0 ≤ ai < bi . (†)
Clearly, there are exactly b0 b1 · · · b5 = 210+9+7+6+3+2 = 237 distinct reduced polynomials.
We show that, for every polynomial P (x) with integer coefficients, there exists a reduced polynomial
P̄ (x) such that P (x) and P̄ (x) have the same remainder sequence.
First note that, for i = 0, . . . , 6, and any polynomial R(x) with integer coefficients P (x) and P (x) −
R(x)Qi (x) have the same remainder sequence. This follows from the fact that Qi (x) is divisible by
210 , for any odd integer x.
If the degree d of P (x) = a0 +· · ·+ad xd is higher than 5 we may replace P (x) by P (x)−ad xd−6 Q6 (x).
Indeed, the polynomial P (x)−ad xd−6 Q6 (x) has smaller degree than P (x) and has the same remainder
sequence as P (x). We may continue this until we obtain a polynomial that of degree at most 5 that
has the same remainder sequence as P (x).
27
We assume now that P (x) has degree no higher than 5. If P (x) is reduced we are done. Otherwise,
let i be the highest degree of a coefficient ai of xi that does not satisfy the range condition (†) If q
is the quotient obtained by dividing ai by bi then P (x) and P (x) − qQi (x) have the same remainder
sequence and the coefficient at degree i in P (x) − qQi (x) is in the correct range 0, . . . , bi − 1.
We repeat this procedure with the next highest degree that has a coefficient out of range until we
reach a reduced polynomial that has the same remainder sequence as P (x).
We now consider the 237 reduced polynomials.
Let a = 29 +1 and b = 1. Then P (a)−P (b) = (a−b)(a1 +a2 A2 +a3 A3 +a4 A4 +a5 A5 ), where A2 = a+b,
A3 = a2 + ab + b2 , A4 = a3 + a2 b + ab2 + b3 and A5 = a4 + a3 b + a2 b2 + ab3 + b4 . Since both a and b
are odd, A2 and A4 are even, A3 and Ad are odd, and the parity of a1 + a2 A2 + a3 A3 + a4 A4 + a5 A5
is the same as the parity of a1 + a3 + a5 . Therefore, if a1 + a3 + a5 is even P (a) − P (b) is divisible
by 210 and the sequence of remainders of P (x) is not a permutation.
For an odd integer x, the parity of P (x) = a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + a5 x5 is the same as the
parity of the sum a0 + a1 + · · · + a5 . Thus, only polynomials with odd sum of coefficients have odd
remainders.
Therefore, there are no more remainder sequences that are permutations of 1, 3, . . . , 1023 than there
are reduced polynomials P (x) = a0 + a1 x + a2 x2 + a3 x3 + a4 x4 + a5 x5 for which both a1 + a3 + a5
and a0 + a1 + a2 + a3 + a4 + a5 are odd.
There are exactly 236 reduced polynomials for which a1 + a3 + a5 is odd. This can be seen by pairing
up every reduced polynomial P (x) in which a1 is even with the polynomial P (x) + x. Exactly one of
the two polynomials in each such pair has odd sum a1 + a3 + a5 .
There are exactly 235 reduced polynomials for which both a1 + a3 + a5 and a0 + a1 + a2 + a3 + a4 + a5
are odd. This can be seen by pairing up every reduced polynomial P (x) in which a1 + a3 + a5 is odd
and a0 is even with the polynomial P (x) + 1. Exactly one of the two polynomials in each such pair
has odd sum a0 + a1 + a2 + a3 + a4 + a5 .
Remark: It can be proved that there are exactly 235 different remainder sequences that are
permutations of 1, 3, . . . , 1023.
28
3 USAMO 2008
37th United States of America Mathematical Olympiad
April 29 and 30, 2008
3.1 USAMO 2008 Problems

1. Prove that for each positive integer n, there are pairwise relatively prime integers k0 , k1 , . . . ,
kn , all strictly greater than 1, such that k0 k1 · · · kn − 1 is the product of two consecutive integers.
(This problem was suggested by Titu Andreescu.)
2. Let ABC be an acute, scalene triangle, and let M, N , and P be the midpoints of BC, CA, and
AB, respectively. Let the perpendicular bisectors of AB and AC intersect ray AM in points D and
E respectively, and let lines BD and CE intersect in point F , inside of triangle ABC. Prove that
points A, N, F , and P all lie on one circle.
(This problem was suggested by Zuming Feng.)
3. Let n be a positive integer. Denote by Sn the set of points (x, y) with integer coordinates such that
¯ ¯
¯ 1 ¯¯
¯
|x| + ¯y + ¯ < n.
2
A path is a sequence of distinct points (x1 , y1 ), (x2 , y2 ), . . . , (x` , y` ) in Sn such that, for i = 2, . . . , `,
the distance between (xi , yi ) and (xi−1 , yi−1 ) is 1 (in other words, the points (xi , yi ) and (xi−1 , yi−1 )
are neighbors in the lattice of points with integer coordinates).
Prove that the points in Sn cannot be partitioned into fewer than n paths (a partition of Sn into m
paths is a set P of m nonempty paths such that each point in Sn appears in exactly one of the m
paths in P).
(This problem was suggested by Gabriel Carroll.)
4. Let P be a convex polygon with n sides, n ≥ 3. Any set of n − 3 diagonals of P that do not intersect
in the interior of the polygon determine a triangulation of P into n − 2 triangles. If P is regular and
there is a triangulation of P consisting of only isosceles triangles, find all the possible values of n.
5. Three nonnegative real numbers r1 , r2 , r3 are written on a blackboard. These numbers have the
property that there exist integers a1 , a2 , a3 , not all zero, satisfying a1 r1 + a2 r2 + a3 r3 = 0. We are
permitted to perform the following operation: find two numbers x, y on the blackboard with x ≤ y,
then erase y and write y − x in its place. Prove that after a finite number of such operations, we can
end up with at least one 0 on the blackboard.
(This problem was suggested by Kiran Kedlaya.)
29
6. At a certain mathematical conference, every pair of mathematicians are either friends or strangers.
At mealtime, every participant eats in one of two large dining rooms. Each mathematician insists
upon eating in a room which contains an even number of his or her friends. Prove that the number
of ways that the mathematicians may be split between the two rooms is a power of two (i.e., is of
the form 2k for some positive integer k).
30
3.2 USAMO 2008 Solutions
1. Prove that for each positive integer n, there are pairwise relatively prime integers k0 , k1 , . . . , kn , all
strictly greater than 1, such that k0 k1 · · · kn − 1 is the product of two consecutive integers.
Solution 1. We proceed by induction. The case n = 1 is clear, since we may pick k0 = 3 and k1 = 7.
Let us assume now that for a certain n there are pairwise relatively prime integers 1 < k0 < k1 <
· · · < kn such that k0 k1 · · · kn − 1 = an (an − 1), for some positive integer an . Then choosing
kn+1 = a2n + an + 1 yields
k0 k1 · · · kn+1 = (a2n − an + 1)(a2n + an + 1) = a4n + a2n + 1,
so k0 k1 · · · kn+1 − 1 is the product of the two consecutive integers a2n and a2n + 1. Moreover,
gcd(k0 k1 · · · kn , kn+1 ) = gcd(a2n − an + 1, a2n + an + 1) = 1,
hence k0 , k1 , . . . , kn+1 are pairwise relatively prime. This completes the proof.
Solution 2. Write the relation to be proved as
4k0 k1 · · · kn = 4a(a + 1) + 4 = (2a + 1)2 + 3.
There are infinitely many primes for which −3 is a quadratic residue. Let 2 < p0 < p1 < . . . < pn be
such primes. Using the Chinese Remainder Theorem to specify a modulo pn , we can find an integer
a such that (2a + 1)2 + 3 = 4p0 p1 · · · pn m for some positive integer m. Grouping the factors of m
appropriately with the pi ’s, we obtain (2a + 1)2 + 3 = 4k0 k1 · · · kn with ki pairwise relatively prime.
We then have k0 k1 · · · kn − 1 = a(a + 1), as desired.
Solution 3. We are supposed to show that for every positive integer n, there is a positive integer x
such that x(x + 1) + 1 = x2 + x + 1 has at least n distinct prime divisors. We can actually prove a
more general statement.
Claim. Let P (x) = ad xd + · · · + a1 x + 1 be a polynomial of degree d ≥ 1 with integer coefficients.
Then for every positive integer n, there is a positive integer x such that P (x) has at least n distinct
prime divisors.
The proof follows from the following two lemmas.

Lemma 1. The following set is infinite:
Q = {p | p a prime for which there is an integer x such that p divides P (x)}.
Proof. The proof is analogous to Euclid’s proof that there are infinitely many primes. Namely, if
we assume that there are only finitely many primes p1 , p2 , . . . , pk in Q, then for each integer m,
P (mp1 p2 · · · pk ) is an integer with no prime factors, which must equal 1 or −1. However, since P has
degree d ≥ 1, it takes each of the values 1 and −1 at most d times, a contradiction.
Lemma 2. Let p1 , p2 , . . . , pn , n ≥ 1 be primes in Q. Then there is a positive integer x such that

P (x) is divisible by p1 p2 · · · pn .
Proof. For i = 1, 2, . . . , n, since pi ∈ Q we can find an integer ci such that P (x) is divisible by
pi whenever x ≡ ci (mod pi ). By the Chinese Remainder Theorem, the system of n congruences
x ≡ ci (mod pi ), i = 1, 2, . . . , n has positive integer solutions. For every positive integer x that solves
this system, P (x) is divisible by p1 p2 · · · pn .
31
2. Let ABC be an acute, scalene triangle, and let M, N , and P be the midpoints of sides BC, CA,
and AB, respectively. Let the perpendicular bisectors of segments AB and AC intersect ray AM in
points D and E respectively, and let lines BD and CE intersect in point F , inside of triangle ABC.
Prove that points A, N, F , and P all lie on one circle.
Solution 1. Assume without loss of generality that AB > AC, and we consider the configuration
shown below. Our proof can be modified for other configurations.
Let O be the circumcenter of triangle ABC. Point D lies on P O, which is the perpendicular bisector
of segment AB. Thus, ABO is an isosceles triangle with AD = BD. Likewise, AEC is isosceles with
AE = EC. Set x = ∠ABD = ∠BAD and y = ∠CAE = ∠ACE. (Hence x + y = ∠BAC.)
Applying the law of sines to triangles ABM and ACM gives
BM AB CM AC
= and = .
sin x sin ∠BM A sin y sin ∠CM A
Dividing the two equations yields
BM sin y AB sin ∠BM A AC
· = · = ,
CM sin x AC sin ∠CM A AB
since sin ∠BM A = sin ∠CM A (as they are supplementary angles). Therefore,
sin x AC
BM = M C if and only if = . (∗)
sin y AB
Applying the law of sines to triangles ABF and ACF gives
AF AB AF AC
= and = .
sin x sin ∠AF B sin y sin ∠AF C
Dividing these two equations and noting (∗) yields
sin y AB sin ∠AF B
= · or sin ∠AF B = sin ∠AF C.
sin x AC sin ∠AF C
Since ∠ADF is an exterior angle of triangle ABO, ∠EDF = 2x. Likewise, ∠DEF = 2y. Thus
∠EF D = 180◦ − 2x − 2y = 180◦ − 2∠BAC. Hence ∠BF C = 2∠BAC (and so BOF C is cyclic),
implying that ∠AF B + ∠AF C = 360◦ − 2∠BAC > 180◦ (since ∠BAC < 90◦ ). We must have
∠AF B = ∠AF C = 180◦ − ∠BAC. Since BOF C is cyclic and BOC is isosceles with vertex angle
BOC = 2∠BAC, ∠OF B = ∠OCB = 90◦ − ∠BAC. Therefore,
∠AF O = ∠AF B − ∠OF B = 180◦ − ∠BAC − (90◦ − ∠BAC) = 90◦ .
Since ∠AP O = ∠AF O = ∠AN O = 90◦ , points A, P, O, F, N lie on a circle.
A A
R
P
E N
P E N
O F
B D C O F
M
D
B M C
32
Solution 2. (Based on work by Richard Rusczyk) We maintain the notations in the first proof. Let
R denote the intersection of lines M P and CE. Let ω denote the circumcircle of AP ON . We claim
that R lies on ω. Note that AEC is isosceles, so that M ER is also isosceles, and so ARM C is an
isosceles trapezoid, hence cyclic. Since P N k M C, ARP N is also cyclic.
As in the above proof, we can show that BF OC is cyclic. Then ∠RF O = ∠OBC = 90◦ − ∠A.
Moreover, ∠RAO = ∠RAC − ∠OAC = ∠C − (90◦ − ∠B) = 90◦ − ∠A. Thus AROF is cyclic, so that
ARP OF N is cyclic.
Solution 3. (By Oleg Golberg) We maintain the notations in the first proof. Assume without loss
of generality that AB > AC, and we consider the configuration shown below. Our proof can be
modified for other configurations.
Since ∠OP B = ∠OM B = ∠OM C = ∠ON C = 90◦ , BP OM and CN OM are cyclic. Let ray AM
meet the circumcircles of BP OM and CN OM at X and Y (different from M ), respectively. We
have
∠OXY = ∠OXM = ∠OBM = ∠OBC = ∠BCO = ∠M CO = ∠M Y O = ∠XY O;
that is, triangle OXY is similar to triangle OBC. and the rotation centered at O that takes one
to the other. In particular, ∠XOY = ∠BAC. Note also that AP ON is cyclic, implying that
∠N OD = ∠P AN = ∠BAC. Hence, ∠XOY = 2∠N OD. Combining the facts that OX = OY
and ∠XOY = 2∠N OD, we conclude that the image of of X under the reflection through line OD
(denoted by ROP ) is the same as that of Y under the reflection through line ON (denoted by RON ).
Let F1 denote this common image. Note that the image of line AY E under RON is line CF1 E, and
the image of line ADX under ROP is line BDF1 . Therefore, F1 lies on lines CE and BD; that is,
F1 = F .
On the other hand, since X lies on the circumcircle of M BP O, F = F1 must lie on its image under
ROP , which is the circumcircle of N AP O.
A
P N IB
Y IP
E ID IZ IC
O F/F 1 P N
D E
B M C I F/I G
F/G IN
D
X B C
M
Solution 4. (By Gabriel Carroll) Invert the figure about a circle centered at A, and let IX denote
the image of the point X under this inversion. Find point G so that AIB IG IC is a parallelogram and
let IZ denote the center of this parallelogram. Note that triangle BAC is similar to triangle IC AIB
and triangles BAD and ID AIB . Because M is the midpoint of AB and IZ is the midpoint of IB IC ,
we also have the similarity between the triangles BAM and IC AIZ . Thus
∠AIG IB = ∠IG AIC = ∠ZAIC = ∠M AB = ∠DAB = ∠DBA = ∠AID IB .
33
Hence quadrilateral AIB ID IG is cyclic and, by a similar argument, quadrilateral AIC IE IG is also
cyclic. Because the images under the inversion of lines BDF and CF E are circles that intersect in
A and IF , it follows that IG = IF ; that is, G = F .
Next note that IB , IZ , and IC are collinear and are the images of IP , IF (or IG ), and IN , respectively,
under a homothety centered at A and with ratio 1/2. It follows that IP , IF and IN are collinear, and
then that the points A, P, F and N lie on a circle.
P
E N
O F
B D C
M
A
P N
E
D
F/F 2
B C
M T
Solution 5. (Based on work by Evan O’Dorney) Let A be the origin and denote the complex number
of each point by the corresponding lowercase letter. Note that 2n = c and 2p = b. Define point F2
such that
2np cp bn
f2 = = = ;
m m m
that is
f2 p f2 n
= and = . (∗∗)
c m b m
The first equation in (∗∗) implies that triangles AF2 C and AP M are similar (and have the same
orientation). Hence
∠ACF2 = ∠AM P = ∠M AC = ∠EAC = ∠ECA;
that is, F2 lies on line EC. Similarly, the second equation in (∗∗) shows that F2 lies on line BD.
Thus, F2 lies on both BD and CE; that is, F2 = F .
We have established the similarity between triangle AF C (which is AF2 C) and triangle AP M .
Thus, triangle ACM is similar to triangle AF P (because of spiral similarities about A). Hence
∠AF P = ∠ACM = ∠AN P , implying that AN F P is cyclic.
Remark: This problem combines the following facts: (1) F lies on the circumcircle of triangle BOC;
(2) F lies on a symmedian of triangle ABC (since ∠BAM = ∠CAF ; (3) line AT is a symmedian
of triangle ABC is T B ⊥ OB and T C ⊥ OC. Facts (1) and (2) have been established in the first
solution and fact (3) is well known and has been established as a lemma in the solution of TST 2007
34
problem 5. Hence F is the intersection of the symmedian passing through A and the circumcircle
of triangle BOC. Further note that OT is a diameter of the circumcircle of triangle BOC. Thus,
OF ⊥ AT . Indeed, line OF is the radical axis of the circumcircles of triangle AN P and BOC. The
problem is developed from IMO 1997 problem 2:
The angle at A is the smallest angle of triangle ABC. The points B and C divide the
circumcircle of the triangle into two arcs. Let U be an interior point of the arc between
B and C which does not contain A. The perpendicular bisectors of AB and AC meet
the line AU at V and W , respectively. The lines BV and CW meet at T. Show that
AU = T B + T C.
3. Let n be a positive integer. Denote by Sn the set of points (x, y) with integer coordinates such that
¯ ¯
¯ 1 ¯¯
¯
|x| + ¯y + ¯ < n.
2
A path is a sequence of distinct points (x1 , y1 ), (x2 , y2 ), . . . , (x` , y` ) in Sn such that, for i = 2, . . . , `,
the distance between (xi , yi ) and (xi−1 , yi−1 ) is 1 (in other words, the points (xi , yi ) and (xi−1 , yi−1 )
are neighbors in the lattice of points with integer coordinates).
Prove that the points in Sn cannot be partitioned into fewer than n paths (a partition of Sn into m
paths is a set P of m nonempty paths such that each point in Sn appears in exactly one of the m
paths in P).
Solution. Color the points in Sn as follows (see Figure 1):
- if y ≥ 0, color (x, y) white if x + y − n is even and black if x + y − n is odd;

