41st United States of America Mathematical Olympiad

Day I, II 12:30 PM – 5 PM EDT

April 24-25, 2012

USAMO 1. First we prove that any n ≥ 13 is a solution of the problem. Suppose that a1 , a2 , ..., an
satisfy max(a1 , a2 , ..., an ) ≤ n · min(a1 , a2 , ..., an ), and that we cannot find three that are
the side-lengths of an acute triangle. We may assume that a1 ≤ a2 ≤ ... ≤ an . Then
a2i+2 ≥ a2i + a2i+1 for all i ≤ n − 2. Let (Fn ) be the Fibonacci sequence, with F1 = F2 = 1
and Fn+1 = Fn + Fn−1 . It is easy to check that Fn < n2 for n ≤ 11, F12 = 122 and Fn > n2
for n > 12 (the last inequality follows by an immediate induction, while the first one can
be checked by hand). The inequality a2i+2 ≥ a2i + a2i+1 and the fact that a1 ≤ a2 ≤ ... ≤ an
imply that a2i ≥ Fi · a21 for all i ≤ n. Hence, if n ≥ 13, we obtain a2n > n2 · a21 , contradicting
the hypothesis. This shows that any n ≥ 13 is a solution of the problem.
√
By taking ai = Fi for 1 ≤ i ≤ n, we have max(a1 , a2 , ..., an ) ≤ n · min(a1 , a2 , ..., an ), for
any n < 13, but it is easy to see that no three ai ’s can be the side-lengths of an acute
triangle. Hence the answer to the problem is: all n ≥ 13.
This problem and solution were suggested by Titu Andreescu.

USAMO 2. Let R, G, B, Y denote the sets of Red, Green, Blue, Yellow points, respectively, and let
r, g, b, y denote a generic Red, Green, Blue, Yellow point, respectively. For integers 0 ≤
k ≤ 431, let Tk denote the counterclockwise rotation of 360k


432
degree around the center of
the circle. Furthermore, for a set S, let |S| denote the number of elements in S.
First, we claim that there is some index i1 such that |Ti1 (R) ∩ G| ≥ 28. Indeed, for each
k, set Tk (R) ∩ G consists of all Green points that are the images of Red points under the
rotation Tk . Hence the sum

s1 = |T0 (R) ∩ G| + |T1 (R) ∩ G| + · · · + |T431 (R) ∩ G|

is equal to the number of pairs of points (r, g) such that g = Tk (r) for some k. On the
other hand, for each r and each g, there is a unique rotation Tk with Tk (r) = g, form which
it follows that s1 = 1082 = 11664. Clearly, |T0 (R) ∩ G| = |R ∩ G| = 0 (because R ∩ G = ∅).
By the Pigeonhole principle, there is some index i1 such that
l s m  11664 
1
|Ti1 (R) ∩ G| ≥ = = d27.06 . . .e = 28,
431 431

establishing our claim. Let RG denote the set Ti1 (R) ∩ G, and let rg denote a generic
point in RG.
Second, we claim that there is some index i2 such that |Ti2 (RG) ∩ B| ≥ 8. Again, for each
k, set Tk (RG) ∩ B consists of all Blue points that are the images of the points in RG under
the rotation Tk . Hence the sum

s2 = |T0 (RG) ∩ B| + |T1 (RG) ∩ B| + · · · + |T431 (RG) ∩ B|

1
 is equal to the number of pairs of points (rg, b) such that b = Tk (rg) for some k. On the
other hand, for each rg and each b, there is a unique rotation Tk with Tk (rg) = b, form
which it follows that s2 ≥ 28 · 108 = 3024. Clearly, RG is a subset of B, which is disjoint
with B, so |T0 (RG) ∩ B| = 0. Furthermore, T432−i1 (Ti1 ) is the identity transformation,
implying that T432−i1 (Ti1 (R)) = R and T432−i1 (RG) is a subset of R which is disjoint with
B. In particular, |T432−i1 (RG) ∩ B| = 0. By the Pigeonhole principle, there is some index
i2 such that
l s m  3024 
2
|Ti2 (RG) ∩ B| ≥ ≥ = d7.0325 . . .e = 8,
430 430
establishing our claim. Let RGB denote the set Ti2 (RG) ∩ B, and let rgb denote a generic
point in RGB.
Third, we claim that there is some index i3 such that |Ti3 (RGB) ∩ Y | ≥ 3. We repeated
our previous process one more time. We note that

