The Forty-Sixth Annual William Lowell Putnam Competition

Saturday, December 7, 1985

Done all

A–1 Determine, with proof, the number of ordered triples (A1 , A2 , A3 ) of sets which have the property that
(i) A1 ∪ A2 ∪ A3 = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}, and
(ii) A1 ∩ A2 ∩ A3 = ∅.
Express your answer in the form 2a 3b 5c 7d , where a, b, c, d are nonnegative integers.

Solution: For each integer i from 1 to 10 you must put i in one of 6 sets: A1 ∩ Ac2 ∩ Ac3 or A1 ∩ Ac2 ∩ A3 or
A1 ∩ A2 ∩ Ac3 or Ac1 ∩ A2 ∩ A3 or Ac1 ∩ Ac2 ∩ A3 or Ac1 ∩ A2 ∩ Ac3 . Thus there are 6 choices for each of the
10 values of i giving the answer

610 = 210 310 .

A–2 Let T be an acute triangle. Inscribe a rectangle R in T with one side along a side of T . Then inscribe a rectangle S in the
triangle formed by the side of R opposite the side on the boundary of T , and the other two sides of T , with one side along
the side of R. For any polygon X, let A(X) denote the area of X. Find the maximum value, or show that no maximum
exists, of A(R)+A(S)
A(T ) , where T ranges over all triangles and R, S over all rectangles as above.

Solution: The problem is evidently scale invariant so we take the side of the triangle on which R lies on the
x-axis running from 0 to 1. Put the vertex at (x, y) with y > 0 and 0 < x < 1 to make the triangle acute. We
take the combined height of R and S to be λy with 0 < λ < 1 and the height of triangle R to be λθy for some
0 < θ < 1.
The corners of R are at

(λθx, 0); (λθx, λθy); (1 − λθ(1 − x), 0); (1 − λθ(1 − x), λθy).

The area is then

A(R) = (1 − λθ)λθy

The corners of S are at

(λx, λθy); (λx, λy); (1 − λ(1 − x), λθy); (1 − λ(1 − x), λy).

and then

A(S) = λ(1 − λ)(1 − θ)y

Finally

A(T ) = y/2.

This makes
A(S) + A(R)
= 2 {(1 − λθ)λθ + λ(1 − λ)(1 − θ)}
A(T )

which simplifies to

2(−λ2 θ2 + λ2 θ + λ − λ2 ).
 2

At λ = 0 this ratio vanishes. The derivative with respect to θ is

2λ2 − 4λ2 θ

which vanishes for λ > 0 at, and only at, θ = 1/2. The second θ derivative is negative so that θ = 1/2
maximizes the ratio for each λ > 0. At θ = 1/2 the ratio becomes

2(λ − 3λ2 /4) = 2/3 − 3(λ − 2/3)2 /2.

This is evidently maximized at λ = 2/3 and the maximum value is 2/3.

A–3 Let d be a real number. For each integer m ≥ 0, define a sequence {am (j)}, j = 0, 1, 2, . . . by the condition

am (0) = d/2m ,
am (j + 1) = (am (j))2 + 2am (j), j ≥ 0.

Evaluate limn→∞ an (n).

Solution: Let bm (j) = am (j) + 1 so that

bm (j + 1) − 1 = (bm (j) − 1)2 + 2(bm (j) − 1)

or

bm (j + 1) = bm (j)2 = bm (j − 1)4 = · · · .
m
This makes bm (m) = bm (0)2 . Since bm (0) = 1 + d/2m we get
m
am (m) = bm (m) − 1 = (1 + d/2m )2 − 1 → ed − 1.

A–4 Define a sequence {ai } by a1 = 3 and ai+1 = 3ai for i ≥ 1. Which integers between 00 and 99 inclusive occur as the
last two digits in the decimal expansion of infinitely many ai ?