- if y < 0, color (x, y) white if x + y − n is odd and black if x + y − n is even.
•
(0,2)
• ◦ •
(−1,1) (0,1) (1,1)
• ◦ •(0,0) ◦ •
(−2,0) (−1,0) (1,0) (2,0)
• ◦ • ◦ •
(−2,−1) (−1,−1) (0,−1) (1,−1) (2,−1)
• ◦ •
(−1,−2) (0,−2) (1,−2)
•(0,−3)
Figure 1: Coloring of S3
Consider a path (x1 , y1 ), (x2 , y2 ), . . . , (x` , y` ) in Sn . A pair of successive points (xi−1 , yi−1 ) and (xi , yi )
in the path is called a pair of successive black points if both points in the pair are colored black.
Suppose now that the points of Sn are partitioned into m paths and the total number of successive
pairs of black points in all paths is k. By breaking the paths at each pair of successive black points,
35
we obtain k + m paths in each of which the number of black points exceeds the number of white
points by at most one. Therefore, the total number of black points in Sn cannot exceed the number
of white points by more than k + m. On the other hand, the total number of black points in Sn
exceeds the total number of white points by exactly 2n (there is exactly one more black point in each
row of Sn ). Therefore,
2n ≤ k + m.
There are exactly n adjacent black points in Sn (call two points in Sn adjacent if their distance is
1), namely the pairs
(x, 0) and (x, −1),
for x = −n + 1, −n + 3, . . . , n − 3, n − 1. Therefore k ≤ n (the number of successive pairs of black
points in the paths in the partition of Sn cannot exceed the total number of adjacent pairs of black
points in Sn ) and we have 2n ≤ k + m ≤ n + m, yielding
n ≤ m.
4. Let P be a convex polygon with n sides, n ≥ 3. Any set of n − 3 diagonals of P that do not intersect
in the interior of the polygon determine a triangulation of P into n − 2 triangles. If P is regular and
there is a triangulation of P consisting of only isosceles triangles, find all the possible values of n.
Solution. The answer is n = 2m+1 + 2k , where m and k are nonnegative integers. In other words, n
is either a power of 2 (when m + 1 = k) or the sum of two nonequal powers of 2 (with 1 = 20 being
considered as a power of 2).
We start with the following observation.
Lemma. Let Q = Q0 Q1 . . . Qt be a convex polygon with Q0 Q1 = Q1 Q2 = · · · = Qt−1 Qt . Suppose
that Q is cyclic and its circumcenter does not lie in its interior. If there is a triangulation of Q
consisting only of isosceles triangles, then t = 2a , where a is a positive integer.
Proof. We call an arc minor if its arc measure is less than or equal to 180◦ . By the given conditions,
\
points Q1 , . . . , Qt−1 lie on the minor arc Q0 Qt of the circumcircle, so none of the angles Qi Qj Qk
(0 ≤ i < j < k ≤ t) is acute. (See the left-hand side diagram shown below.) It is not difficult to see
that Q0 Qt is longer than each other side or diagonal of Q. Thus Q0 Qt must be the base of an isosceles
triangle in the triangulation of Q. Therefore, t must be even. We write t = 2s. Then Q0 Qs Qt is an
isosceles triangle in the triangulation. We can apply the same process to polygon Q0 Q1 . . . Qs and
show that s is even. Repeating this process leads to the conclusion that t = 2a for some positive
integer a.
The results of the lemma can be generalized by allowing a = 0 if we consider the degenerate case
Q = Q0 Q1 .
P1 P 13
Q0 Qs
P1
Qt
P 11 P9 P5
36
We are ready to prove our main result. Let P = P1 P2 . . . Pn denote the regular polygon. There
is an isosceles triangle in the triangulation such that the center of P lies within the boundary of
the triangle. Without loss of generality, we may assume that P1 Pi Pj , with P1 Pi = P1 Pj (that is,
Pj = Pn−i+2 ), is this triangle. Applying the Lemma to the polygons P1 . . . Pi , Pi . . . Pj , and Pj . . . P1 ,
we conclude that there are 2m − 1, 2k − 1, 2m − 1 (where m and k are nonnegative integers) vertices in
the interiors of the minor arcs P[ d [ m k
1 Pi , Pi Pj , Pj P1 , respectively. (In other words, i = 2 + 1, j = 2 + i.)
Hence
n = 2m − 1 + 2k − 1 + 2m − 1 + 3 = 2m+1 + 2k ,
where m and k are nonnegative integers. The above discussion can easily lead to a triangulation
consisting of only isosceles triangles for n = 2m+1 + 2k . (The middle diagram shown above illustrates
the case n = 18 = 23+1 + 21 . The right-hand side diagram shown above illustrates the case n = 16 =
22+1 + 23 .)
5. Three nonnegative real numbers r1 , r2 , r3 are written on a blackboard. These numbers have the
property that there exist integers a1 , a2 , a3 , not all zero, satisfying a1 r1 + a2 r2 + a3 r3 = 0. We are
permitted to perform the following operation: find two numbers x, y on the blackboard with x ≤ y,
then erase y and write y − x in its place. Prove that after a finite number of such operations, we can
end up with at least one 0 on the blackboard.
Solution. If two of the ai vanish, say a2 and a3 , then r1 must be zero and we are done. Assume at
most one ai vanishes. If any one ai vanishes, say a3 , then r2 /r1 = −a1 /a2 is a nonnegative rational
number. Write this number in lowest terms as p/q, and put r = r2 /p = r1 /q. We can then write
r1 = qr and r2 = pr. Performing the Euclidean algorithm on r1 and r2 will ultimately leave r and 0
on the blackboard. Thus we are done again.
Thus it suffices to consider the case where none of the ai vanishes. We may also assume none of the
ri vanishes, as otherwise there is nothing to check. In this case we will show that we can perform an
operation to obtain r10 , r20 , r30 for which either one of r10 , r20 , r30 vanishes, or there exist integers a01 , a02 ,
a03 , not all zero, with a01 r10 + a02 r20 + a03 r30 = 0 and
|a01 | + |a02 | + |a03 | < |a1 | + |a2 | + |a3 |.
After finitely many steps we must arrive at a case where one of the ai vanishes, in which case we
finish as above.
If two of the ri are equal, then we are immediately done by choosing them as x and y. Hence we may
suppose 0 < r1 , r2 < r3 . Since we are free to negate all the ai , we may assume a3 > 0. Then either
a1 < − 12 a3 or a2 < − 21 a3 (otherwise a1 r1 + a2 r2 + a3 r3 > (a1 + 12 a3 )r1 + (a2 + 21 a3 )r2 > 0). Without
loss of generality, we may assume a1 < − 12 a3 . Then choosing x = r1 and y = r3 gives the triple
(r10 , r20 , r30 ) = (r1 , r2 , r3 − r1 ) and (a01 , a02 , a03 ) = (a1 + a3 , a2 , a3 ). Since a1 < a1 + a3 < 21 a3 < −a1 , we
have |a01 | = |a1 + a3 | < |a1 | and hence this operation has the desired effect.
6. At a certain mathematical conference, every pair of mathematicians are either friends or strangers.
At mealtime, every participant eats in one of two large dining rooms. Each mathematician insists
upon eating in a room which contains an even number of his or her friends. Prove that the number
of ways that the mathematicians may be split between the two rooms is a power of two (i.e., is of
the form 2k for some positive integer k).
Solution 1. Let n be the number of participants at the conference. We proceed by induction on n.
37
If n = 1, then we have one participant who can eat in either room; that gives us total of 2 = 21
options.
Let n ≥ 2. The case in which some participant, P , has no friends is trivial. In this case, P can eat
in either of the two rooms, so the total number of ways to split n participants is twice as many as
the number of ways to split (n − 1) participants besides the participant P . By induction, the latter
number is a power of two, 2k , hence the number of ways to split n participants is 2 × 2k = 2k+1 , also
a power of two. So we assume from here on that every participant has at least one friend.
We consider two different cases separately: the case when some participant has an odd number of
friends, and the case when each participant has an even number of friends.
Case 1: Some participant, Z, has an odd number of friends.
Remove Z from consideration and for each pair (X, Y ) of Z’s friends, reverse the relationship between
X and Y (from friends to strangers or vice versa).
Claim. The number of possible seatings is unchanged after removing Z and reversing the relationship
between X and Y in each pair (X, Y ) of Z’s friends.
Proof. Suppose we have an arrangement prior to Z’s departure. By assumption, Z has an even
number of friends in the room with him.
If this number is 0, the room composition is clearly still valid after Z leaves the room.
If this number is positive, let X be one of Z’s friends in the room with him. By assumption, person
X also has an even number of friends in the same room. Remove Z from the room; then X will
have an odd number of friends left in the room, and there will be an odd number of Z’s friends in
this room besides X. Reversing the relationship between X and each of Z’s friends in this room will
therefore restore the parity to even.
The same reasoning applies to any of Z’s friends in the other dining room. Indeed, there will be an
odd number of them in that room, hence each of them will reverse relationships with an even number
of individuals in that room, preserving the parity of the number of friends present.
Moreover, a legitimate seating without Z arises from exactly one arrangement including Z, because
in the case under consideration, only one room contains an even number of Z’s friends.
Thus, we have to double the number of seatings for (n − 1) participants which is, by the induction
hypothesis, a power of 2. Consequently, for n participants we will get again a power of 2 for the
number of different arrangements.
Case 2: Each participant has an even number of friends.
In this case, each valid split of participants in two rooms gives us an even number of friends in either
room.
Let (A, B) be any pair of friends. Remove this pair from consideration and for each pair (C, D),
where C is a friend of A and D is a friend of B, change the relationship between C and D to the
opposite; do the same if C is a friend of B and D is a friend of A. Note that if C and D are friends
of both A and B, their relationship will be reversed twice, leaving it unchanged.
Consider now an arbitrary participant X different from A and B and choose one of the two dining
rooms. [Note that in the case under consideration, the total number of participants is at least 3, so
such a triplet (A, B; X) can be chosen.] Let A have m friends in this room and let B have n friends
in this room; both m and n are even. When the pair (A, B) is removed, X’s relationship will be
38
reversed with either n, or m, or m + n − 2k (for k the number of mutual friends of A and B in the
chosen room), or 0 people within the chosen room (depending on whether he/she is a friend of only
A, only B, both, or neither). Since m and n are both even, the parity of the number of X’s friends
in that room will be therefore unchanged in any case.
Again, a legitimate seating without A and B will arise from exactly one arrangement that includes
the pair (A, B): just add each of A and B to the room with an odd number of the other’s friends,
and then reverse all of the relationships between a friend of A and a friend of B. In this way we create
a one-to-one correspondence between all possible seatings before and after the (A, B) removal.
Since the number of arrangements for n participants is twice as many as that for (n − 2) participants,
and that number for (n − 2) participants is, by the induction hypothesis, a power of 2, we get in turn
a power of 2 for the number of arrangements for n participants. The problem is completely solved.
Solution 2. Let n be the number of mathematicians. Represent the problem as a graph G, with
vertices and edges representing mathematicians and friendships, respectively. We seek partitions
V0 ∪ V1 such that in each induced subgraph G[Vi ], every vertex has even degree.
Label the vertices {v1 , . . . , vn }, and let di be the degree of vi . Let the notation vi ∼ vj indicate
adjacency in the graph. We introduce A = (aij ), an n × n matrix over the finite field F2 = {0, 1}.
For every i 6= j, let the entry aij = 1 if vi ∼ vj , and 0 otherwise. On the diagonal, let each aii equal
di modulo 2 (recall that our matrix has elements in F2 ).
Let d be the n × 1 vector with entries equal to the degrees di . Consider an n × 1 vector of variables
x. We claim that there is a bijective correspondence between the solutions of the matrix equation
Ax = d and the valid partitions of the graph. Indeed, for any solution vector x = (x1 , . . . , xn )T ,
partition the vertices by putting vi in V0 if xi = 0, and in V1 otherwise. It is easy to check that
Ax = d if and only if this is a valid partition.
Basic linear algebra tells us that the number of solutions of Ax = d is either 0 or a power of
|F2 | = 2. Therefore, it remains to show that there is at least one solution. In fact, this is precisely
the statement of Gallai’s Cycle-Cocycle Partition Theorem (c.f. Exercise 1.35 in Graph Theory, by
Reinhard Diestel, Problem 5.17 in Combinatorial Problems and Solutions, by László Lovász, with
solution on page 287 of the second book, or the paper “Simple proofs to three parity theorems,” by
Yair Caro, published in 1996 in Ars Combinatoria, on pages 175–180 of volume 42).
In the remainder of this solution, we provide an alternate proof of this final existence result. We
need to show that d is in the column space of A. By linear algebra, the column space of A is the
orthogonal complement of the nullspace of AT . Therefore, it suffices to show that for every vector
y = (y1 , . . . , yn )T in the null space of AT (i.e., AT y = 0), we always have yT d = 0.
Consider the submatrix B of A consisting of the rows and columns i for which yi = 1. Since the
diagonal of A is d, the product yT d is precisely the number of 1’s on the diagonal of B. Now
AT y = 0, so the sum of every row in B must be 0 over F2 , hence the total number of 1’s in B is even.
But B is symmetric, so it has an even number of off-diagonal 1’s. Therefore, B indeed has an even
number of 1’s on its diagonal, so our previous observation implies that yT d = 0 over F2 , as desired.
Remark: This problem is closely related to the notion of odd-parity covers, studied by K. Sutner
(see the above paper by Y. Caro for references). Let each vertex of a graph by equipped with a switch
and a light bulb. Flipping the switch at a vertex v toggles the state of the bulb at v, as well as all
of the bulbs at the vertices adjacent to v. Initially, all lights are off. It turns out that it is always
possible to flip a collection of switches such that all lights are simultaneously on. Interestingly, this
problem appeared in 1997 on the Iranian Olympiad (in different terms). We challenge the interested
reader to show that this result implies the existence result that we needed above, and vice versa.
39
4 Team Selection Test 2008
49th IMO Team Selection Test
Washington, D.C.
June 7 and 8, 2008
4.1 TST 2008 Problems

