HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

Putnam Solutions, 2007

These solutions are based on discussions among our contestants (and our friends from
Carlton College) Saturday afternoon and evening, with revisions as various errors were
caught by careful reviewers!

2007 A1

Find all values of α for which the curves y = αx2 + αx + 1/24 and x = αy2 + αy + 1/24 are
tangent to each other.

Solution. The reflection (x, y) à (y, x) interchanges the curves. If one curve lies
entirely on one side of the diagonal line y = x, then it cannot intersect its reflection, which
lies on the other side. Therefore, if these curves intersect at all, they will have a point of
intersection on the diagonal. This is enough to show that they cannot have a common
tangent at off-diagonal points, because a common tangent would be an intersection of
multiplicity at least two and, in conjunction with its mirror image, would imply the two
curves intersect in at least five points (counting multiplicity), an impossibility for two
polynomials of degree two. Therefore, we can restrict our search for mutually tangent
points to those of the form (t, t).

The slope of the first curve at (t, t) is 1/(dx/dy) = 1/(2αt + α) and the slope of the second
curve at (t, t) is dy/dx(t) = 2αt + α. For the two curves to be tangent to each other at (t, t),
we need these slopes to be the same. Whence

t = αt2 + αt + 1/24 and 1/(2αt + α) = 2αt + α, or 1 = (2αt + α)2.

Writing u = t + ½ gives

0 = α2 – (13/6)α + 4αu – 4(αu)2 and u2 = (1/2α)2.

There are two cases: u and α have the same or different signs; that is, either u = 1/(2α) or
u = -1/(2α).

(1) Same sign: 0 = α2 – (13/6)α + 1; solutions are (13 ± 5)/12 = {3/2, 2/3}.

(2) Different sign: 0 = α2 – (13/6)α – 3; solutions are (13 ± √601)/12.

Page 1/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

Comment. The figures illustrate the four solutions.

3 3
Reflection
Alpha=-0.96
2 Diagonal 2

1 1

0 0

-1 -1

-2 -2 Reflection
Alpha=03.126
Diagonal
-3 -3
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3

3 3

2 2

1 1

0 0

-1 -1

-2 Reflection -2 Reflection
Alpha=0.667 Alpha=01.5
Diagonal Diagonal
-3 -3
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3

2007 A2

Find the least possible area of a convex set in the

plane that intersects both branches of the hyperbola (-w,1/w)
xy = 1 and both branches of the hyperbola xy = -1. (x,1/x)

Solution. The answer is 4. Let X = (x, 1/x), Y =

(y, -1/y), Z = (-z, -1/z), and W = (-w, 1/w) be four of
the points of intersection, one per branch, with x, y, z,
and w all positive. The convex hull of these four (y,-1/y)
points lies inside the convex set and therefore cannot (-z,-1/z)

have greater area. We can further reduce the area if

Page 2/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

we can decrease the area of any of the triangles formed by three of these vertices, such as
XYW. That triangle has base YW. The value of x making the height minimum occurs
where the slope of the tangent, -1/x2, is parallel to the base; that is,

-1/x2 = (-1/y – 1/w) / (y + w).

This together with three similar formulas obtained from contemplating the other three
triangles implies x = y = z = w. The areas in all these cases are 4. Because there are no
other critical points and nothing is achieved by letting the variables approach zero or
infinity, this is the global minimum area.

2007 A3

Let k be a positive integer. Suppose that the integers 1, 2, 3, …, 3k + 1 are written down
in random order. What is the probability that at no time during this process, the sum of
the integers that have been written up to that time is a positive integer divisible by 3?
Your answer should be in closed form, but may include factorials.

Solution. The probability is (k+1)!k! / [(3k+1)(2k)!]. Because divisibility by three

is all that matters, we work modulo three. Evidently the positions of the zeros do not
change any of the divisibility properties. We ask, then, for the equivalent probability that
a random sequence of k zeros, k+1 ones, and k twos has no zero partial sums (modulo 3).