s3 = |T0 (RGB) ∩ Y | + |T1 (RGB) ∩ Y | + · · · + |T431 (RGB) ∩ Y | ≥ 8 · 108 = 864

and
|T0 (RGB) ∩ Y | = |T432−i2 (RGB) ∩ Y | = |T432−i2 −i1 (RGB) ∩ Y | = 0.
By the Pigeonhole principle, there is some index i3 such that
l s m  864 
3
|Ti3 (RGB) ∩ Y | ≥ ≥ = d2.01 . . .e = 3,
429 429

establishing our claim.

Let y1 , y2 , y3 be three points in Ti2 (RGB) ∩ Y . Then

(y1 , y2 , y3 ), (b1 , b2 , b3 ) = T432−i3 (y1 , y2 , y3 ), (g1 , g2 , g3 )

= T432−i3 −i2 (y1 , y2 , y3 ), (r1 , r2 , r3 )
= T432−i3 −i2 −i1 (y1 , y2 , y3 )

are twelve points that we are looking for.

This problem and solution were suggested by Gregory Galperin.

USAMO 3. We will show that the sequence exists for all n ≥ 3.

For n = 2, the sequence cannot exist: If it existed, we would have ak = −2a2k for all k,
from which a1 = (−2)r a2r for all r by induction. Then a1 would have to be divisible by 2r
for all r, which is impossible for a1 6= 0.
Now fix n ≥ 3. We will show that the desired sequence exists. The construction is
basically a repeated application of the Chinese Remainder Theorem, but the details require
substantial care.
First we prove two lemmas.
Lemma 1 It is possible to partition the positive integers into subsets S1 , S2 , S3 , . . . so that
for every positive integer k,
2
 (i) the numbers (n − 1)k and nk are in the same subset, and
(ii) the numbers k, 2k, . . . , (n − 2)k are all in strictly earlier subsets than (n − 1)k.

Proof To show this, define a function f from positive integers to positive reals as follows.
Let Pn be the set of primes dividing n. No element of Pn divides n − 1. For any number
k, write its prime factorization k = pe11 pe22 · · · perr , and then define
Y Y
f (k) = pei i · (pei i )logn (n−1) .
pi ∈P
/ n pi ∈Pn

Notice that for every positive integer k,

f ((n − 1)k) = (n − 1)f (k) = f (nk) (1)
whereas for each t = 1, 2, . . . , n − 2,
f (tk) ≤ tf (k) < f ((n − 1)k). (2)
Also notice that for each k, f (k) ≥ k logn (n−1) , which implies that for any fixed C, there can
only be finitely many values of k with f (k) < C. The latter fact means that the elements
of the image of f can be arranged in increasing order, x1 < x2 < x3 < · · · . Now just let
Si = f −1 (xi ) for each i. The sets Si are a partition of the positive integers, and (1) and
(2) ensure that they satisfy (i) and (ii) respectively.
Lemma 2 Let p, q be relatively prime positive integers and t1 , t2 , . . . , tr arbitrary integers.
Then it is possible to choose nonzero integers b1 , b2 , . . . , br+1 such that
pbi + qbi+1 = ti for i = 1, 2, . . . , r. (3)

