13th Putnam 1953

Problem A1

Show that (2/3) n3/2 < ∑1n √r < (2/3) n3/2 + (1/2) √n.

Solution

The gradient of √x is falling from x = 0 to x = n, so 2/3 n3/2 = ∫0n √x dx < √1 + √2 + √3 + ... + √n. That gives
the first inequality.

To get the second, we note that the chords joining (r, √r) to (r+1, √(r+1) ) all lie under the curve, so if we
subtract the area of the little triangles from ∑1n √r then we get something less than the integral. The triangles
have area 1/2 √1 + 1/2 (√2 - √1) + 1/2 (√3 - √2) + ... + 1/2 (√n - √(n-1) ) = 1/2 √n. That gives the second
inequality.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem A2

The complete graph with 6 points and 15 edges has each edge colored red or blue. Show that we can find 3
points such that the 3 edges joining them are the same color.

Solution

Take any point A. It has 5 edges, so at least 3 of them must be the same color. wlog it is red. So we have B,
C, D with AB, AC, AD all red. Now if any of the three edges BC, CD, DB is red, then that gives us a red
triangle. But if they are all blue, then BCD is a blue triangle.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem A3

a, b, c are real, and the sum of any two is greater than the third. Show that 2(a + b + c)(a2 + b2 + c2)/3 > a3 +
b3 + c3 + abc.

Solution

wlog we may take a ≥ b ≥ c.

c (and hence also a and b) must be positive. For if c ≤ 0, then a ≥ b + c. Contradiction.

Multiplying out the required relation and cancelling, we must prove that 2(a2b + a2c + b2a + b2c + c2a + c2b)
> a3 + b3 + c3 + 3abc. But we have a2(b + c) > a3, b2(c + a) > b3, c2(a + b) > c3. So it is sufficient to prove
that a2b + a2c + b2a + b2c + c2a + c2b > 3abc. But a2b ≥ abc and ab2 ≥ abc. Also b2c + bc2 = bc(b + c) > abc.
Similarly, ac(a + c) > abc. So we have in fact proved the slightly stronger result with abc replaced by 4abc/3.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem A4

Using sin x = 2 sin x/2 cos x/2 or otherwise, find ∫0π/2 ln sin x dx.

Solution

Let I = ∫0π/2 ln sin x dx. Making the suggested substitution, we get I = π/2 ln 2 + ∫0π/2 ln sin x/2 cos x/2 dx.
Putting y = x/2 we get ∫0π/2 ln sin x/2 cos x/2 dx = 2 ∫0π/4 ln sin y + ln cos y dy. But cos y = sin(π/2 - y), so
∫0π/4 ln cos y dy = ∫π/4π/2 ln sin y dy. Hence I = π/2 ln 2 + 2I, so I = - π/2 ln 2.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem A5

S is a parabola with focus F and axis L. Three distinct normals to S pass through P. Show that the sum of the
angles which these make with L less the angle which PF makes with L is a multiple of π.

Solution

We start by finding a formula for tan(A+B+C+D). Applying tan(A+B) = (tan A + tan B)/(1 - tan A tan B)
twice and writing a = tan A, b = tan B, c = tan C, d = tand D, we get tan(A+B+C+D) = (∑a - ∑abc)/(1 - ∑ab
+ abcd).

Take the parabola as y = x2. The gradient at (k, k2) is 2k, so the normal has slope -1/2k. The normal has
equation 2k3 + (1 - 2y)k - x = 0. For given x, y this is a cubic in k, so has 1 or real 3 roots. Their slopes
satisfy the cubic 4x h3 + 2(1 - 2y) h2 + 1 = 0 (*).

The focus is at (0, 1/4), so the line joining (x, y) has slope (y - 1/4)/x and negative is (1/4 - y)/x. The quartic
with this root and the 3 roots of (*) is (h + (y - 1/4)/x )(4x h3 + 2(1 - 2y) h2 + 1) (**). So the angles the
normals make with L less the angle which PF makes with L is a multiple of π iff (∑a = ∑abc), where a, b, c, d
are the slopes of the four lines and hence iff the coefficients of h3 and h in (**) are the same. But the
coefficient of h3 is 2(1 - 2y) + 4x (y - 1/4)/x = 1 and the coefficient of h is 1.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem A6

Show that √7, √(7 - √7), √(7 - √(7 + √7)), √(7 - √(7 + √(7 - √7))), ... converges and find its limit.