A sequence satisfying the conditions cannot begin with a zero. This leaves (2k+1)/(3k+1)
of the sequences to consider, but for them, the positions of the zeros do not matter. If a
partial sum is ever 1, the next value in the sequence cannot be 2, and if a partial sum is
ever 2, the next value cannot be 1. A straightforward induction shows the only possible
sequence (ignoring the zeros) is 1, 1, 2, 1, 2, …, 1, 2. There are (k+1)!k! such sequences
out of all (2k+1)! sequences and they are equally probable. Multiplying by (2k+1)/(3k+1)
produces the answer.

2007 A4

A repunit is a positive integer whose digits in base 10 are all ones. Find all polynomials f
with real coefficients such that if n is a repunit, then so is f(n).

Solution. The polynomials are all those in the form

f(x) = (10e(1 + 9x)d – 1) / 9, d = 1, 2, 3, …; e = 0, 1, 2, … .

The repunits are the integers (10k – 1)/9, k = 1, 2, 3, … . This inspires us to define
g(x) = 1 + 9f((x – 1)/9). The condition on f is equivalent to saying that for each positive
integer k, there exists a positive integer m such that g(10k) = 10m. As k grows arbitrarily
large, the leading term of g dominates, say gdxd, showing that m = kd + log10(gd),
implying g(x) = 10exd for integers d ≥ 1 and e. The stated solution is obtained from
f(x) = (g(1 + 9x) – 1)/9. The restriction e ≥ 0 is needed to make f(1) an integer (because
it’s a repunit), then an easy check shows all such f actually work.

Page 3/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

2007 A5

Suppose that a finite group has exactly n elements of order p, where p is a prime. Prove
that either n = 0, or p divides n + 1.

Solution. There are 1 + kp Sylow p-subgroups and, because they are all conjugate
and every element of order p is contained in some Sylow p-subroup, they partition the n
elements of order p into 1 + kp equal-size collections. The number of elements of order p
in any p-group is always one less than a power of p, implying n + 1 ≡ (1 + kp)(-1) + 1 ≡ 0
modulo p.

Comment. OK, it seems a bit like cheating to invoke the Sylow theorems. Let’s
provide a more detailed analysis. Let the group be G. We do not need to assume it is
finite, but we will suppose the subset A ⊂ G of order-p elements has a finite number n > 0
members. We are going to count A modulo p.

Pick any a ∈ A and define P to be the subgroup of G generated by a, so that P = {ε, a, a2,
…, ap-1}. G acts on A (that is, permutes its elements) via conjugation, because (gbg-1)p =
gbpg-1 = gg-1 = ε for any g in G and b in A. As a permutation of A, the generator a
consists of a collection of disjoint cycles, all of a size dividing p. Because p is prime, all
cycles must have length p or 1. This implies the orbits of the P-action on A each have
just one element (fixed points) or exactly p elements. Modulo p, the size of A equals the
number of fixed points.

A fixed point is an element of A that commutes with P. Suppose b and c are fixed points.
Then

a(bc) = (ab)c = (ba)c = b(ac) = b(ca) = (bc)a

implies bc commutes with a and therefore with P. Thus, the fixed points generate a
subgroup Z of G in which P is central (and therefore normal).

We can now proceed inductively. Let n be the smallest integer for which the proposition
fails. The canonical projection from Z to Z/P is well-defined as a group homomorphism,
because P is normal in Z. It is a p-to-one mapping which sends elements in Z of order p
to elements whose order is either p or 1. Because all elements of P map to the element of
order 1 in Z/P, and P has a nontrivial element (a), the number of elements of order p in
Z/P is less than n. Consequently the proposition is true in Z/P: either there are no
elements of order p or else the number of elements of order n, plus one, is a multiple of p.
In either case the number of elements of order n or one in Z (namely, the number of fixed
points plus one) is still a multiple of p. Therefore n + 1 is also a multiple of p, showing
our initial assumption about n is false: there can be no integers for which the proposition
fails, QED.