1. There is a set of n coins with distinct integer weights w1 , w2 , . . . , wn . It is known that if any coin
with weight wk , 1 ≤ k ≤ n, is removed from the set, the remaining coins can be split into two groups
of the same weight. (The number of coins in the two groups can be different.) Find all n for which
such a set of coins exists.
2. Let P, Q, and R be the points on sides BC, CA, and AB of an acute triangle ABC such that
triangle P QR is equilateral and has minimal area among all such equilateral triangles. Prove that
the perpendiculars from A to line QR, from B to line RP , and from C to line P Q are concurrent.
3. For a pair A = (x1 , y1 ) and B = (x2 , y2 ) of points on the coordinate plane, let d(A, B) = |x1 − x2 | +
|y1 − y2 |. We call a pair (A, B) of (unordered) points harmonic if 1 < d(A, B) ≤ 2. Determine the
maximum number of harmonic pairs among 100 points on the plane.
(This problem was suggested by Zuming Feng and Oleg Golberg.)
4. Prove that for no integer n is n7 + 7 a perfect square.
(This problem was suggested Titu Andreescu.)
5. Two sequences of integers, a1 , a2 , a3 , . . . and b1 , b2 , b3 , . . ., satisfy the equation
(an − an−1 )(an − an−2 ) + (bn − bn−1 )(bn − bn−2 ) = 0
for each integer n greater than 2. Prove that there is a positive integer k such that ak = ak+2008 .
(This problem was suggested by Gabriel Carroll.)
6. Let n be a positive integer not divisible by the cube of a prime. Given an integer-coefficient polynomial
f (x), define its signature modulo n to be the (ordered) sequence f (1), . . . , f (n) modulo n. Of the nn
such n-term sequences of integers modulo n, how many are the signature of some polynomial f (x)?
(This problem was suggested by Po-Ru Loh.)
40
4.2 TST 2008 Solutions
1. There is a set of n coins with distinct integer weights w1 , w2 , . . . , wn . It is known that if any coin
with weight wk , 1 ≤ k ≤ n, is removed from the set, the remaining coins can be split into two groups
of the same weight. (The number of coins in the two groups can be different.) Find all n for which
such a set of coins exists.
Solution. The only such n are 1 and odd n at least 7. We divide into cases: n even, n = 1, n = 3,
n = 5, and n ≥ 7 odd.
Case 1: n even. Suppose for contradiction that such a set exists, and choose one with minimal total
weight. Let s be the sum of the weights, and note that
s − w1 ≡ · · · ≡ s − wn ≡ 0 (mod 2).
Hence
w1 ≡ . . . ≡ wn (mod 2).
Since n is even, it follows that s is even, and so also that wi is even for all i. But then the set {wi /2}
is another set of n weights with the desired property, contradicting minimality. Hence n odd fails.
Case 2: n = 1. Trivially yes.
Case 3: n = 3. Removing any weight leaves two unequal weights.
Case 4: n = 5. Suppose for contradiction that such a set exists: order the weights such that
w1 < w2 < w3 < w4 < w5 . Let t = w2 + w3 + w4 . For each i ∈ {2, 3, 4}, the only possible ways to
split the weights other than wi are as w5 = t − wi + w1 or w5 + w1 = t − wi (in all other combinations,
the side with w5 is strictly heavier than the other). By pigeonhole, One of the equations is satisfied
for two values of i, and the corresponding weights are equal, contradiction.
Case 5: n ≥ 7 odd. Checking the cases n = 7, n = 9, and n = 11 is straightforward casework: the
sets {1, 3, 5, . . . , 2n − 1} suffice. For example, if n = 7, the set is {1, 3, 5, . . . , 13} and
3 + 5 + 7 + 9 = 11 + 13
1 + 9 + 13 = 5 + 7 + 11
1 + 3 + 7 + 11 = 9 + 13
1 + 9 + 11 = 3 + 5 + 13
1 + 3 + 5 + 11 = 7 + 13
1 + 5 + 13 = 3 + 7 + 9
1 + 3 + 5 + 9 = 7 + 11.
Suppose there exists such a set for n = k and n = j, let w1 < w2 < · · · < wk and v1 < · · · <
vj be the corresponding set of weights and let s = v1 + · · · + vj . Consider the set of weights
w1 v1 , w1 v2 , . . . , w1 vj , w2 s, w3 s, . . . , wk s. These are distinct (each is less than the next in the list by
assumption). If weight w1 vi is removed, then by the inductive hypothesis the other v’s can be sorted
into two groups of equal weight, as can the w’s other than w1 , so the remaining weights can be divided
into two groups of equal weight. If weight wi s is removed, then put all the weights w1 v1 , . . . , w1 vj
together to get weights w1 s, w2 s . . . , wi−1 s, wi+1 s, . . . , wk s. By the inductive hypothesis these can be
divided into two groups of equal weight. Hence if such a set exists for n = k and n = j, it exists for
k + j − 1.
Taking j = 7, we see that if a set exists for n = k, one also exists for n = k + 6. By induction, it
follows that a set exists for all n of the form 6m + 7, 6m + 9, and 6m + 11, where m is a non-negative
integer. Since this covers all odd numbers greater than 7, we are done.
41
2. Let P, Q, and R be the points on sides BC, CA, and AB of an acute triangle ABC such that
triangle P QR is equilateral and has minimal area among all such equilateral triangles. Prove that
the perpendiculars from A to line QR, from B to line RP , and from C to line P Q are concurrent.
R1
R
X
B Q
P1 Q1
P
Solution. (By Zuming Feng) By Miquel’s theorem (which can be shown by simple angle chasing),
the circumcircles of triangles AQR, BRP , and CP Q meet at a common point X. The key observation
is that XR ⊥ AB, XP ⊥ BC, and XQ ⊥ CA. Indeed, if P1 Q1 R1 is an inscribed equilateral triangle,
and the circumcircles of triangles AQ1 R1 , BR1 P1 , and CP1 Q1 meet at a common point X. Let P, Q,
and R be the feet of the perpendiculars from X to the sides of the triangle. Quick angle chasing
(∠RR1 X = ∠P P1 X = QQ1 X) shows that right triangles XP P1 , XQQ1 , and XRR1 are similar, and
so triangles P QR and P1 Q1 R1 are similar. Clearly, P QR is a smaller triangle, and this establishes
our observation.
B a
a
Y
a X
P Q
It is then straightforward to check that the perpendiculars from A to QR, B to RP , and C to P Q

meet at the isogonal conjugate of X.
Remark: The proof shows that the condition of triangle P QR being equilateral is unnecessary.
42
3. For a pair A = (x1 , y1 ) and B = (x2 , y2 ) of points on the coordinate plane, let d(A, B) = |x1 − x2 | +
|y1 − y2 |. We call a pair (A, B) of (unordered) points harmonic if 1 < d(A, B) ≤ 2. Determine the
maximum number of harmonic pairs among 100 points on the plane.
Solution. The answer is 3750.

We claim that there do not exist five points on the plane such that they are pairwise harmonic.
Suppose such a set of five points exists. We say (x1 , y1 ) precedes (x2 , y2 ), denoted by (x1 , y1 ) ≺
(x2 , y2 ), if and only if x1 ≤ x2 and y1 ≤ y2 . This defines a partial order on the set. Because
it has five elements, it has either a chain or an antichain of three elements. Let A, B, C in the
order of their x-coordinates form a chain or an antichain under the ordering we defined. Then
d(A, B) + d(B, C) = d(A, C) and all pairs of A, B, C cannot be good. (This may also be proved by
invoking the pigeon-hole principle to prove the more general result of Erdős-Szekeres: a sequence of
mn + 1 real numbers either contains a non-increasing sequence of length m + 1 or a non-decreasing
sequence of length n + 1.)
Consider the graph with vertices at 100 points of the plane and edges connecting good pairs. Then
due to the above statement the graph has no complete subgraph with 5 vertices. By Turan’s theorem
it has at most µ ¶ µ ¶
100 25
−4· = 3750 = 6 · 252
2 2
edges. Therefore, every set of 100 points of the plane has at most 3750 harmonic pairs.
To complete our proof, we show that the bound 3750 can be attained. Consider the square with
vertices at A(0, 1), B(1, 0), C(0, −1), D(−1, 0). On each of the sides AB, BC, CD, DA choose 25
points close to A, B, C, D, respectively. Among the chosen 100 points, any pair of two points lying
on different sides of the square is good. Therefore, 6 · 252 = 3750 is the maximum number of good
pairs among 100 points of the plane.
4. Prove that for no integer n is n7 + 7 a perfect square.
Solution. Assume on the contrary that n7 + 7 = m2 for some integer m.

For n even, n7 + 7 ≡ 3 (mod 4), so it cannot be a perfect square. For n ≡ 3 (mod 4), n7 + 7 ≡ 2
(mod 4), so again it cannot be a perfect square. We must have n ≡ 1 (mod 4) and m even.
Note that n7 + 27 = n7 + 7 + 121 = m2 + 112 . Thus n + 2 is a divisor of m2 + 112 . But n + 2 ≡ 3
(mod 4), and if gcd(m, 11) = 1, this is impossible by a well-known fact about sums of two squares.
We must have gcd(m, 11) = 11, then m = 11m1 for some even integer m1 and
(n + 2)(n6 − 2n5 + 22 · n4 − 23 · n3 + 24 · n2 − 25 · n + 26 ) = n7 + 27 = 112 (m21 + 1).
Since n + 2 ≡ 3 (mod 4) and m21 + 1 ≡ 1 (mod 4), n6 − 2n5 + 22 · n4 − 23 · n3 + 24 · n2 − 25 · n + 26

is congruent to 3 modulo 4. Since all the divisors of m21 + 1 are congruent to 1 modulo 4, 11 must
divide each of n + 2 and n6 − 2n5 + 22 · n4 − 23 · n3 + 24 · n2 − 25 · n + 26 . But this is impossible,
because 11 | (n + 2) implies that n ≡ −2 (mod 11) and
n6 − 2n5 + 22 · n4 − 23 · n3 + 24 · n2 − 25 · n + 26 ≡ 7 · 26 6≡ 0 (mod 11).
Therefore, our assumption was wrong and n7 + 7 is not a perfect square for any integer n.
43
Remark: We can also prove that if
µ ¶
n7 + 27
d = gcd n + 2, = gcd(n + 2, n6 − 2n5 + 22 · n4 − 23 · n3 + 24 · n2 − 25 · n + 26 ≡ 7 · 26 ),
n+2
then d | 7. Indeed, from the identities a7 + b7 = (a + b)7 − 7ab(a + b)(a2 + ab + b2 )2 and a2 + ab + b2 =

(a + b)2 − ab, it follows that
a7 + b7
= (a + b)6 − 7ab(a + b)4 + 14a2 b2 (a + b)2 − 7a3 b3 .
a+b
7 7
For a = n and b = 2 we have d | (a+b) and d | aa+b
+b
, hence d | 7a3 b3 = 7(2n)3 . But gcd(n+2, 2n) = 1,
as n is odd, so gcd(d, (2n)3 ) = 1 and so d | 7.
5. Two sequences of integers, a1 , a2 , a3 , . . . and b1 , b2 , b3 , . . ., satisfy the equation
(an − an−1 )(an − an−2 ) + (bn − bn−1 )(bn − bn−2 ) = 0
for each integer n greater than 2. Prove that there is a positive integer k such that ak = ak+2008 .
Solution 1. Define d(i, j) = (ai − aj )2 + (bi − bj )2 . Notice that
(an − an−1 )2 + (an − an−2 )2 − (an−1 − an−2 )2 = 2(an − an−1 )(an − an−2 )
and
(bn − bn−1 )2 + (bn − bn−2 )2 − (bn−1 − bn−2 )2 = 2(bn − bn−1 )(bn − bn−2 ).
Adding these two equations and using the given implies that
d(n, n − 1) + d(n, n − 2) − d(n − 1, n − 2) = 0,
hence
d(n, n − 1) = d(n − 1, n − 2) − d(n, n − 2) ≤ d(n − 1, n − 2),
where equality holds if and only if an = an−2 and bn = bn−2 . Therefore, the sequence
{d(n, n − 1)}∞
n=2 = (d(2, 1), d(3, 2), d(4, 3), . . . )
is nonincreasing. Since all the terms of this sequence are nonnegative integers, the sequence must
eventually become constant: d(n, n−1) = d(n−1, n−2) for all sufficiently large n, which in particular
implies an = an−2 for all sufficiently large n. So whenever n is large enough, we have
an+2008 = an+2006 = an+2004 = · · · = an .
Solution 2. (Based on work by Palmer Mebane) Let xn = an − an−1 and yn = bn − bn−1 . Then the
given equation reads
xn (xn + xn−1 ) + yn (yn + yn−1 ) = 0.
We wish to show that xk + xk+1 + · · · + xk+2007 = 0.
We claim that, for each n, one of the following holds:
• xn = −xn−1 and yn = −yn−1 (in particular, x2n + yn2 = x2n−1 + yn−1

2 )
44
• x2n + yn2 < x2n−1 + yn−1
2
To see this, note that 0 ≤ (xn−1 + xn )2 + (yn−1 + yn )2 , with equality if and only if xn = −xn−1 and
yn = −yn−1 . Expanding and using the given equation we obtain
0 = 2xn (xn + xn−1 ) + 2yn (yn + yn−1 ) ≤ (xn−1 + xn )2 + (yn−1 + yn )2
or
x2n + yn2 ≤ x2n−1 + yn−1
2
,
with equality if and only if xn = −xn−1 and yn = −yn−1 . Thus we have the claim.
Therefore, because x2n + yn2 ≥ 0 is an integer, by the well-ordering principle, x2n + yn2 < x2n−1 + yn−1
2
can only hold finitely many times. Thus there is a k such that for all n > k, xn = −xn−1 and
yn = −yn−1 . This gives us the desired k such that ak = ak+2008 .
6. Let n be a positive integer not divisible by the cube of a prime. Given an integer-coefficient polynomial
f (x), define its signature modulo n to be the (ordered) sequence f (1), . . . , f (n) modulo n. Of the nn
such n-term sequences of integers modulo n, how many are the signature of some polynomial f (x)?
Solution. Let p1 · · · pr q12 · · · qs be the prime factorization of n. Then the number of achievable
signatures is pp11 · · · ppr r q13q1 · · · qs3qs .
We begin by reducing to the case in which n is a prime or the square of a prime. Assume for
now that our claim holds in these cases. Observe that the signature of a polynomial f (x) modulo
n is completely determined by the collection of its signatures modulo p1 , . . . , pr , q12 , . . . , qs2 . Indeed,
for each such modulus mi , the sequence f (1), . . . , f (mi ) modulo mi determines the entire sequence
f (1), . . . , f (n) modulo mi because a ≡ b (mod mi ) implies f (a) ≡ f (b) (mod mi ). Since the moduli
p1 , . . . , pr , q12 , . . . , qs2 are relatively prime with product n, it follows by the Chinese Remainder Theorem
that specifying f (1), . . . , f (n) modulo all mi is equivalent to specifying the sequence modulo n.
Conversely, we need to show that every sequence y1 , . . . , yn (mod n) obtained by applying the above
procedure to a collection of signatures fi (1), . . . , fi (mi ) (mod mi ) is actually the signature of a poly-
nomial f (x). To do this, we apply the Chinese Remainder Theorem once more: define f (x) by
choosing its coefficients such that
[xk ]f (x) ≡ [xk ]fi (x) (mod mi )
for all i. Then f (x) ≡ fi (x) (mod mi ) for all i, and the claim follows.
By the above, it suffices to show that the number of achievable signatures modulo p is pp and the
number of achievable signatures modulo p2 is p3p . In the first case, this amounts to showing that all
p-term sequences modulo p can be realized as f (1), . . . , f (p) for some polynomial f (x). There are a
number of quick proofs of this result, all of which ultimately use the fact that the integers modulo p
form a field. A direct approach is to use Lagrange interpolation; another method is to use Lagrange’s
theorem to deduce that no two polynomials in Z/pZ of degree less than p have the same signature
modulo p. Yet another approach is to show that the linear transformation (over Z/pZ) that sends
the coefficient vector ([x0 ]f (x), . . . , [xp−1 ]f (x)) to the values (f (1), . . . , f (p)) is nonsingular; this can
be done by evaluating the Vandermonde determinant.
The modulus p2 is trickier to handle. We begin by using an idea similar to Lagrange interpolation
to exhibit a set of polynomials achieving p3p distinct signatures modulo p2 , thus obtaining a lower
45
bound. We then establish a matching upper bound by finding a (different!) set of p3p polynomials
such that all integer-coefficient polynomials can be easily “reduced” to one of these, keeping the
signature fixed during the reduction.
To prove our lower bound, we begin by defining
fk (x) = ((x − 1) · · · (x − (k − 1))(x − (k + 1)) · · · (x − p))2
and
gk (x) = (x − k)fk (x).
Consider all polynomials of the form
p
X
ak fk (x) + bk gk (x)
k=1
where 1 ≤ ak ≤ p2 and 1 ≤ bk ≤ p. We claim that no two such polynomials have the same signature
modulo p2 ; this will establish our lower bound.
By subtracting, it is enough to show that the only polynomial
p
X
f (x) = ak fk (x) + bk gk (x)
k=1
satisfying f (1) ≡ · · · ≡ f (p2 ) ≡ 0 (mod p2 ) with |ak | < p2 and |bk | < p is the zero polynomial. In
fact, it suffices to evaluate f (x) modulo p2 at x = 1, . . . , 2p. For j = 1, . . . , p, we have gk (j) = 0
for all k and fk (j) = 0 unless k = j, in which case fj (j) is invertible modulo p2 . It follows that
0 ≡ f (j) = aj fj (j) and hence aj ≡ 0 (mod p2 ), so ak = 0 for all k.
Next we evaluate f (x) at j = p + 1, . . . 2p. This time, all gk (j) vanish modulo p2 except when
k = j − p, in which case gj−p (j) is divisible by p but not p2 . Thus, 0 ≡ f (j) ≡ bj−p gj−p (j) (mod p2 )
and hence bj−p ≡ 0 (mod p), so bk = 0 for all k, completing the proof of the lower bound.
It remains to prove that no more than p3p signatures modulo p2 are attainable. Given two polynomials
f (x) and g(x), we say that they are equivalent, denoted f (x) ∼ g(x), if they have the same signature
modulo p2 . It is easy to see that f (x) ∼ g(x) implies f (x)h(x) ∼ g(x)h(x) and f (x) + h(x) ∼ g(x) +
h(x) for any polynomial h(x). We claim that every polynomial f (x) is equivalent to a polynomial
g(x) such that
[xk ]g(x) ∈ {1, . . . , p2 } for 0 ≤ k < p
and
[xk ]g(x) ∈ {1, . . . , p} for p ≤ k < 2p.
The key observation is that by Fermat, xp − x vanishes modulo p when evaluated at any integer.
Therefore, (xp − x)2 and p(xp − x) both vanish modulo p2 at all integers, i.e.
x2p ∼ 2xp+1 − x2 and pxp ∼ px.
Using the properties of ∼ stated above, the first relation allows us to reduce f (x) to a polynomial
of degree less than p; the second relation then allows us to further reduce to a polynomial with all
coefficients of xp , . . . , x2p−1 in {1, . . . , p}. Finally, the remaining coefficients can be reduced modulo
p2 to lie in {1, . . . , p2 }, completing the proof.
Remark: One can also try to apply the above techniques to more general n, but the lower bound
already breaks down in the special case p = 2 when we allow p3 | n.
46
5 IMO 2006
47th International Mathematical Olympiad
Ljubljana, Slovenia
July 12 and 13, 2007
5.1 IMO 2006 Problems