Solution

Answer: 2.

Let xn be the nth number in the sequence. We have xn+2 = √(7 - √(7 + xn)). Hence if xn < 2, then xn+2 > 2,
and if xn > 2, then xn+2 < 2. So if the sequence converges, its limit must be 2.

If xn = 2 + ε, with 0 < ε < 1, then 9 + ε < 9 + ε + ε2/36 = (3 + ε/6)2, so xn+2 > √(7 - (3 + ε/6)) = √(4 - ε/6). But
certainly ε/6 < ε/3 - ε2/144, so √(4 - ε/6) > 2 - ε/12. Thus xn+2 differs from 2 by less than ε/12. Similarly, if xn
= 2 - ε, with 0 < ε < 1, then 9 - ε > 9 - ε/2 + ε2/144 = (3 - ε/12)2, so xn+2 < √(4 + ε/12) < 2 + ε/48.

So we have established that if |xn - 2| < 1, then |xn+2 - 2| < 1/12 |xn - 2|. But we certainly have |x1 - 2| < 1, and
|x2 - 2| < 1, so xn converges to 2.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem A7

p(x) ≡ x3 + ax2 + bx + c has three positive real roots. Find a necessary and sufficient condition on a, b, c for
the roots to be cos A, cos B, cos C for some triangle ABC.

Solution

If A + B + C = 180o, then cos A = - cos(B + C) = sin B sin C - cos B cos C. Squaring, we get cos2A + cos2B
+ cos2C + 2 cos A cos B cos C = 1. So a necessary condition is a2 - 2b - 2c = 1.

Conversely, suppose that this condition holds. Then if the roots are p, q, r, we have p2 + q2 + r2 + 2pqr = 1
(*). We are given that the roots are all positive, so 2pqr > 0, hence p2 < 1 and so p < 1. Similarly for q and r.
So we can find angles A, B, C greater than 0 and less than 90o such that p = cos A, q = cos B, r = cos C. Now
we can rewrite (*) as (1 - cos2B)(1 - cos2C) = cos2A + 2 cos A cos B cos C + cos2B cos2C = (cos A + cos B
cos C)2. But (1 - cos2B) = sin2B, (1 - cosC) = sin2C, so we have sin B sin C = ±(cos A + cos B cos C). But
we know that A, B, C are between 0 and 90o, so cos A, cos B, cos C, sin B, sin C are all positive. Hence we
must use the + sign and we have cos A = sin B sin C - cos B cos C, so A + B + C = 180o. Hence the condition
is also sufficient.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B1

Does ∑1∞ 1/n1 + 1/n converge?

Solution

Answer: no.

x < ex for all x ≥ 0, so x1/x < e < 3. Hence 1/n1 + 1/n > 1/3n. But ∑ 1/n diverges.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B2

p(x) is a real polynomial of degree n such that p(m) is integral for all integers m. Show that if k is a
coefficient of p(x), then n! k is an integer.

Solution

Note that n! is best possible, because 1/n! x(x + 1) ... (x + n - 1) is always integral for integral x (it is the
binomial coefficient (x+n-1)Cn ).

We need a standard result from the calculus of differences. Let Δf(x) = f(x + 1) - f(x). Then p(x) = p(0) + Δ1x
+ 1/2! Δ2 x(x - 1) + 1/3! Δ3 x(x - 1)(x - 2) + ... + 1/n! Δn x(x - 1) ... (x - n + 1) (*), where Δm = Δmp(0) (thus
Δ1 = p(1) - p(0), Δ2 = p(2) - 2p(1) + p(0) etc).

Assume this is true. Then if p(x) is integral for all integral x, all the Δm must be integral. So the result above
gives n! p(x) = an integral combination of integral polynomials. Hence all the coefficients of n! p(x) are
integral.

