Solutions to the selected USAMO problems

The USA(J)MO Editorial Board

June 2020

The solutions to all four problems we mentioned are included in full below. In addition
to showing how to solve the problems, we think they are good templates to show how
we expect a correct solution to be written. (Actually, they are a little bit on the verbose
side, and during the competition you could get away with being more succinct. However,
if in doubt it is always better to include more details.)

Contents
1 Solution to 2012/1 2

2 Solution to 2014/4 3

3 Solution to 2004/3 4

4 Solution to 2013/5 5

1
§1 Solution to 2012/1
The answer is all n ≥ 13.
Define (Fn ) as the sequence of Fibonacci numbers, by F1 = F2 = 1 and Fn+1 =
Fn + Fn−1 . We will find that Fibonacci numbers show up naturally when we work
through the main proof, so we will isolate the following calculation now to make the
subsequent solution easier to read.
Lemma. For positive integers m, we have Fm ≤ m2 if and only if m ≤ 12.

Proof. A table of the first 14 Fibonacci numbers is given below.

F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 F13 F14
1 1 2 3 5 8 13 21 34 55 89 144 233 377
By examining the table, we see that Fm ≤ m2 is true for m = 1, 2, . . . 12,
and in fact F12 = 122 = 144. However, Fm > m2 for m = 13 and m = 14.
Now it remains to prove that Fm > m2 for m ≥ 15. The proof is by induction
with base cases m = 13 and m = 14 being checked already. For the inductive
step, if m ≥ 15 then we have
Fm = Fm−1 + Fm−2 > (m − 1)2 + (m − 2)2
= 2m2 − 6m + 5 = m2 + (m − 1)(m − 5) > m2
as desired.

We now proceed to the original problem. Denote by (†) the standing assumption that
max(a1 , a2 , . . . , an ) ≤ n · min(a1 , a2 , . . . , an ). The solution is divided into two parts.
Proof that all n ≥ 13 have the property. We first show now that every n ≥ 13 has
the desired property. Suppose for contradiction that no three numbers are the sides
of an acute triangle. Assume without loss of generality (by sorting the numbers) that
a1 ≤ a2 ≤ · · · ≤ an . Then since ai−1 , ai , ai+1 are not the sides of an acute triangle for
each i ≥ 2, we have that a2i+1 ≥ a2i + a2i−1 ; writing this out gives
a23 ≥ a22 + a21 ≥ 2a21
a24 ≥ a23 + a22 ≥ 2a21 + a21 = 3a21
a25 ≥ a24 + a23 ≥ 3a21 + 2a21 = 5a21
a26 ≥ a25 + a24 ≥ 5a21 + 3a21 = 8a21
and so on. The Fibonacci numbers appear naturally and by induction, we conclude that
a2i ≥ Fi a21 . In particular, a2n ≥ Fn a21 .
However, we know max(a1 , . . . , an ) = an and min(a1 , . . . , an ) = a1 , so (†) reads an ≤
n · a1 . Therefore we have Fn ≤ n2 , and so n ≤ 12, contradiction!
Proof that no n ≤ 12 have the property. Assume that n ≤ 12. √ The above calculation
also suggests a way to pick the counterexample: we √ choose a i = Fi for every i. Then
min(a1 , . . . , an ) = a1 = 1 and max(a1 , . . . , an ) = Fn , so (†) is true as long as n ≤ 12.
And indeed no three numbers form the sides of an acute triangle: if i < j < k, then
a2k = Fk = Fk−1 + Fk−2 ≥ Fj + Fi = a2j + a2i .

This problem was proposed by Titu Andreescu.

2
§2 Solution to 2014/4
The answer is k = 6. The solution is divided into two parts.
Proof that A cannot win if k = 6. We give a strat-
egy for B to prevent A’s victory. Shade in every
third cell, as shown in the right figure. Then A can
never cover two shaded cells simultaneously on her
turn. Now suppose B always removes a counter
on a shaded cell (and otherwise does whatever he
wants). Then he can prevent A from ever getting
six consecutive counters, because any six consecu-
tive cells contain two shaded cells.
Example of a winning strategy for A when k = 5.
We describe a winning strategy for A explicitly. Note that after B’s first turn there is
one counter, so then A may create an equilateral triangle, and hence after B’s second
turn there are two consecutive counters. Then, on her third turn, A places a pair of
counters two spaces away on the same line. Label the two inner cells x and y as shown
below.

x y

Now it is B’s turn to move; in order to avoid losing immediately, he must remove either
x or y. Then on any subsequent turn, A can replace x or y (whichever was removed) and
add one more adjacent counter. This continues until either x or y has all its neighbors
filled (we ask A to do so in such a way that she avoids filling in the two central cells
between x and y as long as possible).
So, let’s say without loss of generality (by symmetry) that x is completely surrounded
by tokens. Again, B must choose to remove x (or A wins on her next turn). After x is
removed by B, consider the following figure.

x y

We let A play in the two marked green cells. Then, regardless of what move B plays,
one of the two choices of moves marked in red lets A win. Thus, we have described a
winning strategy when k = 5 for A.