1. Let ABC be a triangle with incenter I. A point P in the interior of the triangle satisfies
∠P BA + ∠P CA = ∠P BC + ∠P CB.
Show that AP ≥ AI, and that equality holds if and only if P = I.
(This problem was suggested by the Republic of Korea.)
2. Let P be a regular 2006-gon. A diagonal of P is called a good segment if its endpoints divide the
boundary of P into two parts, each composed of an odd number of sides of P. The sides of P are
also called good segments.
Suppose P has been dissected into triangles by 2003 diagonals, no two of which have a common point
in the interior of P. Find the maximum number of isosceles triangles having two good segments that
could appear in such a configuration.
(This problem was suggested by Serbia and Montenegro.)
3. Determine the least real number M such that the inequality

|ab(a2 − b2 ) + bc(b2 − c2 ) + ca(c2 − a2 )| ≤ M (a2 + b2 + c2 )2
holds for all real numbers a, b, and c.
(This problem was suggested by Ireland.)
4. Determine all pairs (x, y) of integers such that

1 + 2x + 22x+1 = y 2 .
(This problem was suggested by the United States of America.)
5. Let P (x) be a polynomial of degree n > 1 with integer coefficients and let k be a positive integer.
Consider the polynomial
Q(x) = P (P (. . . (P (x) . . . ))
| {z }
k P 0s
Prove that there are at most n integers t such that Q(t) = t.
(This problem was suggested by Romania.)
6. Assign to each side b of a convex polygon P the maximum area of a triangle that has b as a side and
is contained in P. Show that the sum of the areas assigned to the sides of P is at least twice the area
of P.
(This problem was suggested by Serbia and Montenegro.)
47
5.2 IMO 2006 Solutions
1. Let ABC be a triangle with incenter I. A point P in the interior of the triangle satisfies
∠P BA + ∠P CA = ∠P BC + ∠P CB.
Show that AP ≥ AI, and that equality holds if and only if P = I.
Solution. We begin by proving a well-known fact.
A
I
P
B C
Lemma. Let ABC be a triangle with circumcenter O, circumcircle γ, and incenter I. Let M be the
second intersection of line AI with γ. Then M is the circumcenter of triangle IBC.
Proof. Let ∠A = 2α, ∠B = 2β. Note that M is on the opposite side of line BC as A. We have
∠CBM = ∠CAM = α, so that ∠IBM = ∠IBC+∠CBM = β+α. Also, ∠BIM = ∠BAI+∠ABI =
α + β. Thus, triangle IBM is isosceles with BM = IM . Similarly, CM = IM . This proves the
claim.
Back to our current problem, we note that

(∠P BA + ∠P CA) + (∠P BC + ∠P CB) = ∠B + ∠C,
so
1
∠P BA + ∠P CA = ∠P BC + ∠P CB = (∠B + ∠C).
2
In triangle P BC, we have
1
∠BP C = 180◦ − (∠P BC + ∠P CB) = 180◦ − (∠B + ∠C).
2
It is clear that ∠IBC + ∠ICB = 21 (∠B + ∠C), and so in triangle BCI,
1
∠BIC = 180◦ − (∠B + ∠C).
2
We conclude that ∠BP C = ∠BIC; that is, points B, C, I, and P lie on a circle. By the Lemma,
they all lie on a circle centered at M . In particular, we have M P = M I.
In triangle AP M , we have
AI + IM = AM ≤ AP + P M = AP + IM,
implying that AI ≤ AP . Equality holds if and only if AM = AP + P M ; that is, A, P , and M are
collinear, or P = I.
48
2. Let P be a regular 2006-gon. A diagonal of P is called a good segment if its endpoints divide the
boundary of P into two parts, each composed of an odd number of sides of P. The sides of P are
also called good segments.
Suppose P has been dissected into triangles by 2003 diagonals, no two of which have a common point
in the interior of P. Find the maximum number of isosceles triangles having two good segments that
could appear in such a configuration.
Note: Let M denote the maximum we are looking for. The answer is M = 1003.
Let ω denote the circumcircle of P, and let P = P1 P2 . . . P2006 , where points P1 , . . . , P2006 are arranged
in the clockwise direction along ω. Then Pi Pj is good if and only if i − j is odd. We call an isosceles
triangle (in T ) good if it has two good segments. Since 2006 is even, a good triangle has exactly
two good sides. Any set of 2003 diagonals of P that do not intersect in the interior of the polygon
determines a triangulation of P into 2004 triangles. Let T denote such a triangulation.
It is easy to see that M ≥ 1003. We can first use diagonals P1 P3 , P3 P5 , . . . , P2003 P2005 , and P2005 P1
to obtain 1003 good triangles. We can then complete the triangulation easily by a triangulations of
P1 P3 · · · P2005 using 1001 diagonals. We present two solutions showing that M ≤ 1003.
Let Pd i Pj denote the directed (clockwise direction) broken line segment Pi Pi+1 . . . Pj (where P2006+k =
Pk ). We say Pd i Pj is non-major if it contains at most 1003 sides of P.
Solution 1. We start with the following lemma.
Lemma. Let Pi Pj be a diagonal used in T such that Pdi Pj is non-major and contains n segments of
¥ ¦
P. Then, there are at most n2 good triangles with vertices on Pd i Pj . More precisely, there are at
most  j k
 j−i , if i < j
j 2 k
 j−i+2006 , if i > j
2
good triangles with vertices on Pd

i Pj .
Proof. Without loss of generality, we may assume that i < j. We induct on n.

The base cases for n = 1 and n = 2 are trivial. Assume the statement is true for n with n ≤ k and
2 ≤ k < 1003. We consider the case n = k + 1.
Let Pi Pa Pj be a triangle in T with Pa on Pd lie on¦ non-major arc Pd
i Pj . (Note that Pi , Pa , and Pj ¥ i Pj
a−i
on ω in clockwise order). By j the induction
k hypothesis, there are at most 2 good triangles with
[ j−a [
vertices on P P
i a and at most 2 good triangles on P P
a j . We now have two cases:
• Case 1: Pi Pa Pj is not good.

In this case, applying the inductive hypothesis, there are at most
¹ º ¹ º ¹ º
a−i j−i j−i
+ ≤
2 2 2

i Pj , as desired.
49
• Case 2: Pi Pa Pj is good.
Since Pdi Pj is non-major, Pi Pj > Pi Pa and Pi Pj > Pa Pj . We must have that Pi Pa and Pa Pj are
the two equal good sides, so both are in particular good. Hence both a − i and j − a are odd,
giving a total of at most
¹ º ¹ º
a−i j−i a−i 1 j−a 1 j−i
1+ + =1+ − + − =
2 2 2 2 2 2 2

i Pj .
j k
In both cases, there are at most j−i
2 good triangles on Pd
i Pj , so the induction is complete.
Now we prove our main result. Let Pi Pk be the longest diagonal used in T . Let Pi Pj Pk be a non-
obtuse triangle in T . Without loss of generality, we may assume that i < j < k. Since Pi Pj Pk is
non-obtuse, Pd [ [
i Pj , Pj Pk , and Pk Pi are all non-major. By the lemma, there are at most
¹ º ¹ º ¹ º
j−i k−j i − k + 2006 j − i k − j i − k + 2006
+ + ≤ + + = 1003 (∗)
2 2 2 2 2 2
good triangles besides Pi Pj Pk .

If Pi Pj Pk is not good, we are done. If it is, then exactly two of j − i, k − j, and i − k are odd, and so
(∗) is strict inequality. We still have at most 1002 + 1 = 1003 good triangles in this case, completing
our proof.
Solution 2. Let Pi Pj Pk (i < j < k) be a good triangle, with Pi Pj and Pj Pk being good segments.
This means that there are an odd number of sides of P between Pi and Pj and also between Pj and
Pk . We say Pd [
i Pj and Pj Pk belong to triangle Pi Pj Pk .
At least one side in each of these groups does not belong to any other good triangle. This is so
because any odd triangle whose vertices are among the points between Pi and Pj has two sides of
equal length and therefore has an even number of sides belonging to it in total. Eliminating all sides
belonging to any other good triangle in Pd i Pj must therefore leave at least one side that belongs to
no other good triangle. The same argument applies to P [ j Pk . Let us assign these two sides (one in
d [
Pi Pj and one in Pj Pk ) to triangle Pi Pj Pk .
To each good triangle we have thus assigned a pair of sides, with no two good triangles sharing an
assigned side. It follows that at most 1003 good triangles can appear in the triangulation; that is,
M ≤ 1003.
3. Determine the least real number M such that the inequality
|ab(a2 − b2 ) + bc(b2 − c2 ) + ca(c2 − a2 )| ≤ M (a2 + b2 + c2 )2
holds for all real numbers a, b, and c.
Note: Consider the polynomial
P (a, b, c) = ab(a2 − b2 ) + bc(b2 − c2 ) + ca(c2 − a2 ).
50
It is not difficult to check that P (a, a, c) = 0. Hence a − b divides P (a, b, c). Since P (a, b, c) is
cyclically symmetric, we conclude that (a − b)(b − c)(c − a) divides P (a, b, c). Since P (a, b, c) is a
cyclic homogeneous polynomial of degree 4 (each monomial in the expansion of P (a, b, c) has degree
4) and (a − b)(b − c)(c − a) is a cyclic homogeneous polynomial of degree 3,
P (a, b, c) = (a − b)(b − c)(c − a)Q(x),
where Q(x) is a cyclic homogeneous polynomial of degree 1; that is, Q(x) = k(a + b + c) for some
constant k. It is easy to deduce that k = 1 and
P (a, b, c) = (a − b)(b − c)(c − a)(a + b + c).
The given inequality now reads
|(a − b)(b − c)(c − a)(a + b + c)| ≤ M (a2 + b2 + c2 )2 . (∗)
Since the above inequality is symmetric with respect to a, b, and c, we may assume that a ≥ b gec.
(Indeed, we may assume that a > b > c, because otherwise the left-hand side of (∗) is 0, and we have
nothing to prove.) Thus, (∗) reduces to
(a − b)(b − c)(a − c)(a + b + c) ≤ M (a2 + b2 + c2 )2 (∗∗)
for real numbers a > b > c. Note also that (∗∗) is homogeneous (of degree 4). We may further
assume that a + b + c = 1. Then, (∗∗) reduces to
(a − b)(b − c)(a − c) ≤ M (a2 + b2 + c2 )2 (†)
for real numbers a > b > c with a + b + c = 1. Setting a − b = x and b − c = y, we have a − c = x + y.

Note that
(a − b)2 + (b − c)2 + (c − a)2

= 2(a2 + b2 + c2 ) − 2(ab + bc + ca)
= 2(a2 + b2 + c2 ) − [(a + b + c)2 − (a2 + b2 + c2 )]
= 3(a2 + b2 + c2 ) − 1.
We can rewrite (†) as

9xy(x + y) ≤ M [x2 + y 2 + (x + y)2 + 1]2 (‡)
for positive real numbers x and y. It suffices to find the least M satisfying (‡). There are many ways
to finish. We present two typical ones.
Solution 1. We rewrite (‡) as