Proof We first prove existence of a sequence of integers satisfying (3) for each i, by
induction on r. If r = 1, then since p, q are relatively prime, we can find c, d such
that pc + qd = 1. Then, b1 = ct1 and b2 = dt1 satisfy (3). Now suppose we have
b1 , . . . , br satisfying (3) for i = 1, 2, . . . , r − 1. If we choose any integer k, and replace
each bi with b0i = bi + (−1)i pi−1 q r−i k, then (3) still holds for i = 1, 2, . . . , r − 1, and
pb0r = pbr + (−1)r pr−1 k. Since p, q are relatively prime, we can choose k so as to make
pb0r congruent to tr modulo q, and then we take br+1 = (tr − pb0r )/q. Then the numbers
b01 , b02 , . . . , b0r , br+1 satisfy (3) for i = 1, 2, . . . , r.
This shows that we can find b1 , b2 , . . . , br+1 satisfying (3), but they may not all be nonzero.
However, once again, we can make the replacements b0i = bi + (−1)i pi−1 q r+1−i k for any
integer k, and the new sequence still satisfies (3). By an appropriate choice of k, we can
ensure each b0i is nonzero.
Now both lemmas are proven, and we resume the main proof. We will construct terms of
the sequence inductively, but not in the order a1 , a2 , . . ..
Suppose S is any set of positive integers, and we have chosen nonzero integers ak for each
k ∈ S. Say that there is a conflict in S if there exists some k such that k, 2k, . . . , nk are
all in S, and
ak + 2a2k + · · · + nank 6= 0.
Let S1 , S2 , . . . be as given by Lemma 1 We will inductively define our sequence as follows:
3
 (a) Step 1: Choose nonzero values ak for all k ∈ S1 simultaneously, without creating a
conflict in S1 .
(b) Step t ≥ 1: Given the values of ak for k ∈ S1 ∪ · · · ∪ St−1 chosen at previous steps,
choose nonzero integers ak for all k ∈ St simultaneously, without creating a conflict
in S1 ∪ · · · ∪ St .
If we can show that each step of this process can indeed be carried out, then it will
eventually define ak for all positive integers k, meeting the required condition

ak + 2a2k + · · · + nank = 0 (4)

for all k (since no conflicts are created).

For step 1, Lemma 1 implies we can choose ak arbitrarily for k ∈ S1 without creating
any conflicts, since (n − 1)k, nk ∈ / S1 for all k. Now for step t ≥ 1, suppose ak have
been assigned already for all k ∈ S1 ∪ S2 ∪ · · · ∪ St−1 . We need to assign ak for k ∈ St
to avoid creating any new conflicts. This just requires that the new assignments satisfy
(4) for all integers k such that (n − 1)k and nk are in St : for any other value k, either
{k, 2k, . . . , nk} 6⊆ S1 ∪ · · · ∪ St so no conflict can be created, or else Lemma 1 implies
{k, 2k, . . . , nk} ⊆ S1 ∪ · · · ∪ St−1 so that the corresponding constraint (4) has been dealt
with at an earlier step.
Thus for each k such that (n − 1)k, nk ∈ St , we have a constraint

(n − 1)a(n−1)k + nank = Xk , (5)

where Xk = −(ak +· · ·+(n−2)a(n−2)k ) is determined by the assignments made at previous