Here is an example to help make this less abstract. Let G be the group of permutations of
{1,2,3,4}. It contains nine elements of order 2: A = {(12), (13), (14), (23), (24), (34),
(12)(34), (13)(24), (14)(23)}. (This is cycle notation: (12) designates the permutation 1

Page 4/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

à 2 à 1, and implicitly 3 à 3 and 4à 4.) Let a = (12), so that P = {ε, (12)} is acting
on A. Conjugation by (12) is easy to compute: it exchanges the roles of 1 and 2 in the
permutation. For instance, a(13)a-1 = (23). The action of P on A is completely
determined by the action of (12) itself, and looks like this:

(12) à (12)
(13) à (23) à (13)
(14) à (24) à (14)
(34) à (34)
(12)(34) à (12)(34)
(13)(24) à (23)(14) à (13)(24).

Thus the fixed points are {(12), (34), (12)(34)}. The group they generate, Z, also
includes the identity. The canonical map from Z to Z/P sends (12) and ε to εP and sends
both (34) and (12)(34) to (34)P. Within Z/P there is one element of order 2 along with
the identity of order 1. The inverse image of each of these in Z consists of two elements,
for a total of four elements, including ε.

2007 A6

A triangulation T of a polygon P is a finite collection of triangles whose union is P, and

such that the intersection of any two triangles is either empty, or a shared vertex, or a
shared side. Moreover, each side of P is a side of exactly one triangle in T. Say that T is
admissible if every internal vertex is shared by 6 or more triangles. … Prove that there
is an integer Mn, depending only on n, such that any admissible triangulation of a polygon
P with n sides has at most Mn triangles.

Solution. Here is a sketch. The strategy is to decompose admissible triangulations

into smaller ones. There are two ways to do this. If a triangulation includes a diagonal of
the polygon, cut it along the diagonal. Otherwise, remove the polygon and all edges
adjacent to the polygon’s vertices, yielding a triangulation of the “derived” polygon.
This process eventually terminates in a triangulation with no diagonals and no interior
vertices, of which there are only three: the complete graphs on one, two, and three
vertices.

We can prove that the process of derivation (a) decreases the number of vertices on the
polygon by at least six and (b) removes a number of faces equal to the sum of the original
polygon vertices and the new polygon vertices. This is enough to show inductively that
the number of faces of an admissible triangulation never exceeds some crude bound such
as Mn = (n–2)2.

The hard part is showing (a). All the necessary information is encoded by the sequence
of degrees of vertices of P within its triangulation. Specifically, for any boundary vertex
v of P, let its “deficiency” d(v) = max(0, 6 – deg(v)). This is the minimum number of
additional edges needed to bring the degree of v up to 6.

Page 5/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

Let P be the derived polygon. Consider how to build up the original polygon in stages by
erecting edges externally on P. For the new triangulation to be admissible, there must be
at least d(v) edges attached to each vertex v. Proceeding around P, let v and w be two
successive vertices. One new face is formed by going from v to a new vertex, then to w,
then back along the edge of P from w to v. Every successive pair of vertices of P thereby
determines a unique new vertex (and the deficiency of such a vertex is 2). Furthermore,
at any vertex v where d(v) exceeds 2, at least d(v) – 2 additional edges must be erected
and therefore d(v) – 2 additional vertices must be created (and their deficiencies equal 3).
Let D(P) be the sum of d(v) – 2 as v ranges over the vertices of P. Whence, the outer
(original) polygon in the derivation has at least D(P) more vertices than P does. (This
argument breaks down when P is a single point, but that does not ruin the induction, as
we shall show.) Furthermore, there is a natural one-to-one correspondence between the
new faces and the edges of both P and the new outer polygon.

Therefore, all we need to show is that D(P) ≥ 6. This is an easy induction, because D(P)
new vertices of deficiency 3 are created at each stage and in the base cases of 1, 2, and 3
vertices, we compute D(P) = 4, 6, and 6, respectively. It looks like things fail for the case
of a single vertex, but we can fix this by considering all triangulations whose derived
polygon is a single point: clearly these consist of k ≥ 6 triangles sharing that point,
forming a k-vertex polygon. At each vertex v of that polygon d(v) = 3, implying D(P) = k
≥ 6. Thus the induction does go through as planned. Finally, it is straightforward to
check that when two triangulations are sewn together along an outer edge (the reverse of
splitting a triangulation along a diagonal of P), D(P) must increase.