µ ¶2
9(x + y) x2 + y 2 + (x + y)2 + 1
≤ √
M xy
µ ¶2
2(x + y)2 + 1 √
= √ − 2 xy .
xy
Setting
2(x + y)2 + 1 √
A= √ and B = 2 xy,
xy
51
the above inequality becomes
9(x + y)
≤ (A − B)2 .
M
2 2 2
Note that A > B > 0 as A − B = x +y +(x+y) √
xy
+1
> 0. For real numbers x and y with fixed x + y,
√
if we increase the value of xy, the left-hand side (9(x + y)/M ) of the above inequality does not
change its value, while A decreases its value (with fixed numerator and increasing denominator) and
√
B increases its value. Hence, if we increase the value of xy, A − B is a positive term with decreasing
value; that is, the right-hand side of the inequality decreases its value. Therefore, when we increases
√
the value of xy with fixed x + y, the above inequality gets strengthened. Therefore, we may assume
that x = y in the above inequality, and (‡) becomes
18x3 ≤ M (6x2 + 1)2 = M (36x4 + 12x2 + 1)
or
12 1 18
36x + + 3 ≥ .
x x M
It suffices to find the minimum value of the continuous function
12 1
f (x) = 36x + + 3 for x > 0.
x x
Note that
df 12 3 3(2x2 − 1)(6x2 + 1)
= 36 − 2 − 4 = ,
dx x x x4
implying the only critical value in the domain is at x = √12 . It is easy to check that f (x) indeed
√
obtains the global minimum 32 2 at x = √12 in the domain.
√
We conclude the minimum value of M is 9322 , obtained when x = y = a − b = b − c = √1
2
(and
a + b + c = 1); that is, µ ¶
1 1 1 1 1
(a, b, c) = +√ , , −√ .
3 2 3 3 2
Solution 2. (By Aleksandar Ivanov, observer with the Bulgarian team) By the AM-GM Inequal-
ity, we have
x2 + y 2 + (x + y)2 + 1
µ ¶ µ ¶
1 1 (x + y)2 (x + y)2
= x2 + + y2 + + +
2 2 2 2
√ √ (x + y) 2
≥ 2x + 2y + 2xy +
r 2
4
√ √ (x + y)2
≥ 4 ( 2x)( 2y)(2xy) ·
p 2
= 4 4 2x2 y 2 (x + y)2 ,
or √
(x2 + y 2 + (x + y)2 + 1)2 ≥ 16 2xy(x + y).
It is then routine to show (‡). Equality holds only if x2 = y 2 = 21 , and it is straightforward to check
it indeed leads to the equality case.
52
4. Determine all pairs (x, y) of integers such that
1 + 2x + 22x+1 = y 2 .
Note: The answers are (x, y) = (0, ±2) and (x, y) = (4, ±23). It is easy to check that these are
solutions. If (x, y) is a solution then obviously x ≥ 0 and (x, −y) is a solution too. For x = 0, we get
the first two solutions. Now we assume that (x, y) is a solution with x > 0; without loss of generality
confine attention to y > 0.
Solution 1. The equation can be rewritten as
2x (1 + 2x+1 ) = y 2 − 1 = (y − 1)(y + 1),
which shows that gcd(y − 1, y + 1) = 2, so exactly one of y − 1 and y + 1 is divisible by 4. Hence,

x ≥ 3 and one of y − 1 and y + 1 is divisible by 2x−1 but not by 2x . Consequently, we may write
y = 2x−1 m + ², (†)
where m is odd and ² = ±1. Plugging this into the original equation, we obtain
2x (1 + 2x+1 ) = (2x−1 m + ²)2 − 1 = 22x−2 m2 + 2x m²,
or
1 + 2x+1 = 2x−2 m2 + m².
It follows that
1 − m² = 2x−2 (m2 − 8). (‡)
If ² = 1, (‡) becomes m2 − 8 ≤ 0, or m = 1, which fails to satisfy (‡). Thus ² = −1, so (‡) becomes
1 + m = 2x−2 (m2 − 8) ≥ 2(m2 − 8),
implying that 2m2 − m − 17 ≤ 0. Hence m ≤ 3. On the other hand, m 6= 1 by (‡). Because m is

odd, m = 3, leading to x = 4 by (‡). Substituting these into (†) yields y = 23, completing our proof.
Solution 2. It is easy to check that there is no solution for x = 1, 2, and 3. We assume that (x, y)
is a solution with x ≥ 5 and y > 0. Note that
½
1 + 22 + 22x+1 = y 2
1 + 2x+1 + 22x = (1 + 2x )2 .
Subtracting the two equations gives
[y − (1 + 2x )][y + (1 + 2x )] = 22x − 2x = 2x (2x − 1).
It is easy to see that both y and 1 + 2x are odd and that y > 1 + 2x . We must have
½ ½
y − (1 + 2x ) = 2m y − (1 + 2x ) = 2x−1 n
or
y + (1 + 2x ) = 2x−1 n, y + (1 + 2x ) = 2m,
where m and n are positive integers with mn = 2x − 1. It is not difficult to see that the latter case
is not possible. (Indeed, y = 2m − (1 + 2x ) ≤ 2(2x − 1) − (1 + 2x ) = 2x − 3, contradicting the fact
that y > 1 + 2x .) Hence we must have the former case. Solving the system gives
y = m + 2x−2 n and 1 + 2x = 2x−2 n − m. (∗)
53
We claim that n = 5. Note that both m and n are odd. We establish our claim by showing that
3 < n < 7. Since y > 1 + 2x , we have
2x+1 + 2 = 2(1 + 2x ) < y + 1 + 2x = 2x−1 n,
implying that n is greater than 3. Hence n ≥ 5. By the second equation in (∗), we have m =
2x−2 n − 2x − 1 ≥ 5 · 2x−2 − 2x − 1 = 2x−2 − 1. If n ≥ 7, then
2x − 1 = mn > (2x−2 − 1)7 = 2x + 3 · 2x−2 − 7 > 2x − 1
for x ≥ 3. We conclude that 3 < n < 7; that is, n = 5.

Substituting n = 5 into the second equation and then the first equation in (∗) gives m = 5 · 2x−2 −
1 − 2x = 2x−2 − 1 and y = m + 2x−2 n = 3 · 2x−1 − 1. It follows that
(3 · 2x−1 − 1)2 = y 2 = 1 + 2x + 22x+1 ,
or 9 · 22x−2 − 3 · 2x = 2x + 22x+1 . Solving the last equation gives 4 · 2x = 22x−2 , leading to x = 4,

which contradicts the assumption that x ≥ 5. Hence there is no solution for x ≥ 5.
5. Let P (x) be a polynomial of degree n > 1 with integer coefficients and let k be a positive integer.
Consider the polynomial
Q(x) = P (P (. . . (P (x) . . . ))
| {z }
k P 0s
Prove that there are at most n integers t such that Q(t) = t.
Solution. Let Z denote the set of integers. We define
SP = {t | t ∈ Z and P (t) = t} and SQ = {t | t ∈ Z and Q(t) = t}.
Clearly, SP is a subset of SQ . Also note that there are at most n elements in SP . This is so because
that t ∈ SP if and only if t is a root of polynomial P (x) − x = 0 of degree n, which has at most n
roots. If SQ = SP , we have nothing to prove. We assume that SP is a proper subset of SQ , and that
t ∈ SQ but t 6∈ SP .
Consider the sequence {ti }∞i=0 with t0 = t, ti+1 = P (ti ) for every nonnegative integer i. Since t ∈ SQ ,
tk = Q(t0 ) = Q(t) = t = t0 .
Since the polynomial a − b divides the polynomial am − bm (where m is a nonnegative integer), it is
not difficult to see that the polynomial a − b divides the polynomial P (a) − P (b), where P (x) is a
polynomial with integer coefficients. In our current problem, then, we can conclude that the integer
sequence {ti }∞i=0 satisfies the following sequence of divisibility relations
(ti+1 − ti ) | (P (ti+1 ) − P (ti )) = ti+2 − ti+1
for every nonnegative integer i. Since tk+1 − tk = t1 − t0 = P (t) − t 6= 0, each term in the chain of
differences
t1 − t0 , t2 − t1 , . . . , tk − tk−1 , tk+1 − tk
is a nonzero divisor of the next one, and since tk+1 − tk = t1 − t0 , all these differences have equal
absolute values. Let ti = max{t0 , t1 , . . . , tk }. Then ti−1 − ti = −(ti − ti+1 ), or ti−1 = ti+1 . It is then
not difficult to see that ti+2 = ti for every i; that is,
t1 = P (t0 ) and t0 = P (t1 ) or P (P (t0 )) = t0 .
54
Therefore,
SQ = {t | t ∈ Z and P ((P (t)) = t}.
Without loss of generality, we may assume that t0 < t1 . If s0 is another element in SQ , let s1 = P (s0 ).
(It is possible that s0 ∈ SP ; that is, s1 = s0 .) We further assume without loss of generality that
s0 < s1 and t0 < s0 ; that is, t0 < s0 ≤ s1 and t0 < t1 . Note that s1 −t0 divides P (s1 )−P (t0 ) = s0 −t1 .
We must have t0 < s0 < s1 < t1 . Note that s0 − t1 also divides P (s0 ) − P (t1 ) = s1 − t0 , it follows
that s0 − t1 = −(s1 − t0 ); that is,
t0 + t1 = s0 + s1 = s0 + P (s0 ).
In other words, s0 is a root of the polynomial P (x) + x = t0 + t1 . Since P (x) + x has degree n, there
are at most n (integer) roots (including t0 ) of P (x) + x. Hence there are at most n elements in SQ ,
completing our proof.
6. Assign to each side b of a convex polygon P the maximum area of a triangle that has b as a side and
is contained in P. Show that the sum of the areas assigned to the sides of P is at least twice the area
of P.
Solution. Define the weight of a side XY to be the area assigned to it, and define an antipoint of
a side of a polygon to be one of the points in the polygon farthest from that side (and consequently
forming the triangle with greatest area).
Lemma. For any side XY , Z is an antipoint if and only if the line ` through Z parallel to XY does
not go through the interior of the polygon. (Note that this means we can assume Z is a vertex, as
we shall do henceforth).
Proof. Clearly, if Z is an antipoint, ` must not go through the interior of the polygon. Now if `
does not go through the interior of the polygon, assume there is a point Z 0 farther away from XY
than Z. Since the polygon is convex, the point XZ 0 ∩ ` is in the interior of the polygon, which is a
contradiction.
Suppose for the sake of contradiction that the sum of the weights of the sides is less than twice the
area of some polygon. Then let S be the non-empty set of all convex polygons for which the sum
of the weights is strictly less than twice the area. It is easy to check that no polygon in S can be a
triangle, so we may assume all polygons in S have at least 4 sides.
We first prove by contradiction that there is some polygon in S such that all of its sides are parallel
to some other side. Suppose the contrary; then consider one of the polygons in S which has the
minimal number of sides not parallel to any other side (this exists by the well-ordering principle).
Call this polygon P = A1 A2 · · · An , and without loss of generality let An A1 be a side which is not
parallel to any other side of P .
Then let Ai be the unique antipoint of An A1 , and let Au and Av be respective antipoints of Ai−1 Ai
and Ai Ai+1 . Define X to be the point such that Au X||Ai−1 Ai , Av X||Ai Ai+1 .
Now consider the set T ⊂ P of points that are strictly on the same side of Au Av as An A1 . First of
all, for any side in T , Ai must be its antipoint, since the line through Ai parallel to Aj Aj+1 does not
go through the interior of P . Similarly, any vertex in T is not the antipoint of any side.
55
We now look at the polygon P 0 = Av Av+1 · · · Au−1 Au X. It is clear that P 0 has fewer sides which are
not parallel to any other side than P . Using [ · ] to denote area, we have
[P 0 ] − [P ] = [A1 A2 · · · Av−1 Av XAu Au+1 · · · An ].
The weight of the side Aj Aj+1 is the same in both P 0 and P for v ≤ j < u, but for P 0 , the sum of
the weights of the remaining two sides is [XAu Ai Av ], as Ai is an antipoint of both Au X and Av X.
Meanwhile, the sum of the weights of remaining sides for P is [A1 A2 · · · Av−1 Av Ai Au Au+1 · · · An ].
Hence the difference in the sums of weights of P 0 and P is
[XAu Ai Av ] − [A1 A2 · · · Av−1 Av Ai Au Au+1 · · · An ] = [A1 A2 · · · Av−1 Av XAu Au+1 · · · An ],
the same as the difference in area (and both differences were positive). Therefore, if the sum of
weights of P was less than 2[P ], then certainly the sum of weights of P 0 must be less than 2[P 0 ], so
that P 0 ∈ S. However, this contradicts the minimality of the number of non-parallel sides in P , so
there exists a polygon in S with opposite sides parallel.
Now, we will let R be the non-empty set of all polygons in S with all sides parallel to the opposite
side. Note that all polygons in R must have an even number of sides. We will show that there is a
parallelogram in R.
Suppose not, and that Q = B1 B2 · · · B2m is one of the polygons in R with the minimal number
of sides, and m ≥ 3. Let X = B1 B2 ∩ B2m−1 B2m and Y = Bm−1 Bm ∩ Bm+2 Bm+1 . Set Q0 =
XB2 B3 · · · Bm−1 Y Bm+2 · · · B2m . We propose that the increase in the sum of weights going from Q
to Q0 is at most twice the increase in area, so that Q0 ∈ R.
To aid us, we will let hX and hY be the respective distances of X and Y from B2m B1 and Bm Bm+1 .
The increase in weight is
[XBm+1 B1 ] + [XB2m Bm ] + [Y Bm B2m ] + [Y Bm+1 B1 ] − [B1 B2m Bm ] − [B2m Bm Bm+1 ]

= [XB1 Y ] + [XB2m Y ] − [B1 B2m Bm ] + [Y Bm X] + [Y Bm+1 X] − [B2m Bm Bm+1 ]
hY · B1 B2m hX · Bm Bm+1
= [XB1 B2m ] + + [Y Bm Bm+1 ] +
2 2
while the increase in area is [XB1 B2m ] + [Y Bm Bm+1 ]. It remains to show that the first expression
is at most twice the second, or in other words, to show that
hY · B1 B2m hX · Bm Bm+1 hX · B1 B2m hY · Bm Bm+1
+ ≤ [XB1 B2m ] + [Y Bm Bm+1 ] = + ,
2 2 2 2
which is equivalent to
(hX − hY ) (B1 B2m − Bm Bm+1 ) ≥ 0
Noting that triangles B1 B2m X and Bm+1 Bm Y are similar, we have hX /hY = B1 B2m /Bm Bm+1 , so
the above inequality holds.
With the inequality proven, we now know that Q0 ∈ R, and yet Q0 has fewer sides than Q. This
contradicts the minimality of the number of sides of Q, so there exists a parallelogram in R. However,
the sum of the weights of a parallelogram clearly equals twice its area, so this contradicts the entire
existence of S, as desired.
56
6 IMO 2007
48th International Mathematical Olympiad
Hanoi, Vietnam
July 25 and 26, 2007
6.1 IMO 2007 Problems

1. Real numbers a1 , a2 , . . . , an are given. For each i (1 ≤ i ≤ n) define
di = max{aj : 1 ≤ j ≤ i} − min{ai : i ≤ j ≤ n}
and let
d = max{di : 1 ≤ i ≤ n}.
(a) Prove that, for arbitrary real numbers x1 ≤ x2 ≤ · · · ≤ xn ,
d
max{|xi − ai | : 1 ≤ i ≤ n} ≥ . (∗)
2
(b) Show that there are real numbers x1 ≤ x2 ≤ · · · ≤ xn such that equality holds in (∗).
(This problem was suggested by New Zealand.)
2. Consider five points A, B, C, D and E such that ABCD is a parallelogram and BCED is a cyclic
quadrilateral. Let ` be a line passing through A. Suppose that ` intersects the interior of segment
DC at F and intersects line BC at G. Suppose also that EF = EG = EC. Prove that ` is the
bisector of angle DAB.
(This problem was suggested by Luxembourg.)
3. In a mathematical competition some competitors are friends. Friendship is always mutual. Call a
group of competitors a clique if each two of them are friends. (In particular, any group of fewer than
two competitors is a clique.) The number of members in a clique is called its size.
Given that, in this competition, the largest size of a clique is even, prove that the competitors can
be arranged in two rooms such that the largest size of a clique contained in one room is the same as
the largest size of a clique contained in the other room.
(This problem was suggested by Russia.)
4. In triangle ABC the bisector of angle BCA intersects the circumcircle again at R, the perpendicular
bisector of side BC at P , and the perpendicular bisector of side CA at points Q. The midpoints of
BC and CA are K and L, respectively. Prove that triangles RP K and RQL have the same area.
(This problem was suggested by the Czech Republic.)
57
5. Let a and b be positive integers. Show that if 4ab − 1 divides (4a2 − 1)2 , then a = b.
(This problem was suggested by the United Kingdom.)
6. Let n be a positive integer. Consider
S = {(x, y, z) : x, y, z ∈ {0, 1, . . . , n}, x + y + z > 0} .
as a set of (n + 1)3 − 1 points in three-dimensional space. Determine the smallest possible number
of planes, the union of which contains S but does not include (0, 0, 0).
(This problem was suggested by the Netherlands.)
58
6.2 IMO 2007 Solutions
1. Real numbers a1 , a2 , . . . , an are given. For each i (1 ≤ i ≤ n) define
di = max{aj : 1 ≤ j ≤ i} − min{ai : i ≤ j ≤ n}
and let
d = max{di : 1 ≤ i ≤ n}.
(a) Prove that, for arbitrary real numbers x1 ≤ x2 ≤ · · · ≤ xn ,

d
max{|xi − ai | : 1 ≤ i ≤ n} ≥ . (∗)
2
(b) Show that there are real numbers x1 ≤ x2 ≤ · · · ≤ xn such that equality holds in (∗).
Solution 1. We handle each part separately.
(a) Let 1 ≤ p ≤ q ≤ r ≤ n be indices for which
d = dq , ap = max{aj : 1 ≤ j ≤ q}, ar = min{aj : q ≤ j ≤ n}
and so d = ap − ar . (These indices are not necessarily unique.)