steps. We just need to show that it is possible to choose ak for all k ∈ St such that all
these constraints are satisfied.
Form a directed graph whose vertices are the elements of St , with an edge leading from
(n − 1)k to nk whenever both numbers are in St . Then every component of this graph is
either a single vertex or a (directed) path. We wish to show that nonzero integer values
can be assigned to elements of St so that for each edge, the corresponding constraint (5)
is satisfied. It suffices to show this for each component of the graph. If the component
is a single vertex, any nonzero value works. Otherwise, it is a path k1 , k2 , . . . , kr+1 , and
Lemma 2 ensures that we can choose nonzero integer values for ak1 , ak2 , . . . , akr+1 so as to
satisfy (5) for each edge.
This shows that each step of our constructive process can indeed be performed successfully,
and iterating eventually constructs every term of the sequence.
This problem and solution were suggested by Gabriel Carroll.
USAMO 4. There are three solutions: the constant functions 1, 2 and the identity function. Let us
prove that these are the only ones. Consider such a function f and suppose first of all
that there exists a > 2 such that f (a) = a. Then a!, (a!)!, ... are all fixed points of f . So
there is an increasing sequence (an )n≥0 of fixed points. If n is any positive integer, ak − n
divides ak − f (n) = f (ak ) − f (n) for all k, and so it also divides f (n) − n for all k. Thus
f (n) = n and since it holds for any n, we are done in this case.
4
 Now suppose that f has no fixed points greater than 2. Let p > 3 be a prime and observe
that (p − 2)! ≡ 1 (mod p) (by Wilson’s theorem), thus f (p − 2)! − f (1) = f ((p − 2)!) − f (1)
is a multiple of p. Clearly f (1) is 1 or 2. As p > 3, the fact that p divides f (p − 2)! − f (1)
implies that f (p − 2) < p. Since (p − 1)! − f (1) is not a multiple of p (again by Wilson), we
deduce that actually f (p − 2) ≤ p − 2. On the other hand, p − 3 divides f (p − 2) − f (1) ≤
f (p − 2) − 1. Thus either f (p − 2) = f (1) or f (p − 2) = p − 2. As p − 2 > 2, the last case
is excluded and so f (p − 2) = f (1) and this for all primes p > 3. Taking n any positive
integer, we deduce that p − 2 − n divides f (1) − f (n) and this holds for all large primes
p. Thus f (n) = f (1) and f is constant. The conclusion is now clear.
This problem and solution were suggested by Gabriel Dospinescu.

USAMO 5. Solution 1: The proof is split into two cases.

Case 1: P is on the circumcircle of ABC. Then P is the Miquel point of A0 , B 0 , C 0
with respect to ABC. Indeed, because ∠A0 B 0 C 0 = ∠CBA = ∠CP A = ∠A0 P C 0 , points
P , A0 , B 0 , C 0 are concyclic, and the same can be said for P , A, B 0 , C 0 and P , A0 , B 0 , C.
Hence ∠CA0 B 0 = ∠CP B 0 = ∠BP C 0 = ∠BA0 C 0 , so A0 B 0 C 0 are collinear.
Case 2: P is not on the circumcircle of ABC. Let Q be isogonal conjugate of P with
respect to ABC (which is not degenerate).
Claim. Let Q0 be the isogonal conjugate of P with respect to AB 0 C 0 . Then Q = Q0 .
Proof of the claim. Note that

∠BQC = ∠BAC + ∠CP B (because P and Q are isogonal conjugates in ABC)

0
= ∠C 0 AB 0 + ∠B 0 P C
= ∠C 0 Q0 B 0 (because P and Q are isogonal conjugates in AB 0 C 0 ).

Let X, Y , Z denote the reflections of P in sides BC, CA, AB, respectively, and let X 0
denote P ’s reflection in side B 0 C 0 of triangle AB 0 C 0 . Then ∠ZXY = ∠BQC (because QC
is orthogonal to XY and QB is orthogonal to XZ), whereas ∠ZX 0 Y 0 = ∠C 0 Q0 B 0 because
Q0 B 0 is orthogonal to X 0 Y and Q0 C 0 is orthogonal to X 0 Z and Q0 C 0 is orthogonal to X 0 Z,
so since ∠C 0 Q0 B 0 = ∠BQC, we get ∠ZXY = ∠ZX 0 Y . It follows that X, Y , Z, X 0 are
concyclic. The center of the XY Z-circle is Q while the center of the X 0 Y 0 Z-circle is Q0 .
Thus Q = Q0 .
Note. We have made use of the well-known fact that the circumcenter of the triangle
determined by the reflections of a point across the sidelines of another given triangle is
precisely the isogonal conjugate of the point with respect to that triangle. For a proof see
R. A. Johnson, Advanced Euclidean Geometry, 1929 ed., reprinted by Dover, 2007.
Similar arguments show that Q is also the isogonal point of P with respect to triangles
A0 BC 0 and A0 B 0 C. Therefore,

