APMO 1989 – Problems and Solutions

Problem 1
Let x1 , x2 , . . . , xn be positive real numbers, and let

S = x1 + x2 + · · · + xn .

Prove that
S2 S3 Sn
(1 + x1 )(1 + x2 ) · · · (1 + xn ) ≤ 1 + S + + + ··· + .
2! 3! n!

Solution 1
Let σk be the kth symmetric polynomial, namely
X Y
σk = xi ,
|S|=k i∈S
S⊆{1,2,...,n}

and more explicitly

σ1 = S, σ2 = x1 x2 + x1 x3 + · · · + xn−1 xn , and so on.

Then
(1 + x1 )(1 + x2 ) · · · (1 + xn ) = 1 + σ1 + σ2 + · · · + σn .
The expansion of

S k = (x1 + x2 + · · · + xn )k = (x1 + x2 + · · · + xn )(x1 + x2 + · · · + xn ) · · · (x1 + x2 + · · · + xn )

| {z }
k times
Q
has at least k! occurrences of i∈S xi for each subset S with k indices from {1, 2, . . . , n}. In fact,
if π is a permutation of S, we can choose each xπ(i) from the ith factor of (x1 + x2 + · · · + xn )k .
Then each term appears at least k! times, and

Sk
S k ≥ k!σk ⇐⇒ σk ≤ .
k!
Summing the obtained inequalities for k = 1, 2, . . . , n yields the result.
Solution 2
By AM-GM,
 n  n
(1 + x1 ) + (1 + x2 ) + · · · + (1 + xn ) S
(1 + x1 )(1 + x2 ) · · · (1 + xn ) ≤ = 1+ .
n n
By the binomial theorem,
n X n    k n n
1 n(n − 1) . . . (n − k + 1) k X S k

S n S X
1+ = = S ≤ ,
n k=0
k n k=0
k! nk k=0
k!

and the result follows.

Comment: Maclaurin’s inequality states that
σ1 σ2 σk σn
r r r
≥ n

≥ · · · ≥ k
n

≥ · · · ≥ n
n .


n 2 k n

n Sk 1 n(n−1)...(n−k+1) k Sk


Then σk ≤ k nk
= k! nk
S ≤ k!
.

1
Problem 2
Prove that the equation
6(6a2 + 3b2 + c2 ) = 5n2
has no solutions in integers except a = b = c = n = 0.
Solution
We can suppose without loss of generality that a, b, c, n ≥ 0. Let (a, b, c, n) be a solution with
minimum sum a + b + c + n. Suppose, for the sake of contradiction, that a + b + c + n > 0.
Since 6 divides 5n2 , n is a multiple of 6. Let n = 6n0 . Then the equation reduces to

6a2 + 3b2 + c2 = 30n20 .

The number c is a multiple of 3, so let c = 3c0 . The equation now reduces to

2a2 + b2 + 3c20 = 10n20 .

Now look at the equation modulo 8:

b2 + 3c20 ≡ 2(n20 − a2 ) (mod 8).

Integers b and c0 have the same parity. Either way, since x2 is congruent to 0 or 1 modulo 4,
b2 + 3c20 is a multiple of 4, so n20 − a2 = (n0 − a)(n0 + a) is even, and therefore also a multiple
of 4, since n0 − a and n0 + a have the same parity. Hence 2(n20 − a2 ) is a multiple of 8, and

b2 + 3c20 ≡ 0 (mod 8).

If b and c0 are both odd, b2 + 3c20 ≡ 4 (mod 8), which is impossible. Then b and c0 are both
even. Let b = 2b0 and c0 = 2c1 , and we find

a2 + 2b20 + 6c21 = 5n20 .

Look at the last equation modulo 8:

a2 + 3n20 ≡ 2(c21 − b20 ) (mod 8).

A similar argument shows that a and n0 are both even.

We have proven that a, b, c, n are all even. Then, dividing the original equation by 4 we find

6(6(a/2)2 + 3(b/2)2 + (c/2)2 ) = 5(n/2)2 ,

and we find that (a/2, b/2, c/2, n/2) is a new solution with smaller sum. This is a contradiction,
and the only solution is (a, b, c, n) = (0, 0, 0, 0).

2
Problem 3
Let A1 , A2 , A3 be three points in the plane, and for convenience,let A4 = A1 , A5 = A2 . For
n = 1, 2, and 3, suppose that Bn is the midpoint of An An+1 , and suppose that Cn is the midpoint
of An Bn . Suppose that An Cn+1 and Bn An+2 meet at Dn , and that An Bn+1 and Cn An+2 meet
at En . Calculate the ratio of the area of triangle D1 D2 D3 to the area of triangle E1 E2 E3 .
25
Answer: .
49
Solution
Let G be the centroid of triangle ABC, and also the intersection point of A1 B2 , A2 B3 , and
A3 B1 .