We consider |xp − ap | and |xr − ar |. Since xp ≤ xr , we conclude that
(ap − xp ) + (xr − ar ) = (ap − ar ) + (xr − xp ) ≥ ap − ar = d.
implying that at least one of the two summands ap − xp and xr − ar is greater than or equal to
d
2 . Thus
d
max{|xi − ai | : 1 ≤ i ≤ n} ≥ max{|xp − ap |, |xr − ar |} ≥ max{ap − xp , xr − ar } ≥ .
2
(b) Define the sequence {xk }nk=1 as
½ ¾
d d
x1 = a1 − , xk = max xk−1 , ak − for 2 ≤ k ≤ n.
2 2
We show that we have equality in (∗) for this sequence.
By the definition, sequence {xk }nk=1 is non-decreasing and
d
x k − ak ≥ − for all 1 ≤ k ≤ n. (†)
2
Next we prove that
d
xk − ak ≤ for all 1 ≤ k ≤ n. (‡)
2
Consider an arbitrary index 1 ≤ k ≤ n. Let ` ≤ k be the smallest index such that xk = x` . We
have either ` = 1 or ` ≥ 2 and x` > x`−1 . In both cases, we have
d
xk = x` = a` − .
2
We also note that
a` − ak ≤ max{aj : 1 ≤ j ≤ k} − min{aj : k ≤ j ≤ n} = dk ≤ d.
59
Combining the last two relations yields
d d d
xk − ak = a` − ak − ≤d− = ,
2 2 2
which is (‡). Putting the relations (∗) and (∗∗) together gives
d d d
− ≤ xk − ak ≤ or |xk − ak | ≤ for every k with 1 ≤ k ≤ n.
2 2 2
Thus
d
max {|xi − ai | : 1 ≤ i ≤ n} ≤ .
2
We have equality because |x1 − a1 | = d2 .
Solution 2. We present another construction of a sequence {xk }nk=1 for part (b).
For each 1 ≤ i ≤ n, let
Mi = max{aj : 1 ≤ j ≤ i} and mi = min{aj : i ≤ j ≤ n}.
For all 1 ≤ i < n, we have
Mi = max{a1 , . . . , ai } ≤ max{a1 , . . . , ai , ai+1 } = Mi+1
and
mi = min{ai , ai+1 , . . . , an } ≤ min{ai+1 , . . . , an } = mi+1 .
Therefore sequences {Mi } and {mi } are non-decreasing. Moreover, since ai is listed in both defini-
tions,
mi ≤ ai ≤ Mi .
To achieve equality in (∗), we define for every 1 ≤ i ≤ n,
Mi + mi
xi = .
2
Since sequences {Mi } and {mi } are non-decreasing, this sequence {xi } is non-decreasing as well. By
definition, di = Mi − mi . Consequently, we obtain that
di mi − Mi Mi − mi di
− = = xi − Mi ≤ xi − ai ≤ xi − mi = = .
2 2 2 2
Therefore ½ ¾
di d
max {|xi − ai | : 1 ≤ i ≤ n} ≤ max :1≤i≤n = .
2 2
Since the opposite inequality has been proved in part (a), we must have equality.
Remark: During the jury meeting, the leader from the Netherlands suggested the following simpler
definition of d:
d = max{ai − aj : 1 ≤ i ≤ j ≤ n}.
Since this is an entry level problem, jury decided to keep the more detailed definition of d.
60
2. Consider five points A, B, C, D and E such that ABCD is a parallelogram and BCED is a cyclic
quadrilateral. Let ` be a line passing through A. Suppose that ` intersects the interior of segment
DC at F and intersects line BC at G. Suppose also that EF = EG = EC. Prove that ` is the
bisector of angle DAB.
Remark: There are two different configurations (depending on the size of ∠DAB) as shown below.
Solution 1. We consider three cases.
• In this case, we assume that CF = CG. Thus, ∠F GC = ∠GF C, hence ∠GAB = ∠GF C =
∠F GC = ∠F AD, and ` bisects ∠DAB.
• In this case, we assume that CF < GC. Let EK and EL be the altitudes in the isosceles
triangles ECF and EGC, respectively. Then in the right triangles EKF and ELC we have
EF = EC and
CF GC
KF = < = LC,
2 2
so p p
KE = EF 2 − KF 2 > EC 2 − LC 2 = LE.
Since quadrilateral BCED is cyclic, we have ∠EDC = ∠EBC, so the right triangles BEL and
DEK are similar. Then KE > LE implies DK > BL, and hence
DF = DK − KF > BL − LC = BC = AD.
But triangles ADF and GCF are similar, so we have

AD GC
1> = ,
DF CF
which contradicts our assumption. Thus, our assumption was wrong.
• In this case, we assume that CF > GC. Proceed as in the previous case, we obtain the converse
inequalities KF > LC, KE < LE, DK < BL, DF < AD. It follows that
AD GC
1< = ,
DF CF
which contradicts our assumption. Thus, our assumption was wrong.
Combining the above, we conclude that CF = CG and ` bisects ∠DAB.
B C L G B C L
G
E K
K
F E
F
A D
A D
61
Solution 2. (We maintain the notations of the first solution.) Assume that CE = 1, ∠BAF = α,
and ∠F AD = β. Since α and β are in the range (0◦ , 180◦ ), α = β if and only if tan α = tan β.
In triangle CF G, ∠CF G = α and ∠F GC = β. Since E is the circumcenter of triangle CF G, it is
well known that
CK = KF = sin β, KE = cos β; and CL = LG = sin α, LE = cos α.
Since triangles ADF and GCF are similar, we may assume that AD = rCG and F D = rF C or
AD = 2r sin α and F D = 2r sin β. Hence
DK = DF + KF = (2r + 1) sin β and BL = BC + CL = AD + CL = (2r + 1) sin α.
Since BCED is cyclic, ∠LBE = ∠CBE = ∠CDE = ∠KDE or tan ∠LBE = tan ∠KDE; that is,
cos α LE KE cos β
= = = .
(2r + 1) sin α BL KD (2r + 1) sin β
Consequently, tan α = tan β or α = β; that is, ` bisects angle DAB.
Solution 3. This is a synthetic write-up of the second solution. Since AD = BC and triangles CGF
and DAF are similar (with K and L as corresponding midpoints), we have
BC AD CL CL BC BC + CL BL
= = or = = = .
DF DF FK FK DF DF + F K DK
On the other hand, we know that triangles BLE and DKE are similar, implying that
BL LE
= .
DK KE
Combing the last two equations yields
CL LE
= .
FK KE
Note that CL2 + LE 2 = CE 2 = F E 2 = F K 2 + KE 2 . Thus CL = F K and CG = CF . It is then
not difficult to see that AF bisects ∠DAB.
Solution 4. (Based on work by Sherry Gong) Let M be the foot of the perpendicular from E to line
BD. Then E lies on the circumcircle of triangle BCD and K, L, M are the feet of the perpendiculars
from E to the sides of the triangle; that is, K, L, M are collinear (the Simson line).
B C L G
E
K
M
F
A D
62
Since KL is the midline of triangle CF G, KL k F G or M L k AG, implying that M L intersects
the AC in the midpoint of AC, which is also the midpoint of BD, so that M is the center of the
parallelogram ABCD. Because M is a the midpoint of segment BD, in triangle BED, EM is both
d = ED.
altitude and median; that is, BED is isosceles with BE = ED. Hence BE d Note that
d + CE
BC d d
BE d
ED
∠GCE = = = = ∠ECD,
2 2 2
from which ∠BAG = ∠GAD easily follows.
3. In a mathematical competition some competitors are friends. Friendship is always mutual. Call a
group of competitors a clique if each two of them are friends. (In particular, any group of fewer than
two competitors is a clique.) The number of members in a clique is called its size.
Given that, in this competition, the largest size of a clique is even, prove that the competitors can
be arranged in two rooms such that the largest size of a clique contained in one room is the same as
the largest size of a clique contained in the other room.
Remark: There were only two complete solutions submitted by the contestants. Attempts us-
ing induction or knowledge from graph theory all failed. Successful solutions required very careful
observation of subtleties.
Thanks to Kos Geza and IIya Bogdanov (members of the IMO 2007 Problem Selection Committee)
for the figures used in the first solution.
Solution 1. We present an algorithm to arrange the competitors. Let the two rooms be Room A
and Room B. At any state of the algorithm, A and B denote the sets of the competitors in the
rooms, and c(A) and c(B) denote the largest sizes of cliques in the rooms, respectively. Our goal is
to reach a stage with c(A) = c(B), and we call such a stage a balanced stage.
We start with an initial arrangement, and then we modify it in several steps (with a maximum of 4
steps) to reach a balanced stage. Within in each step, we modify our arrangement several times, if
necessary, by sending one person at a time to the other room.
• Step 1 In this step, we define our initial arrangement.

Let M be one of the cliques of largest size, |M | = 2m. Send all members of M to Room A and
all other competitors to Room B.
Since M is a clique of the largest size, we have c(A) = |M | ≥ (B). If c(A) = c(B), we are done!
If c(A) > c(B), we move to the next step.
• Step 2 In this step, we modify the arrangement so that c(A) ≤ c(B).
We achieve our goal with the following modifications: (1) Check the values c(A) and c(B);
(2) If c(A) > c(B), send one person from Room A to Room B; (3) Repeat (1) and (2) until
c(A) > c(B) is violated.
63
Room A Room B
A∩M B∩M
Note that c(A) > c(B) implies that Room A is not empty. Thus (2) is possible. On the
other hand, in each step, c(A) decreases by one and c(B) increases by at most one. Thus this
modification process will terminate.
Furthermore, at the end of this step, we reach an arrangement with
c(A) ≤ c(B) ≤ c(A) + 1, (†)
implying that
c(A) = |A| ≥ m. (‡)
(Indeed, if c(A) < m or c(A) ≤ m − 1, then we would have at least m + 1 members of M in
Room B and at most m − 1 in Room A, implying c(B) − c(A) ≥ (m + 1) − (m − 1) = 2, violating
(†).)
We consider our current arrangement. Let k = c(A).
If c(B) = k, then this is a desired arrangement, and we are done!
Otherwise, we must have c(B) 6= k. Then from (†) and (‡), we may assume that
c(B) = k + 1, k = |A| = |A ∩ M | ≥ m, |B ∩ M | ≤ m. (∗)
We proceed to the next step.
• Step 3 In this step, we check the members of the largest clique in B.
If there exists a competitor x ∈ B ∩ M and a clique C ⊂ B such that |C| = k + 1 and x ∈ / C,
then move x to Room A (and stop). This is a balanced arrangement, and we are done! Indeed,
after moving x back to Room A, we will have k +1 members of M in Room A, thus c(A) = k +1.
Due to x ∈/ C, c(B) = |C| is not decreased, and after this step we have c(A) = c(B) = k + 1.
Room A Room B
A∩M B∩M
x C
Otherwise, we assume that there is no such competitor x, and then in Room B, all cliques of
size k + 1 contain B ∩ M as a subset. We proceed to our final step.
• Step 4 In this step, we complete our modification process.
We achieve our goal as the following: (a) Check the value of c(B); (b) if c(B) = k + 1, choose a
clique C ⊂ B such that |C| = k + 1 and move one member of C \ M to Room A; (c) repeat (a)
and (b) until c(B) ≤ k.
64
Room A Room B
A∩M C B∩M
Note that |C| = k + 1 > m ≥ |B ∩ M | (by (∗)), so C \ M cannot be empty. Hence by Step
3, we are allowed to proceed with (a) and (b). Since there are only finitely many cliques in B
with size k + 1, this process terminates. Furthermore, every time we move a single person from
Room B to Room A, the value of c(B) decreases by at most 1. Hence, at the end of this process
we have c(B) = k. We claim that this is indeed a balanced arrangement, and we are done.
To finish our proof, we prove our claim. In Room A we have the clique A ∩ M with size |A ∩ M | = k
thus c(A) ≥ k. Let Q ⊂ A be an arbitrary clique. It suffices to show that |Q| ≤ k.
Room A Room B
A∩M
B∩M
In set Q, there can be two types of competitors:

– members of M . Since M is a clique, they are friends with all members of B ∩ M .
– competitors which were just moved to Room A in this step. Each of them has been in a clique
with B ∩ M so they are also friends with all members of B ∩ M .
Hence, all members of Q are friends with all members of B ∩ M . Sets Q and B ∩ M are cliques
themselves, so Q ∪ (B ∩ M ) is also a clique. Since M is a clique of the largest size,
|M | ≥ |Q ∪ (B ∩ M )| = |Q| + |B ∩ M | = |Q| + |M | − |A ∩ M |,
therefore
|Q| ≤ |A ∩ M | = k.
completing our proof.
Solution 2. (Based on work by Alex Zhai) As before, let M be a maximal size clique. We will
describe an algorithm which yields the desired result.
First, put all the people in M into room A, and put all others in room B. A starts out with the
larger maximum clique. One by one, move students from A to B until it is no longer possible to
move an additional student such that the maximum clique in A is at least that of B.
65
This means that if k + 1 students remain in A, and the maximum clique size of B is either k + 1, in
which case we are done, or it is k.
Let T1 be the set of members of S in A at this point, and let T2 be the set of members of S in B at
this point. For any x ∈ T1 , x must form a k + 1-clique with k students in B, or else we may simply
move x to B to leave A and B both with maximal cliques of size k.
Furthermore, call a set V of k students in B an associate of x whenever x ∪ V is a k + 1-clique. For
any associate V , if there exists t ∈ T2 such that t 6∈ V , we may move t to room A and x to room B,
and both rooms will have a maximum clique of size k + 1. Hence, assume T2 ⊂ V , for every choice
of V .
Then for some x ∈ T1 , consider the sets Li = Vi \ T2 , where Vi are all of x’s associates. They are
non-empty, since
|Li | = |Vi | − |T2 | = k − (|S| − (k + 1)) = 2k + 1 − |S|,
and |S| is even.

S
Let L = Li . We move everyone in L from B to A, and notice that there is no k + 2 clique K in
T1 ∪ L, for if there were, then K ∪ T2 would be a clique of cardinality k + 2 + |S| − k − 1 > |S|.
In addition, every k + 1 clique in T1 ∪ L contains x, for x is connected to everyone in T1 ∪ L. Thus,
by moving x from A to B, we decreased the maximum clique of A from k + 1 to k. We also did not
increase the maximum clique of B past k, since for every associate Vi of x, some non-empty subset
(namely, Li ) has been removed.
Therefore, when this is all done, we have reduced the maximum clique of A to k, a clique which can
be formed by considering T1 \ x, which is still in A. We now move the students in A but not in T1 \ x
to B, one by one. Either at some point the maximum clique in B is also k, or we run out of students
not in T1 \ x.
But this is analogous to our original situation, but with one fewer person in A, and we repeat the
process until it terminates.
4. In triangle ABC the bisector of angle BCA intersects the circumcircle again at R, the perpendicular
bisector of side BC at P , and the perpendicular bisector of side CA at points Q. The midpoints of
BC and CA are K and L, respectively. Prove that triangles RP K and RQL have the same area.
Remark: Let [R] denote the area of region R. If AC = BC then triangle ABC is isosceles, triangles
RQL and RP K are symmetric about the bisector CR and the statement is trivial. If AC 6= BC then
it can be assumed without loss of generality that AC < BC. Denote the circumcenter of triangle
ABC by O. Then perpendicular bisectors LQ and KP meet at O.
Solution 1. Without loss of generality, we assume that the circumradius of triangle ABC is 1. Set
∠CAB = 2x, ∠ABC = 2y, and ∠BCA = 2z. It suffices to show [RQL] is a symmetric function in x
d and so ∠RAC = 2x + z. By the Extended law of sines,
and y. Note that R is the midpoint of AB,
we have
AC = 2 sin 2y, CR = 2 sin(2x + z).
Then
CL · CR · sin z CL2 tan z sin2 2y tan z
[CLR] = = sin 2y sin z sin(2x + z) and [CLQ] = = ,
2 2 2
66
implying that
sin 2y tan z
[RQL] = [CLR] − [CLQ] = · (2 cos z sin(2x + z) − sin 2y).
2
By the product-to-sum formulas, we have 2 sin(2x + z) cos z = sin(2x + 2z) + sin 2x = sin 2y + sin 2x,
by noting that 2x + 2y + 2z = 180◦ . We conclude that
sin 2x sin 2y tan z
[RQL] = ,
2
which is clearly a symmetric function in x and y, completing our proof.
C C
K L K
L O O
Q Q
P P
A B A B
R R
Solution 2. (By Zuming Feng) The right triangles CLQ and CKP have equal angles at vertex C,
so they are similar, ∠CP K = ∠CQL = ∠OQP and
QL CQ
= .
PK CP
Let ` be the perpendicular bisector of chord CR. Clearly, ` passes through the circumcenter O. Due
to the equal angles at P and Q, triangle OP Q is isosceles with OP = OQ. Then line ` is the axis of
symmetry in this triangle as well. Therefore, points P and Q lie symmetrically on line segment CR;
that is,
RP = CQ and RQ = CP.
Triangles RQL and RP K have equal angles at vertices Q and P , respectively. Then
1
[RQL] 2 · QL · QR · sin ∠RQL RQ QL CP CQ
= 1 = · = · = 1.
[RP K] 2 · PR · PK · sin ∠RP K RP P K CQ CP
Solution 3. (By Insuk Seo, observer with the Republic of Korea) delegation) Let the perpendicular
bisector of CR intersect rays CA and CB at U and V , respectively. (Thus, line U V is the line ` is
the previous solution.) Denote by W the intersection of lines KP and AR.
Note that CW RK is a rhombus, and that CW RK is a trapezoid with W R k CK. It follows
that [P RK] = [CP W ]. (Since [CW K] = [CRK] and [P RK] = [CRK] − [CP K] and [CP W ] =
[CW K] − [CP K].)
In rhombus CU RV , sides U R and U C are symmetric with respect to line U V . As we have shown
in the second solution, rays OP and OQ are symmetric with respect to line U V . We conclude that
U LOW is symmetric about line U V . It is then not difficult to see that triangles CP W and RQL are
symmetric about line U V , implying that [CP W ] = [RQL]. It follows that [P RK] = [CP W ] = [RQL].
67
C C
K L K
L O O
V Q
Q
U P P
A B A B
R R
Solution 4. (We maintain the notations in the previous solutions.) As shown in the second solution,
∠OP Q = ∠OQP = 90◦ − z, and triangle OP Q is isosceles with OP = OQ. Furthermore, we also
have ∠P OQ = 2z.
d and ∠ROA = ∠BOR = 2z. Consider the rotation R
Note again that R is the midpoint of AB
around point O by angle 2z with R(A) = R. Then R(R) = B and R(Q) = P ; that is, R sends RQA
to BP R, implying that they have the same area.
Triangles RQL and RQA have RQ as a common side, so the ratio between their areas is
[RQL] d(L, CR) CL 1
= = = .
[RQA] d(A, CR) CA 2
(Here d(X, Y Z) denotes the distance between point X and line Y Z). Likewise,
[RP K] CK 1
= = .
[RP B] CB 2
Our proof can be completed as
[RQA] [BP R]
[RQL] = = = [RP K].
2 2
Solution 5. (We maintain the notations of the previous solutions.) Note that the product CQ · QR
is a power of Q with respect to the circumcircle of triangle ABC. Let r denote the radius of the
circumcircle of triangle ABC. Then CQ · QR = r2 − QO2 . Likewise, CP · P R = r2 − P O2 . As we
have shown in the previous solutions, P O = QO. Hence
CQ · QR = CP · P R.
Note that LQ = CQ sin z and ∠CQL = 90◦ − z. Thus