∠BC 0 A0 = ∠AC 0 A0 = ∠AC 0 P + ∠P C 0 Q + ∠QC 0 A0

= ∠QC 0 B 0 + ∠P C 0 Q + ∠BC 0 P
= ∠BC 0 B 0 = ∠AC 0 B 0 .
5
 This means that A0 , B 0 , C 0 are collinear. 
This problem and solution were suggested by Titu Andreescu and Cosmin Pohoata.
Solution 2: It’s easy to see (say, by law of sines) that

AC 0 AP sin ∠AP C 0 BA0 BP sin ∠BP A0 CB 0 CP sin ∠CP B 0

= , = , = .
BC 0 BP sin ∠BP C 0 CA0 CP sin ∠CP A0 AB 0 AP sin ∠AP B 0
The construction of A0 , B 0 , C 0 by reflections implies that

sin ∠AP C 0 = sin ∠CP A0 , sin ∠BP C 0 = sin ∠CP B 0 , sin ∠BP C 0 = sin ∠CP B 0 .

Hence,
AC 0 BA0 CB 0
· · = 1,
BC 0 CA0 AB 0
and the proof is complete by Menelaus’ theorem.
This second solution was suggested by Li Zhou, Polk State College, Winter Haven FL.

USAMO 6. This problem is a form of Chebyshev’s inequality for random variables. For each set
A ⊆ {1, 2, . . . , n}, define

X X n
X
∆A = 2SA = xi − xi = A (i)xi ,
i∈A i∈{1,2,...,n}\A i=1

where A (i) = 1 if i ∈ A and −1 otherwise. Squaring, we have

n
X X
∆2A = x2i + A (i)A (j)xi xj . (6)
i=1 i,j∈{1,...,n}
i6=j

Now sum the ∆2A ’s over all 2n possible choices of A. For each pair i 6= j, there are 2n−2
sets A with i, j ∈ A, and another 2n−2 sets with i, j ∈ / A; these sets each contributes a
term of +xi xj to the sum in (6). There are also 2n−2 sets A with i ∈ A, j ∈ / A, and 2n−2
sets with i ∈
/ A, j ∈ A. Each of these sets each contributes a term of −xi xj to (6). Hence,
xi xj appears 2n−1 times with a + sign and 2n−1 times with a − sign. Therefore all of these
terms cancel, and we find
X
∆2A = 2n (x21 + · · · + x2n ) = 2n . (7)
A⊆{1,2,...,n}

Now let λ > 0. There cannot be more than 2n−2 /λ2 terms ∆2A whose value greater than
or equal to 4λ2 . If this were not the case, then the sum of these terms would be greater
than 2n , so the sum in (7) would exceed 2n . Hence, there can be at most 2n−2 /λ2 sets A
such that |SA | ≥ λ. (Recall that ∆A = 2SA ). Moreover, these sets can be arranged into
complementary pairs because SA = −S{1,...,n}\A . In each of these pairs, exactly one of the
two members is positive. Therefore there are at most 2n−3 /λ2 sets A with SA ≥ λ.
6
For equality to hold, it must be the case that all positive values of ∆2A are equal to 4λ2 ;
otherwise we would again have a contradiction because the sum of all ∆2A would exceed
2n . In particular, all positive values of ∆2A must be the same. Thus all positive values of
xA must be the same. This will be the case only if at most one of the xi is positive and
at most one of the xi is negative. Because we must have at least one of each, there must
be exactly
√ one positive term and one√negative term. Thus it must be the case that one
xk = 2/2 for some k, one is xj = − 2/2 for some j√6= k, and all other xi = 0. Then the
assumption that every positive ∆2A = 4λ2 yields λ = 2/2.
Conversely, with the xi and λ as described, we have exactly
√ 2n−2 = 2n−3 /λ2 sets A such
√ xA ≥ λ (namely, those sets A that contain the 2/2 term and do not contain the
that
− 2/2 term.) Thus this is indeed the equality case.
This problem and solution were suggested by Gabriel Carroll.

Copyright © Mathematical Association of America

7