To prove the result, notice first that if f(x) = x(x - 1)(x - 2) ... (x - m + 1), then Δrf(0) = m! for r = m and 0
otherwise. So if we call the rhs of (*) q(x) , then Δmq(0) is a sum of terms which are all zero except for
Δm(1/m! Δm x(x - 1) ... (x - m + 1) )(0) = Δm. Hence Δmp(0) = Δmq(0) for m = 1, 2, ... , n. Also p(0) = q(0),
so by a simple induction p(m) = q(m) for m = 0, 1, 2, ... , n. But q(x) is a polynomial of degree at most n. If
polyomials of degree at most n agree at n+1 points, then they must be identical. Hence p(x) = q(x).

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B3

k is real. Solve the differential equations y' = z(y + z)k, z' = y(y + z)k subject to y(0) = 1, z(0) = 0.

Solution

Adding, (y + z)' = (y + z)k+1. Integrating (y + z)k = 1/(1 - kx) (1).

Multiplying opposite sides of the two given equations together, we get yy'(y + z)k = zz' (y + z)k. Hence yy' =
zz'. Integrating y2 - z2 = 1. Hence (y + z)k(y - z)k = 1, so (y - z)k = 1 - kx (2).

For k non-zero, we can immediately solve (1) and (2) to get y = 1/2 ( (1 - kx)1/k + 1/(1 - kx)1/k ), z = 1/2 ( (1 -
kx)1/k - 1/(1 - kx)1/k ).

For k = 0, the original equations simplify to y' = z, z'= y. So y'' = y. So y = A cosh x + B sinh x, z = A cosh x +
B sinh x. Applying the initial conditions, y = cosh x, z = sinh x.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B4

R is the reals. S is a surface in R3 containing the point (1, 1, 1) such that the tangent plane at any point P ∈ S
cuts the axes at three points whose orthocenter is P. Find the equation of S.

Solution

Consider a plane cutting the axes at a = (a, 0, 0), b = (0, b, 0), c = (0, 0, c). If the orthocentre is at p = (x, y,
z), then we have (p - a).(b - c) = (p - b).(a - c) = 0. But a.b = b.c = c.a = 0, so we have p.(b - c) = p.(a - c) =
0. In other words the line from the origin (0, 0, 0) to (x, y, z) is normal to the plane. So the surface satisfies
the condition that all its normals pass through the origin and it passes through (1, 1, 1). This implies that it is
the sphere x2 + y2 + z2 = 3.

Note, however, that for points with one coordinate zero, the tangent plane will meet one axis at infinity, so we
should arguable exclude all such points. That divides the sphere into 8 disconnected pieces. The piece
containing (1, 1, 1) is that in the positive octant (x > 0, y > 0, z > 0).

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B5

The coefficients of the complex polynomial z4 + az3 + bz2 + cz + d satisfy a2d = c2 ≠ 0. Show that the ratio
of two of the roots equals the ratio of the other two.

Solution

We start with a straight slog. Let the roots be p, q, r, s. We have a2d - c2 = (p + q + r + s)2pqrs - (pqr + pqs +
prs + qrs)2. The terms like 2p2q2rs all cancel, leaving p3qrs + pq3rs + pqr3s + pqrs3 - p2q2r2 - p2q2s2 - p2r2s2
- q2 r2 s 2 .

The trick now is to factorise this. We might suspect that pq - rs is a factor. But in that case pr - qs and ps - qr
would presumably also be factors. (pq - rs)(pr - qs)(ps - qr) has degree 6, as required. It also has the correct
number of terms (8). So we try multiplying it out and find that it is the same.

So (pq - rs)(pr - qs)(ps - qr) = 0. But that means that at least one factor must be zero, which gives the result.
Note that the only reason for giving us that a2d is non-zero, is because that implies that none of the roots are
zero (their product d is non-zero) and so having got pq = rs, we can divide to get p/r = s/q.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B6

A and B are equidistant from O. Given k > OA, find the point P in the plane OAB such that OP = k and PA +
PB is a minimum.

Solution

Let C be the circle centre O radius k. Take A' on the ray OA such that OA·OA' = k2 and B' on the ray OB
such that OB·OB' = k2. If A'B' intersects the circle C, then the points of intersection give the positions where
PA + PB is a minimum. If not then the nearest point of C to A'B' (which is on the perpendicular bisector of
AB) gives the minimum.