Solution: We begin by computing some powers of 3 modulo 100 to discover that

320 ≡ 1 mod 100

The powers 3j for j from 1 to 20 are, modulo 100,

3, 9, 27, 81, 43, 29, 87, 61, 83, 49, 47, 41, 23, 69, 7, 21, 63, 89, 67, 1

Any integer n may be written in the form

n = 20j + 4k + l

where 4k + l is the residue class of n modulo 20 and l the residue class of n modulo 4 (so that l ∈ {0, 1, 2, 3}
and k ∈ {0, 1, 2, 3, 4}). In this case

3n ≡ 34k+l mod 100.

Next note that

34 ≡ 1 mod 20

from which it follows that

34k ≡ 1 mod 20.

Let ji , ki , li satisfy

ai = 20ji + 4ki + li
 3

as above. Then

ai+1 ≡ 3li mod 20

We have l1 = 3. If li = 3 then

3li = 27 ≡ 7 mod 20

which makes li+1 = 3. Thus li = 3 for all i. Next

ai+1 ≡ 34ki +li mod 100

We have k1 = 0, k1 = 1. If ki = 1 then

34ki +li = 37 ≡ 87 mod 100

so that ki+1 = 1. Thus we have, for all i ≥ 2 that ki = 1 and li = 3. The residue class of ai+1 modulo 100
is then that of 37 which is 87. Thus only 87 occurs infinitely often as the last two digits of ai – in fact 87 is the
last two digits of ai for all i ≥ 3.
R 2π
A–5 Let Im = 0 cos(x) cos(2x) · · · cos(mx) dx. For which integers m, 1 ≤ m ≤ 10 is Im 6= 0?

Solution: I claim the integral is non-zero only for m ∈ {3, 4, 7, 8} (and in general for m congruent to 0 or 3
mod 4). Write cos(jx) = (exp(jix) + exp(−jix))/2 and put
m
Y m
Y
fm (x) = cos(jx) = 2−m {(exp(jix) + exp(−jix)} .
1 1

Let I denote generic subset of {1, . . . , m} and I¯ its complement. Then the product in fm may be expanded to
get
X Z 2π X X
−m
Im = 2 exp{i{ j− j}x}dx.
0
I j∈I j∈I¯

The integral is 0 unless

X X
j− j = 0.
j∈I j∈I¯

But the two sums in this formula add up to m(m + 1)/2 so the difference is actually
X
2 j − m(m + 1)/2.
j∈I

We thus have
Z X
fm (x)dx = 2−m 2π.
P
I: j∈I j=m(m+1)/4

This is 0 only if there are no subsets I satisfying the criterion. For m ∈ {1, 2, 5, 6, 9, 10} (or in general m
congruent to 1 or 2 modulo 4) the quantity m(m + 1)/4 is not an integer so no such I exists and Im = 0. For
m = 3 take I = {1, 2}. For m = 4 take I = {1, 4}. For m = 7 take I = {1, 2, 4, 7} and for m = 8 take
I = {1, 2, 7, 8}. The examples show that the integral is positive in each case.

A–6 If p(x) = a0 + a1 x + · · · + am xm is a polynomial with real coefficients ai , then set

Γ(p(x)) = a20 + a21 + · · · + a2m .

Let F (x) = 3x2 + 7x + 2. Find, with proof, a polynomial g(x) with real coefficients such that
(i) g(0) = 1, and
 4

(ii) Γ(F (x)n ) = Γ(g(x)n )

for every integer n ≥ 1.

Solution: I remark that I don’t like the notation Γ(p(x)); I want Γ(p) since the quantity in question is a
functional of the function p not its value at x. Check that
Z 2π
1
p eix p e−ix dx



Γ(p) =
2π 0
by writing the product as a double sum and integrating term by term. This means that
Z 2π
1
Γ(F n ) = F n eix F n e−ix dx



2π 0
and we want this to be the same as
Z 2π
1
n
g n eix g n e−ix dx



Γ(g ) =
2π 0

for a suitable polynomial g. Notice that if we can find a g satisfying (a) and such that |z| = 1 implies |g(z)| =
|F (z)| then we will be done since the integrands are just

|F eix |2n

and

|g eix |2n .