2007 B1

Let f be a polynomial with positive integer coefficients. Prove that if n is a positive

integer, then f(n) divides f(f(n) + 1) if and only if n = 1.

Solution. Because f has positive integer coefficients, f(n) ≥ 1 for all positive integral
n. This means it makes sense to compute modulo f(n), where for any n we obtain

f(f(n) + 1) ≡ f(1) modulo f(n).

When n = 1 it is immediate that f(1) ≡ 0 modulo f(n).

Going in the other direction, this result also shows that if f(n) divides f(f(n) + 1)—that is,
f(f(n) + 1) ≡ 0 modulo f(n)—then f(n) divides f(1). Because f(n) and f(1) are both
positive, f(1) ≥ f(n). This would mean f(x) – f(1) has a (real) zero in the interval [1, ∞),
which is impossible for any polynomial all of whose coefficients are positive, QED.

Comment. In an example in The Art and Craft of Problem Solving (section 7.5), Paul
Zeitz shows that “if f(x) is a polynomial with integer coefficients, then for all integers n,
f(f(n) + n) is a multiple of f(n).”

Page 6/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

2007 B2
1
Suppose that f :[0, 1] à R has a continuous derivative and that ∫ f ( x)dx
0
= 0. Prove that
for every α ∈ (0, 1),
α 1
∫ f ( x )dx ≤ max f ' ( x ) .
0 8 0≤ x ≤1
x
Solution. Define F(x) = ∫0
f (t )dt and let a be a value in (0, 1) at which F2 is
maximized. The Fundamental Theorem of Calculus implies F2 is differentiable with
derivative 2Ff; its value at a—a critical point—must be zero. If F(a) = 0 there is nothing
left to show, because then F is identically zero, so we infer f(a) = 0. For any x, Taylor’s
Theorem asserts there is a number c between a and x for which

F(x) = F(a) + (x – a) f (a) + ½(c – a)2f '(a)

= F(a) + ½(c – a)2f '(a).

Let x = 0 and x = 1 in turn:

0 = F(0) = F(a) + ½(c – a)2f '(a) for c in (0, a) and

0 = F(1) = F(a) + ½(c – a)2f '(a) for c in (a, 1).

Taking the values of c that maximize (c – a)2 in the closures of these intervals (namely, 0
and 1) gives

|F(a)| ≤ ½(a)2f '(a) and |F(a)| ≤ ½(1 – a)2f '(a),

respectively. In at least one of the two expressions |1 – a| cannot exceed ½, yielding

|F(x)| = √F2(x) ≤ √F2(a) =|F(a)| ≤ ½(½)2| f '(a) | ≤ 1/8 max[0 < x < 1]| f '(x) |,

QED.

2007 B3

Let x = 1 and for n ≥ 0, let xn+1 = 3xn + xn√5. In particular, x1 = 5, x2 = 26, x3 = 136, and
x4 = 712. Find a closed-form expression for x2007. (a means the largest integer ≤ a.)

Solution. An answer is x2007 = (α2007 + β2007)/2 + (α2007 – β2007)/√5, where

α = 3 + √5 and β = 3 – √5 (the conjugate of α). More generally, we claim

xn = (αn + βn)/2 + (αn – βn)/√5 = f(n).

The proof is by induction, with the base case n = 0 an easy calculation. Assume then that
xn = (αn + βn)/2 + (αn – βn)/√5 for a particular n. By definition,

Page 7/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

xn+1 = 3xn + xn√5 = αxn

= α[(αn + βn)/2 + (αn – βn)/√5]
= (αn+1 + βn+1)/2 + (αn+1 – βn+1)/√5 + αβn/2 – αβn/√5 – βn+1/2 + βn+1/√5
= f(n+1) + βn(α – β)(½ – 1/√5)
= f(n+1) + βn(α – β)(½ – 1/√5).

The induction will be complete if we can show the second term is always zero; that is,
0 < βn(α – β)((½ – 1/√5) < 1 for all n. However, because 0 < β < 1,

βn(√5 – 2) < (√5 – 2) < 1, and

0 < βn(α – β)(½ – 1/√5) = βn(2√5)(½ – 1/√5) = βn(√5 – 2) < 1,

QED.