This problem was proposed by Palmer Mebane.

3
§3 Solution to 2004/3
Answer: the dissection is possible for every k > 0 except for k = 1.
Construction for k > 1. For every integer n ≥ 2 and real number r ≥ 1, we define a
shape R(n, r) as follows.
• We start with a rectangle of width 1 and height r. To its left, we glue a rectangle
of height r and width r2 to its left.
• Then, we glue a rectangle of width 1 + r2 and height r3 below our figure, followed
by a rectangle of height r + r3 and width r4 to the left of our figure.
• Next, we glue a rectangle of width 1 + r2 + r4 and height r5 below our figure,
followed by a rectangle of height r + r3 + r5 and width r6 to the left of our figure.
. . . and so on, until we have 2n pieces. The picture R(3, r) is shown below.
Observe that by construction, the entire
shape R(n, r) is a rectangle which consists of r6 r4 r2 1
r1
two similar “staircase” polygons, with similar-
ity ratio r. Note that R(n, r) is similar to a r3
1 × fn (r) rectangle where
1 + r2 + · · · + r2n
fn (r) = r5
r + r3 + · · · + r2n−1
is the aspect ratio of R(n, r).
Now, given some k > 1, choose n such that
1+ n1 < k. Note fn (1) = 1+ n1 but fn (k) > k. Since fn is continuous, by the intermediate
value theorem there exists some value of r > 1 with fn (r) = k, as needed.
Construction for k < 1. Reduces to the previous case since a
1 × k rectangle and a 1 × k −1 rectangle are similar.
Proof of impossibility for k = 1. Suppose we have a square
dissected into two similar polygons P ∼ Q. Let Γ be their
P Q
common boundary. By counting the number of sides of P and
Q we see Γ must run from one side of the square to an opposite Γ

side (possibly ending at a corner of the square). We orient the

figure so Γ runs from north to south, with P to the west and Q
to the east. Let s be the longest length of a segment in Γ.
Claim. The longest side length of P is max(s, 1). Similarly, the longest side
length of Q is max(s, 1) as well.

Proof. The only edges of P not in Γ are the west edge of our original square,
which has length 1, and the north/south edges of P (if any), which have
length at most 1. An identical argument works for Q.

It follows the longest sides of P and Q have the same length! Hence the two polygons
are in fact congruent, ending the proof.

This problem was proposed by Ricky Liu.

4
§4 Solution to 2013/5
Some optional motivation.1 To give an example, consider the case m = 4 and n = 23.
It happens to be true that
4 23
103 · = 97 + .
41 41
4 23
This equation implies that the repeating decimal representations of 41 and 41 will be
cyclic shifts of each other by 3 places. Indeed, a calculation gives that
1 4 23
= 0.02439, = 0.09756, = 0.56097
41 41 41
where the bar denotes the usual repeating decimal.
However, if the number under the bar of 1/41 is denoted by c (that is, c = 2439), then
4 23
the number under the bar of 41 is 4c and the number under the bar of 41 is 23c. So in
the case (m, n) = (4, 23), the choice c = 2439 works great.
Solution. To get the above argument to work for bigger values of m and n, we’ll need
to replace 41 with some larger denominator D satisfying three ingredients:

• gcd(D, 10) = 1, so the decimal representation of 1/D is indeed periodic;

m n
• D > max(m, n), so D and D are both in in the interval (0, 1); and
m n
• for some exponent e, 10e · D − D is an integer; this gave us the shifting property.

We prove this is possible now for any m and n.

Claim. There exists a number D satisfies which satisfies all three conditions.

Proof. Let’s describe a recipe to find D. Suppose we pick some exponent e

and define the number
A = 10e n − m.

Suppose 2 divides m exactly r ≥ 0 times. Then if e > r, it follows that 2

divides A exactly r times too. Similarly, if 5 divides m exactly s ≥ 0 times,
then as long as e > s it follows that 5 divides A exactly s times too.
Now choose any e larger than max(r, s), so the previous paragraph applies,
and also big enough that A > 2r 5s max(m, n). Then the number D = 2rA5s
obtained by deleting all factors of 2 and 5 from A should work. Indeed, by
construction, gcd(D, 10) = 1 and D > max(m, n). And the particular form
of A gives us
n m A
10e · − = = 2r 5s
D D D
which is an integer.

1
During the USAMO, it is not necessary to include motivation like this if the solution stands without
it. However, we chose to supply it anyways so you can see how one might come up with the solution,
and to make it easier for you to follow along.

5
Now we take c to be the number under the bar of 1/D (leading zeros removed). Then
the decimal representation of mD is the decimal representation of cm repeated (possibly
n
including leading zeros). Similarly, D has the decimal representation of cm repeated
(possibly including leading zeros). Finally, since
m n
10e · − is an integer
D D
it follows that these repeating decimal representations are rotations of each other by e
places, so in particular they have the same number of nonzero digits.

This problem was proposed by Richard Stong. Incidentally, for many students the number
142857 might ring a bell, and is quite related to the problem proposed here. It is possible
(but not necessary) to find a D which is prime, although this requires more care.