Now I will find a and b such that

g(x) = ax2 + bx + 1

solves the problem. If z = exp ix then

|g(z)| = |a cos(2x) + b cos(x) + 1 + i(a sin(2x) + b sin(x))|

while

|F (z)| = |3 cos(2x) + 7 cos(x) + 2 + i(3 sin(2x) + 7 sin(x))|.

The equation |F (z)| = |g(z)| becomes

62 + 42(sin(2θ) sin(θ) + cos(2θ) cos(θ)) + 12 cos(2θ) + 28 cos(θ)

= a2 + b2 + 1 + 2ab(sin(2θ) sin(θ) + cos(2θ) cos(θ)) + 2a cos(2θ) + 2b cos(θ).

A standard trigonometric formula reduces this to

62 + 70 cos(θ) + 12 cos(2θ) = a2 + b2 + 1 + 2b(a + 1) cos(θ) + 2a cos(2θ).

Take a = 6 to make the coefficients of cos(2θ) match. Then 14b = 70 gives b = 5 and finally

62 + 52 + 1 = 62

which establishes the identity. Thus

g(x) = 6x2 + 5x + 1.

B–1 Let k be the smallest positive integer for which there exist distinct integers m1 , m2 , m3 , m4 , m5 such that the polynomial

p(x) = (x − m1 )(x − m2 )(x − m3 )(x − m4 )(x − m5 )

has exactly k nonzero coefficients. Find, with proof, a set of integers m1 , m2 , m3 , m4 , m5 for which this minimum k is
achieved.
 5

Solution: I claim k = 3 and that one set is m1 = 2, m2 = 1, m3 = 0, m4 = −1 and m5 = −2. First

(x − 2)(x − 1)x(x + 1)(x + 2) = x5 − 5x3 + 4x.

Q5
showing k ≤ 3. If p(x) = 1 (x − mi ) has k = 2 then

p(x) = x5 + axr

for some integers a and r with 0 ≤ r ≤ 4. If r ≥ 2 then x = 0 is a double root of p(x) while if r = 1 the
roots of p are 0 and the 4 roots of −a in the complex plane – at most 2 of which can be integers. If r = 0 then
the roots of p are the 5 fifth roots of −a at least 3 of which are complex. And if a = 0 then p(x) = x5 has a
multiple root at 0.

B-2 Define polynomials fn (x) for n ≥ 0 by f0 (x) = 1, fn (0) = 0 for n ≥ 1, and

d
fn+1 (x) = (n + 1)fn (x + 1)
dx
for n ≥ 0. Find, with proof, the explicit factorization of f100 (1) into powers of distinct primes.

Solution: Define

fn (x) = (x + n)n − n(x + n)n−1 .

Check that

fn (0) = nn − nnn−1 = 0

and
d
fn+1 (x) = (n + 1)(x + n + 1)n − (n + 1)n(x + n + 1)n−1 = fn (x + 1).
dx
Then

fn (1) = (n + 1)n − n(n + 1)n−1 = (n + 1)n−1 (n + 1 − n) = (n + 1)n−1

so that

f100 (1) = 10199

is the desired factorization.

B–3 Let
a1,1 a1,2 a1,3 ...
a2,1 a2,2 a2,3 ...
a3,1 a3,2 a3,3 ...
.. .. .. ..
. . . .

be a doubly infinite array of positive integers, and suppose each positive integer appears exactly eight times in the array.
Prove that am,n > mn for some pair of positive integers (m, n).