QL · QR sin ∠CQL CQ · QR sin z cos z CQ · QR sin 2z
[RQL] = = = .
2 2 4
In exactly the same way, we can show that
CP · P R sin 2z
[RP K] = .
4
Combining the above relations, we easily see that [RQL] = [RP K].
68
5. Let a and b be positive integers. Show that if 4ab − 1 divides (4a2 − 1)2 , then a = b.
Solution 1. Call a pair (a, b) of positive integers bad if 4ab − 1 divides (4a2 − 1)2 but a 6= b. In order
to prove that bad pairs do not exist, we present two properties of them which provide an infinite
descent.
• Property (i): If (a, b) is a bad pair then (b, a) is also bad.

Indeed, we consider congruence modulo 4ab − 1. Since 1 = 12 ≡ (4ab)2 (mod 4ab − 1), we have
¡ ¢2
(4b2 − 1)2 ≡ 4b2 − (4ab)2 = 16b4 (4a2 − 1)2 ≡ 0 (mod 4ab − 1).
Hence, the number 4ab − 1 divides (4b2 − 1)2 as well.

• Property (ii): If (a, b) is a bad pair and a < b, then there exists a positive integer c < a such
that (a, c) is also bad.
Indeed, let
(4a2 − 1)2
r= .
4ab − 1
Then r is an integer. Considering congruence modulo 4a yields
r = −r · (−1) ≡ −r(4ab − 1) = −(4a2 − 1)2 ≡ −1 (mod 4a)
that is, r = 4ac − 1 for some positive integer c. Since a < b we obtain that
(4a2 − 1)2
4ac − 1 = < 4a2 − 1
4ab − 1
and therefore c < a. By the construction, the number r = 4ac − 1 is a divisor of (4a2 − 1)2 , so
(a, c) is a bad pair.
Now suppose that there exists at least one bad pair. Take a bad pair (a, b) such that 2a + b attains
its smallest possible value. If a < b then Property (ii) provides a bad pair (a, c) with c < b and thus
2a + c < 2a + b, violating the minimality assumption on (a, b). Otherwise, if b < a, Property (i)
yields that pair (b, a) is also bad while 2b + a < 2a + b, again violating the minimality assumption
on (a, b). In either case, we reach a contradiction. Hence our assumption was wrong and there are
no bad pairs.
Solution 2. Define bad pair as in the first solution. We present two new properties of bad pairs.
Assume that (a, b) is a bad pair.
• Property (iii): (a, b) is a bad pair if and only if 4ab − 1 divides (a − b)2 .
Considering modulo 4ab − 1, we observe that
(4a2 − 1)2 ≡ (4a2 − 4ab)2 ≡ 16a2 (a − b)2 (mod 4ab − 1).
Since gcd(4a, 4ab − 1) = 1, it follows that 4ab − 1 divides (4a2 − 1)2 if and only if 4ab − 1 divides
(a − b)2 (which is a symmetric expression in a and b). Note that this implies that (a, b) is a bad
pair if and only if (b, a) is bad.
69
• Property (iv): If (a, b) is a bad pair, then there exists a positive integer c 6= b such that (b, c) is
also bad.
By Property (iii), we can can write (a − b)2 = m(4ab − 1) for some positive integer m. We
consider the quadratic
f (x) = (x − b)2 − m(4bx − 1) = x2 − 2(2m + 1)bx + b2 + m.
Let x1 and x2 be zeros of f . Since f (a) = 0, we write x1 = a. Note that x1 x2 = b2 + m, which
is positive. Hence x2 is positive. Note also that x1 + x2 = 2(2m + 1), which is an integer. Thus
x2 = c is also an positive integer. Finally, we note that c 6= b since f (b) = −m(4b2 − 1) < 0.
Thus c and b are distinct positive integers such that (c − b)2 = m(4cb − 1). By Property (iii),
(c, b) is a bad pair.
By Property (iii), we may assume that a > b. We further assume that a is minimal. By properties
(iii) and (iv), we obtain another bad pair (b, c). Note that for the quadratic f with positive leading
coefficient, we have the following facts: f (a) = f (c) = 0, f (b) < 0, and a > b. We conclude that
c < b < a. For bad pair (b, c), we have b > c. But b < a violates the minimality assumption on a.
Hence our assumption was wrong and there is no bad pair.
Note: In general, the following statement is true:

Let a be a given integer greater than 1. Prove that for arbitrary positive integers x and y,
the number axy − 1 divides (ax2 − 1)2 if and only if x = y.
The original proposed version of the problem is the following:
Let k be a positive integer. Prove that the number (4k 2 − 1)2 has a positive divisor of the
form 8kn − 1 if and only if k is even.
The current version is the key lemma to solve the original version. Since a medium level number
theory problem was needed during the problem selection, the current version was suggested by Ana
Devic, the leader of Switzerland delegation, and was approved by the jury.
6. Let n be a positive integer. Consider

S = {(x, y, z) : x, y, z ∈ {0, 1, . . . , n}, x + y + z > 0} .
as a set of (n + 1)3 − 1 points in three-dimensional space. Determine the smallest possible number
of planes, the union of which contains S but does not include (0, 0, 0).
Remark: The answer is 3n.

It is easy to find 3n such planes. For example, planes x = i, y = i or z = i (i = 1, 2, . . . , n)
cover the set S but none of them contains the origin. Another such collection consists of all planes
x + y + z = k for k = 1, 2, . . . , 3n. (There are many other collections of 3n planes covering the set S
but not covering the origin. We encourage the interested reader to find some of the other collections.)
Consider the analogous problems in one and two dimensions. The problem in one dimension is to
determine the smallest number of points whose union contains the points with coordinates 1, 2, . . . , n,
but not the origin. Clearly the answer is n.
In two dimensions, we will prove the following statement by induction on a + b:
70
Claim. Integers a and b are given. Suppose the union of a collection of lines contains each of the
points (x, y), where x and y are integers satisfying 0 ≤ x ≤ a, 0 ≤ y ≤ b, and x + y > 0, but does
not contain the origin. Then there are at least a + b lines in the union.
Proof of the claim. The base case a + b = 1 is trivial.

Suppose a + b > 1, and a collection of lines is given. If one of the lines in the collection is x = a, then
the remaining lines cover the points (x, y) satisfying 0 ≤ x ≤ a − 1, 0 ≤ y ≤ b, and x + y > 0. By
the inductive hypothesis, at least a − 1 + b lines are needed to cover these points, so at least a + b
lines are required to cover the points in the original set. Similarly, if one of the lines is y = b, then
at least a + b lines are needed.
Assume that neither the line x = a nor the line y = b is in the collection. Since no line can contain
the origin, the lines x = 0 and y = 0 are also not in the collection. Now consider the 2a+2b−1 points
(0, i), (a, i), where 1 ≤ i ≤ b, and (i, 0), (i, b), where 1 ≤ i ≤ a. Each of these points is contained in
the union of the lines, but each line contains at most two of the points. Thus there are at least a + b
lines. This completes the proof.
In the above problem, a + b is the best possible bound, since the collection of lines x = 1, x =
2, . . . , x = a, y = 1, y = 2, . . . , y = b satisfies the conditions of the problem.
Now consider the general problem in three dimensions. As in two dimensions, there is a set of a+b+c
planes that contains all the points. We conjecture that a + b + c planes are actually necessary to
cover all the points. If c = 0, the result reduces to the two-dimensional problem above. However,
combinatorial arguments like those used above fail when abc > 0.
For the original problem, in which a = b = c = n, we present two proofs that 3n is the smallest
possible number.
Solution 1. We establish the following key lemma.
Lemma 1. Consider a nonzero polynomial P (x1 , . . . , xk ) in k variables. Suppose that P vanishes at

all points (x1 , . . . , xk ) such that x1 , . . . , xk ∈ {0, 1, . . . , n} and x1 + · · ·+ xk > 0, while P (0, 0, . . . , 0) 6=
0. Then deg P ≥ kn.
Proof. We induct on k. The base case k = 0 is clear since P 6= 0. We assume that the statement is
true for k = ` − 1 for some positive integer `. Now we consider the case k = `. Denote for clarity
y = xk = x` .
Let R(x1 , . . . , xk−1 , y) be the residue of P modulo Q(y) = y(y − 1) . . . (y − n). Polynomial Q(y)
vanishes at each y = 0, 1, . . . , n, hence P (x1 , . . . , xk−1 , y) = R(x1 , . . . , xk−1 , y) for all x1 , . . . , xk−1 , y ∈
{0, 1, . . . , n}. Therefore, R also satisfies the condition of the Lemma; moreover, degy R ≤ n. Clearly,
deg R ≤ deg P , so it suffices to prove that deg R ≥ nk.
Now, expand polynomial R in the powers of y:
R(x1 , . . . , xk−1 , y) = Rn (x1 , . . . , xk−1 )y n + Rn−1 (x1 , . . . , xk−1 )y n−1 + · · · + R0 (x1 , . . . , xk−1 ).
We show that the polynomial Rn (x1 , . . . , xk−1 ) satisfies the condition of the induction hypothesis.
Consider the polynomial T (y) = R(0, . . . , 0, y) of degree ≤ n. This polynomial has n roots y =
1, . . . , n; on the other hand, T (y) 6≡ 0 since T (0) 6= 0. Hence deg T = n, and its leading coefficient is
Rn (0, 0, . . . , 0) 6= 0. (For example, in the case k = 1 we obtain that coefficientRn is nonzero.)
71
Similarly, take any numbers a1 , . . . , ak−1 ∈ {0, 1, . . . , n} with a1 +· · ·+ak−1 > 0. Substituting xi = ai
into R(x1 , . . . , xk−1 , y), we get a polynomial in y which vanishes at all points y = 0, . . . , n and has
degree ≤ n. Therefore, this polynomial is null, hence Ri (a1 , . . . , ak−1 ) = 0 for all i = 0, 1, . . . , n. In
particular, Rn (a1 , . . . , ak−1 ) = 0.
Thus, the polynomial Rn (x1 , . . . , xk−1 ) satisfies the condition of the induction hypothesis. So, we
have deg Rn ≥ (k − 1)n and deg P ≥ deg R ≥ deg Rn + n ≥ kn.
Now we can finish the solution. Suppose that there are N planes covering all the points of S but not
containing the origin. Let their equations be ai x + bi y + ci z + di = 0. Consider the polynomial
N
Y
P (x, y, z) = (ai x + bi y + ci z + di ).
i=1
It has total degree N . This polynomial has the property that P (x0 , y0 , z0 ) = 0 for any (x0 , y0 , z0 ) ∈ S,
while P (0, 0, 0) 6= 0. Hence by Lemma 1 we get N = deg P ≥ 3n, as desired.
Solution 2. Suppose r planes are given. As in the first solution, let A1 , . . . , Ar be (nonzero) linear
functions over R3 such that the equations of the planes are Ai (x, y, z) = 0. Define the polynomial
P0 = A1 A2 · · · Ar , whose degree is r. If a lattice point (x, y, z) satisfies 0 ≤ x ≤ a, 0 ≤ y ≤ b,
0 ≤ z ≤ c, and x + y + z > 0, then P0 (x, y, z) = 0; however, P0 (0, 0, 0) 6= 0.
Polynomials P1 , . . . , Pa are defined recursively by Pi+1 (x, y, z) = Pi (x + 1, y, z) − Pi (x, y, z). By
induction, we see that Pi (x, y, z) = 0 if 0 ≤ x ≤ a − i, 0 ≤ y ≤ b, 0 ≤ z ≤ c, and x + y + z > 0, while
Pi (0, 0, 0) 6= 0. Furthermore, if Pi is a nonzero polynomial, then its degree is r − i.
Let Q0 = Pa , and construct polynomials Q1 , . . . , Qb as above. Specifically, define Qi+1 (x, y, z) =
Qi (x, y + 1, z) − Qi (x, y, z). Again, we have Qi (x, y, z) = 0 provided x = 0, 0 ≤ y ≤ b − i, 0 ≤ z ≤ c,
and x + y + z > 0, while Qi (0, 0, 0) 6= 0. If Qi is nonzero, then its degree is r − a − i.
Finally, let R0 = Qb , and define R1 , . . . , Rc by Ri+1 (x, y, z) = Ri (x, y, z + 1) − Ri (x, y, z). As above,
Ri (x, y, z) = 0 if x = y = 0, 0 ≤ z ≤ c − i, and x + y + z > 0. If Ri is nonzero then its degree is
r − a − b − i.
Consider the polynomial Rc (x, y, z). Its value at (0, 0, 0) is nonzero, so it is a nonzero polynomial.
Its degree, which must be nonnegative, is r − a − b − c. Therefore, we have r ≥ a + b + c, as desired.
Therefore, a + b + c planes are necessary, and it is possible to cover all the points with a + b + c
planes. In the original problem, a = b = c = n, so the answer is 3n.
Note that the solution above implicitly uses the method of finite differences, which allows us to write
polynomials in terms of binomial coefficients. The proof could be made more explicit by writing P0
as µ ¶µ ¶µ ¶
X x y z
cd1 d2 d3 .
d1 d2 d3
d1 +d2 +d3 ≤r
This method extends to allow us to solve the following more general problem.
Real numbers x0 , . . . , xa , y0 , . . . , yb , z0 , . . . , zc are given such that the xi are distinct, the yi are
distinct, and the zi are distinct. If a collection of planes covers all points (xi , yj , zk ) except (x0 , y0 , z0 ),
the collection contains at least a + b + c points.
To prove this we need a more general form of the method of finite differences used above.
72
Lemma 2. Suppose distinct reals t0 , . . . , tn are given. Then there exist weights w0 , . . . , wn , with
w0 6= 0, such that for each nonnegative integer i < n, we have w0 ti0 + · · · + wn tin = 0, while
w0 tn0 + · · · + wn tnn 6= 0.
Proof. The n + 1 vectors (1, tj , t2j , . . . , tn−1j ) are linearly dependent over Rn , so there exist weights
wj such that w0 ti0 + · · · + wn tin = 0 for each i < n. However, since the tj are distinct, the vectors
(1, tj , t2j , . . . , tnj ) are linearly independent over Rn+1 . (This follows from the fact that their determi-
nant, the Vandermonde determinant, is nonzero.) Therefore, we must have w0 tn0 + · · · + wn tnn 6= 0.
Also, the vectors (1, tj , . . . , tn−1
j ), for j = 1, . . . , n, are independent. Thus we cannot have w0 = 0.
Next we prove an essential result about the weights found above.
Lemma 3. Let reals t0 , . . . , tn be given, and weights w0 , . . . , wn be defined as above. Let P (t) be
any polynomial, and define Q(t) = w0 P (t + t0 ) + · · · + wn P (t + tn ). If the degree of P (t) is less than
n, then Q(t) = 0; otherwise, the degree of Q is n less than the degree of P .
s!
Proof. First we prove the result for P (t) = ts . The tr coefficient of Q is r!(s−r)! (w0 ts−r s−r
0 +· · ·+wn tn ).
This is zero if r > s − n and nonzero if r ≤ s − n, provided s ≥ n. If s < n, then all coefficients of Q
are zero.
Now suppose P is a general polynomial of degree s. Then the ts term of P is nonzero. Also, the ts−n
term of Q depends only on the ts term of P , so it is also nonzero. Thus the degree of Q is s − n.
Finally, we solve the general form of IMO Problem 6. Suppose there are r planes in the collection,
so there exists a polynomial P of degree r such that P (xi , yj , zk ) = 0 unless i = j = k = 0. Apply
Lemma 3 to P (regarded as a polynomial in its first variable) with t0 = x0 , . . . , ta = xa , to obtain
a polynomial P1 of degree r − a such that P1 (0, yj , zk ) = 0 unless j = k = 0. Apply Lemma 3
again (this time on the second variable) to produce a polynomial P2 of degree r − a − b such that
P2 (0, 0, zk ) = 0 unless k = 0. Apply Lemma 3 one more time (on the third variable) to produce a
polynomial P3 of degree r − a − b − c such that P3 (0, 0, 0) 6= 0. Since P3 is a nonzero polynomial, we
have r ≥ a + b + c. This completes the proof.
73
7 Appendix
7.1 2006 Olympiad Results