Comment. xn+1 = 6xn – 4xn-1 for all positive n. This follows from our formula for xn.

Comment. One can readily discover the formula for xn by making a small table of
powers of α and comparing them to known values of xn. To prepare for the calculations,
note that

(3 + √5)(a + b√5) = 3a + 5b + (3b + a)√5,

so that the coefficients (a, b) of the nth power determine the coefficients (3a+5b, 3b+a) of
the n+1st power.

n a b xn
0 1 0 1
1 3 1 5
2 14 6 26
3 72 32 136
4 376 168 712

The pattern xn = a + 2b is clear. To obtain a closed formula, we need formulas that

extract the coefficients a and b of αn. However, whenever αn = a + b√5, βn = a – b√5,
implying

a = (αn + βn)/2 and b = (αn – βn)/(2√5),

whence a + 2b = f(n).

Comment. Engel problem 14.72 (the last in the book) proves that “for every positive
integer n, 2 divides 1 + (3 +√5)n.” The solution introduces β, uses the vanishing size
of βn, and exploits the initial comment concerning the recursive relationship. Reinforcing
this is a similar problem (14.66) concerning powers of 3 +√11.

Page 8/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

2007 B4

Let n be a positive integer. Find the number of pairs P, Q of polynomials with real
coefficients such that

(P(X))2 + (Q(X))2 = X 2n + 1

and deg(P) > deg(Q).

Solution. There are 2n+1 solutions. Let ζ be a primitive 4nth root of unity (ζ2n = -1).
Use it to factor the right side and, simultaneously, factor the left side over the complex
numbers:

P(X)2 + Q(X)2 = ( R( X ))( R ( X )) = ∏ (X − ζ 2 j +1 )∏ (X − ζ ) where

n −1 n −1
2 j +1

j =0 j =0

R(X) = ±(P(X) + iQ(X)).

Because C[X] has unique factorization, ±R(X) is determined by choosing a subset S of

{0, 1, …, n-1} and selecting terms from the first product corresponding to S and terms
from the second product corresponding to the complement of S, thereby ensuring the
remaining terms are their complex conjugates. Moreover, ±R(X) completely determines
P and Q as its real and imaginary parts, respectively. The leading term is X 2n, which is
real and corresponds to P, so indeed P does have the higher degree.

The number of subsets is 2n and, independently of that, there are two ways to choose the
sign of R(X), giving 2*2n = 2n+1 solutions.

Comment. As a check, and to show how this works in a concrete case, a

straightforward calculation gives the 21+1 solutions (P(X), Q(X)) = (X, 1), (X, -1), (-X, 1),
and (-X, -1) when n = 1. They correspond to the factorization X 2 + 1 = (X – i)(X + i).

The case n = 2 is less trivial. Four of the 22+1 = 8 solutions come directly from the
factorization

X 4 + 1 = (X – ζ)(X – ζ3)(X – ζ7)(X – ζ5)

with ζ = (1 + i)/√2 (ζ7 is the conjugate of ζ and ζ5 is the conjugate of ζ3):

(X – ζ)(X – ζ3) = X 2 – 1 + i(√2X), P(X) = X 2 – 1, Q(X) = √2X,

(X – ζ)(X – ζ5) = X 2 + i(-1), P(X) = X 2, Q(X) = -1,
(X – ζ3)(X – ζ7) = X 2 + i(1), P(X) = X 2, Q(X) = 1,
(X – ζ7)(X – ζ5) = X 2 – 1 + i(-√2X) , P(X) = X 2 – 1, Q(X) = -√2X.

The other four solutions come from negating these.

Page 9/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

2007 B5

Let k be a positive integer. Prove that there exist polynomials P0(n), P1(n), …, Pk-1(n)
(which may depend on k) such that for any integer n,
k k −1
n  n n 
 k  = P0 ( n ) + P1 (n )  k  + L + Pk −1 ( n )  k  .

(a means the largest integer ≤ a.)