Solution: Consider the pairs Mj = 2K−j and Nj = 2j for j = 0, . . . , K where K is a positive integer.
Note that Mj Nj = 2K . Let Cj = {(m, n) : m ≤ Mj and n ≤ Nj }. Let B0 = C0 and for j ≥ 1 put
Bj = Cj \ Cj−1 . Note that the cardinality of Bj is 2K−1 for j ≥ 1 and 2K for j = 0. Thus the cardinality of

B = ∪K
0 Bj

is

2K + K2K−1
 6

If K is large enough so that

2K + K2K−1 > 82K

or

(1 + K/2) > 8

or

K > 14

then under the conditions of the theorem there must be a pair (m, n) ∈ B with

am,n > 2K ≥ mn

B–4 Let C be the unit circle x2 + y 2 = 1. A point p is chosen randomly on the circumference C and another point q is chosen
randomly from the interior of C (these points are chosen independently and uniformly over their domains). Let R be the
rectangle with sides parallel to the x and y-axes with diagonal pq. What is the probability that no point of R lies outside
of C?

Solution: Write p = (cos θ, sin θ) with θ uniformly distributed on (0, 2π). Let S be the rectangle with corners
at (± cos θ, ± sin θ). Then R lies inside C if and only if q ∈ S. The area of S is |4 cos θ sin θ|. The unit circle
has area π so given θ the probability that R lies inside C is

4| cos θ sin θ|
.
π
Now average over θ to find that the desired probability is
Z 2π
4| cos θ sin θ|
dθ/(2π).
0 π
This is just
Z π/2
8 4
cos θ sin θdθ = .
π2 0 π2

R∞ −1 R∞ 2 √
B–5 Evaluate 0
t−1/2 e−1985(t+t )
dt. You may assume that −∞
e−x dx = π.

Solution: I haven’t bothered to check all the algebra here. The method works but the algebra might be defective.
Define
Z ∞
Gn (θ) = t−n/2 e−t e−θ/t dt.
0

Note that
√
G1 (0) = Γ(1/2) = π;

this is a consequence via a change of variables of the hint given. Differentiate G1 with respect to θ under the
integral sign to see that for θ > 0 (needed to justify an application of the dominated convergence theorem) we
have
∂G
= θG5 (θ).
∂θ
Then in G5 integrate by parts taking dv = θ−1 t−2 exp(−θ/t) and u = exp(−t) to get

∂G
= −G1 (θ) − 2−1 G3 (θ).
∂θ
 7

Then make the substitution u = θ/t in the integral defining G3 . This shows
√
G1 (θ) = θG3 (θ).

This gives

∂G √
= −(1 + (2 θ)−1 )G1 .
∂θ
In turn we see
∂ log G √
= −(1 + (2 θ)−1 ).
∂θ
This gives
√
log G1 (θ) = −θ + θ + log G(0)

or
√ √
G1 (θ) = π exp{ θ − θ}.

.
Finally: if we substitute u = 1985t in the integral in the question we get
√ √
1985G1 (19852 ) = 1985π exp{1985 − 19852 }.

Technical points. The Dominated Convergence Theorem applies only for θ > 0 to permit differentiation under
the integral sign. Having integrated the resulting differential equation for θ > 0, however, we are entitled to
evaluate at θ = 0 because G1 is continuous at θ = 0 as may be verified by the DCT again.

B–6 Let G
Pbe a finite set of real n × n matrices {Mi }, 1 ≤ i ≤ r, which form a groupPunder matrix multiplication. Suppose
r r
that i=1 tr(Mi ) = 0, where tr(A) denotes the trace of the matrix A. Prove that i=1 Mi is the n × n zero matrix.

Solution: If M , M1 and M2 are in the group G then Pr M M1 = M M2 implies M1 = M2 because M is

invertible. Thus {M M1 , . . . , M Mr } = G. If A = i=1 Mi /r then we see that M A = A for all M ∈ G
and so A2 = A. If λ is an eigenvalue of A with non-zero eigenvector v then A2 v = Av so λ2 v = λv so
λ(1 − λ) = 0. That is each eigenvalue of A is either 1 or 0. But the trace of any matrix is the sum of the
roots of its characteristic polynomial. Since each such root is an eigenvalue we see that the trace of A is the
multiplicity of 1 as a characteristic value of A. Since the sum is 0 the characteristic polynomial of A is λn .
Now from AA = A we learn that each non-zero column of A is an eigenvector of A for the eigenvalue 1. But
there are no such eigenvectors so every column of A is 0.