7th Putnam 1947

Problem A1

The sequence an of real numbers satisfies an+1 = 1/(2 - an). Show that limn→∞an = 1.

Solution

Fairly easy.

This is slightly messy. First, since k = 1/(2 - k) implies k = 1, it is obvious that if the sequence tends to a
limit, then the limit is 1.

Next, if 0 < an < 1, then 1 < 2 - an < 2, so 1/2 < an+1 < 1. So once the sequence gets into the interval (0, 1) it
stays there. But an+1 - an = (an - 1)2/(2 - an) > 0 for an < 2 (*). So once the sequence gets into the interval (0,
1) it is monotonic increasing and bounded above by 1, and hence tends to a limit (which must be 1).

If an < 0, then 0 < an+1 < 1, so we are also home if a member of the sequence is negative. Similarly, an > 2
implies an+1 < 0. If an = 1, then all following terms are 1 and so the limit is 1. So the only issue is if 1 < a1 <
2.

But then (*) shows that whilst the sequence remains in the interval (1, 2) it is monotonic increasing. It cannot
tend to a limit, because that limit would have to be 1, which is impossible. So it cannot stay in the interval.
We cannot have am = 2, because then we would not have am+1 = 1/(2 - am), so we must have am > 2 for some
m and then we are home.

The official solutions have a neater solution:

Let bn = 1/(1 - an). Then bn+1 = (2 - an)/(1 - an) = bn + 1. Hence for some β, bn = n + β, and hence an = (n + β
- 1)/(n + β). With a little tidying up, this is a solution.

7th Putnam 1947

© John Scholes
[email protected]
3 Nov 1999
7th Putnam 1947

Problem A2

R is the reals. f : R → R is continuous and satisfies f(r) = f(x) f(y) for all x, y, where r = √(x2 + y2). Show that
f(x) = f(1) to the power of x2.

Solution

Induction on n shows that f(x √n) = f(x)n, and hence f(n x) = f(x) to the power of n2. In particular, taking x =
1/n, f(1) = f(1/n) to the power of n2. Hence, provided f(1) is non-zero, f(1/n) = f(1) to the power of 1/n2.
Hence f(m/n) = f(1) to the power of (m/n)2. So we have established that f(x) = f(1) to the power of x2 for all
rational x. But f is continuous, so the relation holds for all x.

If f(1) = 0, then the same reasoning establishes that f(x) = 0 for all x.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem A3

ABC is a triangle and P an interior point. Show that we cannot find a piecewise linear path K = K1K2 ... Kn
(where each KiKi+1 is a straight line segment) such that: (1) none of the Ki do not lie on any of the lines AB,
BC, CA, AP, BP, CP; (2) none of the points A, B, C, P lie on K; (3) K crosses each of AB, BC, CA, AP, BP,
CP just once; (4) K does not cross itself.

Solution

Each time K crosses the boundary of a triangle it moves from the outside to the inside or vice versa. K has
two endpoints, so we can find one of the three triangles ABP, BCP, CAP in which it does not start or finish.
But that is impossible - on the first crossing it must go from outside to inside, on the second from inside to
outside and on the third from outide to inside.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem A4

Take the x-axis as horizontal and the y-axis as vertical. A gun at the origin can fire at any angle into the first
quadrant (x, y ≥ 0) with a fixed muzzle velocity v. Assuming the only force on the pellet after firing is gravity
(acceleration g), which points in the first quadrant can the gun hit?

Solution

Let the angle of the gun to the x-axis be θ. Then the equations of motion are: x = t v cos θ, y = t v sin θ - 1/2 g
t2. So the pellet moves along the parabola y = x tan θ - x2 g/2v2 sec2θ (*).

We can view (*) as an equation for θ given x, y. Put k = tan θ, then the equation becomes g/2v2 x2 k2 - x k +
y + gx2/2v2. This has real roots iff x2 ≥ 2g/v2 x2y - g2/v4 x4 and hence iff y ≤ 2g2/v2 - x2/v4 x2 (**). Since x
≥ 0, we see directly from the quadratic that the sum and the product of the roots are both non-negative, so
(**) is the condition for the equation to have at least one non-negative root in k and hence at least one root
for θ in the range 0 to π/2. Thus the gun can hit points in the first quadrant under (or on) the parabola given
by (**).