For given P on C, the triangles OAP and OPA' are similar, so PA = (AO/PO) PA'. Similarly, PB = (BO/PO)
PB', so PA + PB = (AO/k) (PA' + PB'), so minimising PA + PB is equivalent to minimising PA' + PB'. If A'B'
intersects C, then clearly the points of intersection minimise. If not, let Q be the point on C closest to A'B'.
Let L be the tangent to the circle at that point (so that L is parallel to A'B'). Then QA + QB < RA + RB for
other points R on L. Given another point P on C, take the perpendicular to L through P. If it intersects L at R,
then PA + PB > RA + RB > QA + QB.

Comment. Ptolemy's theorem gives a partial solution OA·PB + OB·PA ≥ OP·AB with equality iff O, A, B, P
are concyclic or collinear. But OA = OB, so PA + PB ≥ k·AB/OA with equality iff P lies on the circle (or line)
OAB. Hence if the circle OAB (or line) intersects the circle C, then the two points of intersection give the
minimum. However, this does not deal with the case where the circles do not intersect. Moreover, it does not
completely deal with the case where they do, because one has to consider what happens if P is the opposite
side of the line OA to B.

13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002
13th Putnam 1953

Problem B7

Show that we can express any irrational number α ∈ (0, 1) uniquely in the form ∑1∞ (-1)n+1 1/(a1a2 ... an),
where ai is a strictly monotonic increasing sequence of positive integers. Find a1, a2, a3 for α = 1/√2.

Solution

Answer: a1 = 1, a2 = 3, a3 = 8.

Let sn be the sum of the first n terms. The terms alternate in sign and decrease in absolute value, so the odd
terms of the sequence sn decrease and the even terms increase. Every odd term exceeds every even term, so
the odds and the evens must each converge. But sn - sn+1 < 1/2n which tends to zero, so they tend to a
common limit.

Choose an as follows. Take a1 to be the smallest integer whose inverse exceeds α. Having chosen a2n-1, take
a2n to be the largest integer such that s2n < α. Having chosen a2n, take a2n+1 to be the largest integer such that
s2n+1 > α.

We have to show that these choices are always possible, or, in other words, that they yield a strictly
increasing sequence an. This depends on the relation 1/k - 1/k(k+1) = 1/(k+1) (*). For suppose we have just
chosen a2n-1. Then we know that s2n-1 > α, but that if increased a2n-1 by 1, then s2n-1 would be < α. Hence,
using (*), taking a2n = a2n-1 + 1 certainly gives s2n < α. On the other hand, if we take a2n to be sufficiently
large, then s2n will be close to s2n-1 and hence exceed α (note that α is irrational so it cannot equal any sm).
So the choice of a2n will exceed a2n-1. A similar argument shows that a2n+1 exceeds a2n.

So we have established that we can find a sequence an such that all the odd partial sums sn exceed α and all
the even partial sums are less than α. But we have also established that sn tends to a limit, so that limit must
be α. That establishes existence.

Suppose there is another expansion so that α = 1/a1 - 1/a1a2 + ... = 1/b1 - 1/b1b2 + ... . As above, we have that
1/(a1 + 1) < α < 1/a1 and also 1/(b1 + 1) < α < 1/b1. But since a1 and b1 are both integers that implies that a1
= b1. Suppose now that we have established that ai = bi for i ≤ n. Then we have that β = (-1)n a1 ... an(α - 1/a1
+ 1/a1a2 - ... + (-1)n/a1...an ) = 1/an+1 - 1/an+1an+2 + ... . But we also have β = 1/bn+1 - 1/bn+1bn+2 + ... . We
now argue as before that β lies between 1/(an+1 + 1) and 1/an+1 and also between 1/(bn+1 + 1) and 1/bn+1.
Hence an+1 = bn+1. That establishes uniqueness.

Finally, consider α = 1/√2. We have 1/2 < 1/√2 < 1, so a1 = 1. We must pick a2 as the largest integer so that 1
- 1/a2 < 1/√2, or a2 < 2 + √2 = 3.4. So a2 = 3. We must pick a3 as the largest integer so that 1 - 1/3 + 1/3a3 >
1/√2 or 2 + 1/x > 3/√2 or x < 3√2 + 4 = 8.2. So a3 = 8.
13th Putnam 1953

© John Scholes
[email protected]
5 Mar 2002