Solution. Fix k and for each n write n in the form n = n/kk + r with 0 ≤ r < k. Let
x = n/k. We seek polynomials for which

xk = P0(kx + r) + xP1(kx + r) + … + xk-1Pk-1(kx + r)

for all positive integers x and all r with 0 ≤ r < k.

We will look for a solution where the degrees of the polynomials do not exceed k-1, so
that we can write
k −1
Pi(z) = ∑p z
j =0
ij
j
.

Expand the powers of z = kx + r using the Binomial Theorem and arrange by ascending
powers of x. The constant terms (zeroth power) all come from P0(kx + r) and equal

p00 + p01r + p02r2 + … + p0k-1rk-1.

These must be zero for r = 0, 1, …, k=1. It is well-known that there is a unique solution
to this system of equations (using the theory of circulants or Vandermonde determinants),
implying P0 is completely determined.

Suppose inductively that P0, P1, …, Pj-1 have been determined. Examine the coefficients
of xj. Because we are effectively working modulo xj+1, there are no contributions from
Pj+1, …, Pk-1. There are k simultaneous equations for the coefficients of Pj of the form

pj0 + pj1r + … + pj,k-1rk-1 = terms from P’s of lower index (+1 when j = k).

Because this system is nonsingular, it has a unique solution, thereby determining Pj, and
thence by induction all the P’s are uniquely determined, QED.
k −1
 y −i
Really clever solution. Let f(x, y) = ∏  x −
i =0 k 
 , rewrite this in the form

n 
x k = P0 ( y ) + P1 ( y ) x + L + Pk −1 ( y ) x k −1 , and note that f (  , n ) = 0 for integral n.
k 

Page 10/11
HAVERFORD COLLEGE PROBLEM SOLVING GROUP December 2007

2007 B6

For each positive integer n, let f(n) be the number of ways to make n! cents using an
unordered collection of coins, each worth k! cents for some k, 1 ≤ k ≤ n. Prove that for
some constant C, independent of n,
2
/ 2−Cn − n 2 / 4 2
/ 2+Cn − n 2 / 4
nn e ≤ f (n ) ≤ n n e .

Solution. Here is just a sketch, because solving this one seems to require some
mathematical sophistication: generating functions, complex analysis, and asymptotic
analysis.

The generating function to partition arbitrary integers into pieces of size 1, 2, 6, ...,
through n! is F(t) where 1/F(t) = (1–t)(1–t2)(1–t6) ... (1–tn!). (The coefficient of tn! is
f(n).) The “biggest singularity” in its partial fraction expansion comes from the (1-t)n
term, so we expect this to dominate the growth. Its coefficient equals

Lim[t à1] (1-t)n F(t)

= 1/( Lim[t à1] (1+t)(1+t+...+t5)...(1+t+...+t[n!-1]) )
= 1/(2*6*24* ... * n!).

Let's call this function 1/g(n). Expanding the (1-t)n term with the Binomial Theorem
shows it contributes a value of |Comb(-n!, n)|/g(n) to f(n). By several applications of
Stirling’s Formula, the natural log of this value is asymptotically

log((n!+n–1)!) – log(n!!) – log(n!) – log(g(n))

= n (n log(n) – n) – Sum of (k log(k) – k), k from 1 to n

The usual techniques (compare to integrals) estimate the sum as

(n2 / 2 * log(n) – n2 / 4) – n2 / 2
= n2 / 2 * log(n) – 3 n2 / 4.

Subtracting and, as was implicit the foregoing, doing all calculations O(n log(n)), yields

log(f(n)) = n2 / 2 * log(n) – n2 / 4 + O(n log(n)),

which is exactly what we want.

A gap remains: showing that the other 2 + 6 + 24 + ... + n! singularities do not change the
asymptotics. The two poles of order n-1 (that is, at 1/(1+t)(n-1) and 1/(1-t)(n-1)) will
contribute terms only 1/n! as large; the six poles of order n-2 will each contribute terms
only 1/(n!(n!-1)) as large, and generally the k! poles of order n+1-k will each contribute
terms only 1/(n!(n!-1)...(n!-k+1)) as large. Summing these factors and making a crude
estimate shows the remaining poles contribute a truly inconsequential amount, so we’re
done.

Page 11/11