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem A5

The sequences an, bn, cn of positive reals satisfy: (1) a1 + b1 + c1 = 1; (2) an+1 = an2 + 2bncn, bn+1 = bn2 +
2cnan, cn+1 = cn2 + 2anbn. Show that each of the sequences converges and find their limits.

Solution

an+1 + bn+1 + cn+1 = (an + bn + cn)2, so by a trivial induction an + bn + cn = 1. There appears to be symmetry
between the three sequences, so we conjecture that each converges to 1/3.

Suppose an ≤ bn ≤ cn. We have an+1 = an2 + 2bncn ≤ ancn + bncn + cncn = cn(an + bn + cn) = cn. Similarly,
bn+1 = bn2 + 2ancn ≤ bncn + ancn + cn2 = cn, and cn+1 = cn2 + 2anbn ≤ cn2 + ancn + bncn = cn. Hence the
largest of an+1, bn+1, cn+1 is no bigger than the largest of an, bn, cn. An exactly similar argument works for
the smallest. Hence the largest forms a monotonic decreasing sequence which is bounded below and the
smallest forms a monotonic increasing sequence which is bounded above.

Let bn - an = h ≥ 0, cn - bn = k ≥ 0. Then an+1 - bn+1 = (an - bn)(an + bn - 2cn), so |an+1 - bn+1| = h(h + 2k) ≤ (h
+ k)2. Similarly, |bn+1 - cn+1| = |bn - cn| |bn + cn - 2an| = k(2h + k) ≤ (h + k)2, and |cn+1 - an+1| = |cn - an| |cn +
an - 2bn| = |(h + k)(k - h)| ≤ (h + k)2. So for n+1 the difference between the biggest and the smallest is the
square of the difference for n. But a1, b1, c1 are all positive and hence, by a trivial induction, an, bn, cn are
positive. Their sum is 1 so the difference between the biggest and smallest must be less than 1. Hence the
difference tends to zero. Hence an, bn, cn all tend to 1/3.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem A6

A is the matrix
a b c

d e f

g h i

det A = 0 and the cofactor of each element is its square (for example the cofactor of b is fg - di = b2). Show
that all elements of A are zero.

Solution

a2e2 - b2d2 = (ei - fh)(ai - cg) - (fg - di)(ch - bi) = (ae - bd) i2 + (cd - af) hi + (bf - ce) gi = (g3 + h3 + i3) i = 0,
since 0 = det A = g3 + h3 + i3. Hence ae = ±bd. Similarly cd = ±af, bf = ±ce. Multiplying the three equations
together we get abcdef = - abcdef unless at least one of the equations has a plus sign. In the first case, at least
one of a, b, c, d, e, f is zero. In the second case, the element corresponding to the cofactor is zero - for
example ae = bd implies i2 = 0 and hence i = 0. So either a member of the first two rows is zero, or a member
of the last row is zero.

wlog we may assume a = 0. That implies b or d = 0 also. [Note that if, for example, i was the zero element,
then we would have ei = ±fh, by an argument similar to that above and hence f or h = 0). If b = 0, then since
a3 + b3 + c3 = 0, we have also c = 0. Similarly, if d = 0, then g = 0. So we now have a complete row or
column zero. But now the square of any other element is a linear combination of elements in the that row or
column and hence zero. Suppose, for example, a = b = c = 0. Then g2 = bf - ce = 0, and similarly for the other
five elements.

Comment. This is surprisingly hard.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem B1

Let R be the reals. f : [1, ∞) → R is differentiable and satisfies f '(x) = 1/(x2 + f(x)2) and f(1) = 1. Show that
as x → ∞, f(x) tends to a limit which is less than 1 + π/4.

Solution

Clearly f '(x) is always positive. But f(1) = 1, so f(x) > 1 for all x. Hence f '(x) < 1/(x2 + 1) for all x. Hence
f(x) = 1 + ∫1x f '(t) dt < 1 + 1 + ∫1x 1/(1 + t2) dt = 1 + (tan-1t)|1x = 1 + tan-1x - π/4 < 1 + π/2 - π/4 = 1 + π/4.
Since f '(x) is positive, f(x) is monotone increasing. It is bounded above by 1 + π/4, so it must tend to a limit
less than 1 + π/4.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem B2

R is the reals. f :(0, 1) → R is differentiable and has a bounded derivative: |f '(x)| <= k. Prove that : |∫01 f(x)
dx - ∑1n f(i/n) /n| ≤ k/n.