The top twelve students on the 2006 USAMO were (in alphabetical order):
Yakov Berchenko-Kogan Needham B. Broughton High School Raleigh, NC
Yi Han Phillips Exeter Academy Exeter, NH
Sherry Gong Phillips Exeter Academy Exeter, NH
Taehyeon Ko Phillips Exeter Academy Exeter, NH
Brian Lawrence Montgomery Blair High School Silver Spring, MD
Tedrick Leung North Hollywood High School North Hollywood, CA
Richard McCutchen Montgomery Blair High School Silver Spring, MD
Peng Shi Sir John A. MacDonald C.I. Toronto, ON
Yi Sun The Harker School San Jose, CA
Arnav Tripathy East Chapel Hill High School Chapel Hill, NC
Alex Zhai University Laboratory High School Urbana, IL
Yufei Zhao Don Mills Collegiate Institute Toronto, ON
Brian Lawrence was the winner of the Samuel Greitzer–Murray Klamkin award, given to the top scorer(s)
on the USAMO. Brian Lawrence, Alex Zhai, and Yufei Zhao placed first, second, and third, respectively.
They were awarded college scholarships of $20000, $15000, $10000, respectively, by the Akamai Foundation.
The Clay Mathematics Institute (CMI) award, for a solution of outstanding elegance, and carrying a $5000
cash prize, was presented to Brian Lawrence for his solution to USAMO Problem 5.
The USA team members were chosen according to their combined performance on the 35th annual USAMO,
and the Team Selection Test that took place at the Mathematical Olympiad Summer Program (MOSP),
held at the University of Nebraska-Lincoln, June 5 – July 1, 2005. Members of the USA team at the 2006
IMO (Ljubljana, Slovenia) were Zachary Abel, Zarathustra (Zeb) Brady, Taehyeon (Ryan) Ko, Yi Sun,
Arnav Tripathy, and Alex Zhai. Zuming Feng (Phillips Exeter Academy) and Alex Saltman (Stanford
University) served as team leader and deputy leader, respectively. The team was also accompanied by
Steven Dunbar (University of Nebraska-Lincoln), as the observer of the deputy leader.
There were 498 contestants from 90 countries in the 2006 IMO. Gold medals were awarded to students
scoring between 28 and 42 points, silver medals to students scoring between 19 and 27 points, and bronze
medals to students scoring between 15 and 18 points. There were 42 gold medalists, 89 silver medalists,
122 bronze medalists, and honorable mentions (awarded to non-medalists solving at least one problem
completely). There were 3 perfect papers (Iurie Boreico from the Republic of Moldova, Zhiyu Liu from
the People’s Republic of China, and Alexander Magazinov from the Russian Federation) on this difficult
exam, even though it had two relatively easy entry level problems (problems 1 and 4). Tripathy’s 30 tied
for 16th place overall. The team’s individual performances were as follows:
Abel SILVER Medallist Brady GOLD Medallist
Ko SILVER Medallist Sun SILVER Medallist
Tripathy GOLD Medallist Zhai SILVER Medallist
In terms of total score (out of a maximum of 252), the highest ranking of the 90 participating teams were
as follows:
China 214 Germany 157 Japan 146 Taiwan 136
Russia 174 USA 154 Iran 145 Poland 133
Korea 170 Romania 152 Moldova 140 Italy 132
74
7.2 2007 Olympiad Results
Sergei Bernstein Belmont High School Belmont, MA
Sherry Gong Phillips Exeter Academy Exeter, NH
Adam Hesterberg Garfield High School Seattle, WA
Eric Larson South Eugene High School Eugene, OR
Brian Lawrence Montgomery Blair High School Kensington, MD
Tedrick Leung North Hollywood High School Winnetka, CA
Haitao Mao Thomas Jefferson HS of Science and Tech Vienna, VA
Delong Meng Baton Rouge Magnet High School Baton Rouge, LA
Krishanu Sankar Horace Mann High School Hastings on Hudson, NY
Jacob Steinhardt Thomas Jefferson HS of Science and Tech Vienna, VA
Arnav Tripathy East Chapel Hill High School Chapel Hill, NC
Alex Zhai University laboratory High School Champaign, IL
Brian Lawrence was the winner of the Samuel Greitzer–Murray Klamkin award, given to the top scorer(s)
on the USAMO. Sherry Gong and Alex Zhai tied for second place. Brian Lawrence, Sherry Gong, and Alex
Zhai were awarded college scholarships of $20000, $15000, $15000, respectively, by the Akamai Foundation.
The Clay Mathematics Institute (CMI) award, for a solution of outstanding elegance, and carrying a $5000
cash prize, was presented to Andrew Geng for his solution to USAMO Problem 4, presented as the second
solution here.
The USA team members were chosen according to their combined performance on the 36th annual USAMO
and the Team Selection Test held in Washington, D.C. on May 22 and 23, 2007. Members of the USA team
at the 2007 IMO (Hanoi, Vietnam) were Sherry Gong, Eric Larson, Brian Lawrence, Tedrick Leung, Arnav
Tripathy, and Alex Zhai. Zuming Feng (Phillips Exeter Academy) and Ian Le served as team leader and
deputy leader, respectively. The team was also accompanied by Steven Dunbar (University of Nebraska-
Lincoln), as the observer of the deputy leader. The Mathematical Olympiad Summer Program (MOSP)
was held at the University of Nebraska-Lincoln, June 10 – June 30, 2007.
There were 520 contestants from 93 countries and regions in the 2007 IMO. This was a very challenging
contest, in particular for stronger students, because the last problem of each session (problems 3 and
6) was very difficult. (There only 7 complete solutions submitted to these two problems.) The top three
contestants were Konstantin Matveev from Russia (37 points), Peter Scholze from Germany, and Caili Shen
from China (36 points). Gold medals were awarded to students scoring between 29 and 37 points, silver
medals to students scoring between 21 and 28 points, and bronze medals to students scoring between 14
and 20 points. There were 39 gold medalists, 83 silver medalists, 131 bronze medalists, and 149 honorable
mentions (awarded to non-medalists solving at least one problem completely). Gong and Zhai’s 32 tied for
7th place overall. The team’s individual performances were as follows:
Gong GOLD Medallist Larson SILVER Medallist
Lawrence SILVER Medallist Leung BRONZE Medallist
Tripathy SILVER Medallist Zhai GOLD Medallist
In terms of total score (out of a maximum of 252), the highest ranking of the 94 participating teams were
as follows:
Russia 184 Korea (South) 168 Ukraine 154 Taiwan 149
China 181 USA 155 Korea (North) 151 Romania 146
Vietnam 168 Japan 154 Bulgaria 149 Iran and Hungary (tie) 143
75
7.3 2003-2007 Cumulative IMO Results
In terms of total scores (out of a maximum of 1260 points for the last five years), the highest ranking of
the participating IMO teams is as follows:
China 1061 Vietnam 810 Hungary 747 Turkey 622

Russia 942 Romania 808 Ukraine 741 Hong Kong (China) 621
USA 922 Japan 801 Germany 694 Moldova 616
Korea (South) 861 Taiwan (China) 779 United Kingdom 633 Poland 604
Bulgaria 859 Iran 779 Belarus 631 Canada 604
More and more countries now value the crucial role of meaningful problem solving in mathematics edu-
cation. The competition is getting tougher and tougher. A top ten finish is no longer a given for the
traditional powerhouses.
7.4 2008 USAMO and USA IMO Team Selection Results

David Benjamin William Henry Harrison High School West Lafayette, IN

TaoRan Chen Bayside High School Freshmeadows, NY
Paul Christiano The Harker School San Jose, CA
Sam Elder Poudre High School Fort Collins, CO
Shaunak Kishore Unionville-Chaddsford High School West Chester, PA
Delong Meng Baton Rouge Magnet High School Baton Rouge, LA
Evan O’Dorney Venture High School Berkeley, CA
Qinxuan Pan Thomas S Wootton High School Gaithersburg, MD
David Rolnick Home School Rupert, VT
Colin Sandon Essex High School Essex Junction, VT
Krishanu Sankar Horace Mann High School Hastings on Hudson, NY
Alex Zhai University laboratory High School Champaign, IL
Evan O’Dorney and Colin Sandon were the co-winners of the Samuel Greitzer–Murray Klamkin award,
given to the top scorer(s) on the USAMO. Krishanu Sankar and Qinxuan Pan tied for second place.
Shaunak Kishore and Delong Meng tied for third place. These students were awarded college scholarships
of $10000, $7500, $5000, respectively, by the Akamai Foundation. The Clay Mathematics Institute (CMI)
award, for a solution of outstanding elegance, and carrying a $5000 cash prize, was presented to Evan
O’Dorney for his solution to USAMO Problem 2, presented as the fifth solution here.
The USA team members were chosen according to their combined performance on the 37th annual USAMO
and the Team Selection Test held in Washington, D.C. on June 8, 9 and 13, 2008. Members of the USA team
at the 2007 IMO (Hanoi, Vietnam) were Paul Christiano, Shaunak Kishore, Evan O’Dorney, Colin Sandon,
Krishanu Sankar, and Alex Zhai. Zuming Feng (Phillips Exeter Academy) and Razvan Gelca (Texas Tech
University) served as team leader and deputy leader, respectively. The team was also accompanied by
Steven Dunbar (University of Nebraska-Lincoln), as the observer of the deputy leader. The Mathematical
Olympiad Summer Program (MOSP) was held at the University of Nebraska-Lincoln, June 10 – July 3,
2008.
For more information about the USAMO or the MOSP, contact Steven Dunbar at sdunbar@math.unl.edu.
76

Zumning Feng, Yi Sun - USA and International Mathematical Olympiads 2007-2008 - 76p PDF

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Zumning Feng, Yi Sun - USA and International Mathematical Olympiads 2007-2008 - 76p PDF

Transféré par

Droits d'auteur :

Formats disponibles

USA and International Mathematical

Zuming Feng Yi Sun

2 Team Selection Test 2007 13

4 Team Selection Test 2008 40

1.1 USAMO 2007 Problems

(This problem was suggested by Sam Vandervelde.)

(This problem was suggested by Gregory Galperin.)

(This problem was suggested by Andás Gyárfás.)

(This problem was suggested by Reid Barton.)

(This problem was suggested by Titu Andreescu.)

with equality if and only if triangle ABC is equilateral.

(This problem was suggested by Kiran Kedlaya and Sungyoon Kim.)

Solution 1. For k ≥ 1, let

Solution 2. For k ≥ 1, let

sm n m−1 m+1 m−1

Solution 3. For k ≥ 1, let

m − 1 = am+1 = am+2 = am+3 = am+4 = . . . .

(a + b)2 ≤ (a + r)2 + (b + r)2 + (a + r)(b + r),

Remark: The above √ argument

(i) |R̃| ≥ s and |B̃| ≥ s,

x7 + 1 (x + 1)7 − ((x + 1)7 − (x7 + 1))

= (x + 1)6 − 72m (x2 + x + 1)2

= {(x + 1)3 − 7m (x2 + x + 1)}{(x + 1)3 + 7m (x2 + x + 1)}

= 2x2 + 2x + 1 ≥ 113 > 1.

Hence (x7 + 1)/(x + 1) is composite and the proof is complete.

with equality if and only if triangle ABC is equilateral.

and S1 T1 = 2r. We have

x(y + z)2 xa2

Remark: Trigonometric solutions of (∗) are also possible.

2.1 TST 2007 Problems

(This problem was suggested by Zuming Feng and Zhonghao Ye.)

2. Let n be a positive integer and let a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn be two nondecreasing

(This problem was suggested by Kiran Kedlaya.)

(This problem was suggested by Thomas Mildorf.)

(This problem was suggested by Zuming Feng and Zhonghao Ye.)

(r(1), r(3), . . . , r(1023))

∠DQX1 + ∠DQZ = ∠X1 CD + ∠DQZ = ∠DP Z + ∠DQZ = 180◦ ;

that is, X1 , Q, Z are collinear.

∠XAQ = ∠CAQ = ∠CP Q = ∠DP Q = ∠DBQ = ∠XBQ,

establishing the claim.

Since AXQB and BP DQ are cyclic, we have

∠QSC = ∠ABQ = ∠P BQ = ∠CDQ,

implying that XDQC is cyclic.

2. Let n be a positive integer and let a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn be two nondecreasing

{ai − aj | 1 ≤ i < j ≤ n} and {bk − b` | 1 ≤ k < ` ≤ n}.

In particular, it gives us that

Solution 1. Put sn = a1 + · · · + an = b1 + · · · + bn . Then

By (∗), these two quantities are equal, so

Consequently, each of the inequalities a1 + · · · + ai ≤ b1 + · · · + bi for i = 1, . . . , n − 1 must be an

Solution 2. Expanding both sides of (∗∗) yields

Squaring both sides of the given equation a1 + · · · + an = b1 + · · · + an gives

From the above relations we easily deduce that

By the Cauchy-Schwartz inequality, we obtain that

We will prove the inequality

Proof. We proceed by induction on n. There is no harm in assuming that a1 ≤ · · · ≤ an . Let m

2 cos nθ cos θ = cos[(n + 1)θ] + cos[(n − 1)θ].

2(r1 r2 − r3 r4 )x = r12 + r22 − (r32 + r42 ).

Since r1 , r2 , r3 , r4 are rational and x is irrational, we must have r1 r2 − r3 r4 = 0 or

cos kθ cos[(k + 1)θ] = cos(k 2 θ) cos[(k 2 − 1)θ].

By the product-to-sum formulas, we derive

cos[(2k + 1)θ] − cos θ cos[(2k 2 − 1)θ] − cos θ

2 sin[(k − k 2 + 1)θ] sin[(k 2 + k)θ] = 0,