A1

C1 E1
B1 D1
G

A2 C2 B2 A3

By Menelao’s theorem on triangle B1 A2 A3 and line A1 D1 C2 ,

A1 B1 D1 A3 C2 A2 D1 A3 D1 B1 1
· · = 1 ⇐⇒ = 2 · 3 = 6 ⇐⇒ = .
A1 A2 D1 B1 C2 A3 D1 B1 A3 B1 7

Since A3 G = 23 A3 B1 , if A3 B1 = 21t then GA3 = 14t, D1 B1 = 21t

7
= 3t, A3 D1 = 18t, and
GD1 = A3 D1 − A3 G = 18t − 14t = 4t, and
GD1 4 2
= = .
GA3 14 7
Similar results hold for the other medians, therefore D1 D2 D3 and A1 A2 A3 are homothetic with
center G and ratio − 72 .
By Menelao’s theorem on triangle A1 A2 B2 and line C1 E1 A3 ,
C1 A1 E1 B2 A3 A2 E1 B2 1 3 A1 E1 2
· · = 1 ⇐⇒ =3· = ⇐⇒ = .
C1 A2 E1 A1 A3 B2 E1 A1 2 2 A1 B2 5
2
If A1 B2 = 15u, then A1 G = 3
· 15u = 10u and GE1 = A1 G − A1 E1 = 10u − 52 · 15u = 4u, and

GE1 4 2
= = .
GA1 10 5
Similar results hold for the other medians, therefore E1 E2 E3 and A1 A2 A3 are homothetic with
center G and ratio 52 .
Then D1 D2 D3 and E1 E2 E3 are homothetic with center G and ratio − 27 : 25 = − 75 , and the ratio

2
of their area is 57 = 2549
.

3
Problem 4
Let S be a set consisting of m pairs (a, b) of positive integers with the property that 1 ≤ a <
b ≤ n. Show that there are at least
2
(m − n4 )
4m
3n
triples (a, b, c) such that (a, b), (a, c), and (b, c) belong to S.
Solution
Call a triple (a, b, c) good if and only if (a, b), (a, c), and (b, c) all belong to S. For i in
{1, 2, . . . , n}, let di be the number of pairs in S that contain i, and let Di be the set of numbers
paired with i in S (so |Di | = di ). Consider a pair (i, j) ∈ S. Our goal is to estimate the number
of integers k such that any permutation of {i, j, k} is good, that is, |Di ∩ Dj |. Note that i ∈ / Di
and j ∈ / Dj , so i, j ∈ / Di ∩ Dj ; thus any k ∈ Di ∩ Dj is different from both i and j, and {i, j, k}
has three elements as required. Now, since Di ∪ Dj ⊆ {1, 2, . . . , n},

|Di ∩ Dj | = |Di | + |Dj | − |Di ∪ Dj | ≤ di + dj − n.

Summing all the results, and having in mind that each good triple is counted three times (one
for each two of the three numbers), the number of good triples T is at least
1 X
T ≥ (di + dj − n).
3
(i,j)∈S

Each term di appears each time i is in a pair from S, that is, di times; there are m pairs in S,
so n is subtracted m times. By the Cauchy-Schwartz inequality
n
! !
2
1 ( ni=1 di )
P
1 X 2
T ≥ d − mn ≥ − mn .
3 i=1 i 3 n

Finally, the sum ni=1 di is 2m, since di counts the number of pairs containing i, and each pair
P
(i, j) is counted twice: once in di and once in dj . Therefore
2
(2m)2 (m − n4 )
 
1
T ≥ − mn = 4m .
3 n 3n

Comment: This is a celebrated graph theory fact named Goodman’s bound, after A. M. Good-
man’s method published in 1959. The generalized version of the problem is still studied to this
day.

4
Problem 5
Determine all functions f from the reals to the reals for which

(1) f (x) is strictly increasing,

(2) f (x) + g(x) = 2x for all real x, where g(x) is the composition inverse function to f (x).

(Note: f and g are said to be composition inverses if f (g(x)) = x and g(f (x)) = x for all real
x.)

Answer: f (x) = x + c, c ∈ R constant.

Solution
Denote by fn the nth iterate of f , that is, fn (x) = f (f (. . . f (x))).
| {z }
n times
Plug x → fn+1 (x) in (2): since g(fn+1 (x)) = g(f (fn (x))) = fn (x),

fn+2 (x) + fn (x) = 2fn+1 (x),

that is,
fn+2 (x) − fn+1 (x) = fn+1 (x) − fn (x).
Therefore fn (x) − fn−1 (x) does not depend on n, and is equal to f (x) − x. Summing the
corresponding results for smaller values of n we find

fn (x) − x = n(f (x) − x).

Since g has the same properties as f ,

gn (x) − x = n(g(x) − x) = −n(f (x) − x).

Finally, g is also increasing, because since f is increasing g(x) > g(y) =⇒ f (g(x)) >
f (g(y)) =⇒ x > y. An induction proves that fn and gn are also increasing functions.
Let x > y be real numbers. Since fn and gn are increasing,

x + n(f (x) − x) > y + n(f (y) − y) ⇐⇒ n[(f (x) − x) − (f (y) − y)] > y − x

and
x − n(f (x) − x) > y − n(f (y) − y) ⇐⇒ n[(f (x) − x) − (f (y) − y)] < x − y.
Summing it up,
|n[(f (x) − x) − (f (y) − y)]| < x − y for all n ∈ Z>0 .
Suppose that a = f (x) − x and b = f (y) − y are distinct. Then, for all positive integers n,

|n(a − b)| < x − y,

which is false for a sufficiently large n. Hence a = b, and f (x) − x is a constant c for all x ∈ R,
that is, f (x) = x + c.
It is immediate that f (x) = x + c satisfies the problem, as g(x) = x − c.