Solution

The worst case for the difference between 1/n f(i/n) and ∫i/n-1/ni/n f(x) dx is if f '(x) = k (or -k) for the entire
range, in which case the difference is the area of a triangle base 1/n and height k/n. Hence the difference for
the complete Riemann sum is at worst k/(2n).

Comment. Note that the question gives a result which is needlessly too poor by a factor 2.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem B3

Let O be the origin (0, 0) and C the line segment { (x, y) : x ∈ [1, 3], y = 1 }. Let K be the curve { P : for
some Q ∈ C, P lies on OQ and PQ = 0.01 }. Let k be the length of the curve K. Is k greater or less than 2?

Solution

Answer: less.

If we use polar coordinates, then r = cosec θ - .01, so the length is ∫π/4arctan(1/3) √( (cosec θ - .01)2 +
cosec2θcot2θ) dθ. This is obviously horrendous.

The trick is that if we just remove the .01, then the integral gives the curve length for the line segment C,
which is 2. But the presence of the -.01 obviously reduces the integrand at every point of the range, so the
integral above must have value less than 2.

Comment. One hopes, of course, that the questioner has made it easy by taking the two endpoints a distance
greater than 2 apart, or maybe the sum of the distances from each endpoint to the middle greater than 2
(although that is already getting tiresome to check). But no. Or maybe the difference in x-coordinates + the
difference in y-coordinates for the endpoints is less than 2 (because the curve length is surely less). But no. I
wasted time on all those!

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem B4

p(z) ≡ z2 + az + b has complex coefficients. |p(z)| = 1 on the unit circle |z| = 1. Show that a = b = 0.

Solution

In particular, |p(1)| = |p(-1)| = 1, so 1 + a + b and 1 - a + b lie on the unit circle. Hence their midpoint 1 + b
lies in the unit disk. Similarly, |p(i)| = |p(-i)| = 1, so -1 + ia + b and -1 - ia + b lie on the unit circle and hence
their midpoint -1 + b lies in the unit disk. But 1 + b and -1 + b are a distance 2 apart, so they must lie at either
end of a diameter of the unit circle and hence b = 0. Now 1 + a and 1 - a lie on the unit circle, as does their
midpoint 1. Hence they must coincide and so a = 0.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem B5

Let p(x) be the polynomial (x - a)(x - b)(x - c)(x - d). Assume p(x) = 0 has four distinct integral roots and that
p(x) = 4 has an integral root k. Show that k is the mean of a, b, c, d.

Solution

a, b, c, d must be integers. (k - a)(k - b)(k - c)(k - d) = 4 and all of (k - a), (k - b), (k - c), (k - d) are integers.
They all divide 4, so they must belong to {-4, -2, -1, 1, 2, 4}. They are all distinct, so at most two of them
have absolute value 1. Hence none of them can have absolute value 4 - or their product would be at least 8.
Hence they must be -2, -1, 1, 2. Hence their sum is 0, so 4k = a + b + c + d.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002
7th Putnam 1947

Problem B6

P is a variable point in space. Q is a fixed point on the z-axis. The plane normal to PQ through P cuts the x-
axis at R and the y-axis at S. Find the locus of P such that PR and PS are at right angles.

Solution

Answer: sphere centre Q, radius QO, where O is the origin, excluding the two circles formed by the
intersection of the sphere with the y-z plane and the x-z plane.

Let Q be (0, 0, r). Let P be (a, b, c). If R is (x, 0, 0), then PR and PQ are perpendicular, so their dot product is
0, so a(a - x) + b2 + c(c - r) = 0, hence ax = (a2 + b2 + c2 - cr). Similarly, if S is (0, y, 0), then by = (a2 + b2 +
c2 - cr). We require PR and PS perpendicular so a(a - x) + (b - y)b + c2 = 0, hence ax + by = a2 + b2 + c2. So
a2 + b2 + c2 - 2cr = 0 and hence a2 + b2 + (c - r)2 = r2, which shows that P lies on the sphere centre Q radius
QO.

Conversely, suppose P lies on the sphere. Then SP and SO are tangents to the sphere and hence equal.
Similarly, RP = RO, so PRS and ORS are similar. Hence ∠RPS = ∠ROS = 90o. However, if P lies in the z-x
plane, then S is not a finite point and if P lies in the y-z plane, then R is not a finite point. So we must exclude
points lying on these two circles of the sphere.

7th Putnam 1947

© John Scholes
[email protected]
5 Mar 2002