Professional Documents
Culture Documents
The 62nd William Lowell Putnam Mathematical Competition Saturday, December 1, 2001
The 62nd William Lowell Putnam Mathematical Competition Saturday, December 1, 2001
A-1 Consider a set and a binary operation , i.e., for each B-2 Find all pairs of real numbers 6/ QE
satisfying the sys-
, . Assume
for all tem of equations
. Prove that
for all .
" "
A-2 You have coins . For each , ! is bi- ( 6 / (SR Q
&RE/ ( Q
/ % Q
" "
of fallings heads. If the * coins are tossed, what is the 2 Q %$ Q 1 2T/ 1
U
probability that the number of heads is odd? Express / %
the answers as a rational function of * .
A-3 For each integer + , consider the polynomial B-3 For any positive integer * , let VW*YX denote the closest in-
,.- teger to Z * . Evaluate
&/
0 /$12) &%'+3(54
/ ( 6+725%
,8- [
For what values of + is &/
the product of two non- \ %E_ ` (M%aY_ ^`
constant polynomials with integer coefficients?
^]8 %
A-4 Triangle 9:; has an area 1. Points < => lie, respec-
tively, on sides :; , ?9 , 9?: such that 9?< bisects : =
at point @ , : = bisects > at point , and > bisects
B-4 Let
b denote the set of rational
hgi numbers
P different from
9< at point A . Find the area of the triangle @ A . 2;" D "dc . Define e7f by e. 6/ /j23"#'/ .
Prove or disprove that
A-5 Prove that there are unique positive integers , * such [
that
B8 2) (C"
%ED'DF" . k
eml ^n
o)p$
A-6 Can an arc of a parabola inside a circle of radius 1 have ]8
a length greater than 4?
B-1 Let * be an even positive integer. Write the numbers where e l n denotes e composed with itself * times.
" % G * in the squares of an *IHJ* grid so that the
-th row, from left to right, is B-5 Let and be real numbers in the interval 6D "#E%
,
and let q be a continuous real-valued function such
6K2L"
&*
()" 6;2M"
*N(M% O 6;2M"
*P(5* that qY rqs 6/
O
tu qs 6/
( / for all real / . Prove that
Color the squares of the grid so that half of the squares qs 6/
Cv / for some constant v .
in each row and in each column are red and the other
half are black (a checkerboard coloring is one possi- B-6 Assume that
$w. is an increasing sequence of pos-
bility). Prove that for each coloring, the sum of the itive real numbers such that xryrz F#'* D . Must
numbers on the red squares is equal to the sum of the there exist infinitely many positive integers * such that
numbers on the black squares. ( ^B % for } " % G *K2)" ?
a{ {|
A1 Let k be a fixed positive integer. The n-th derivative of Each player, in turn, signs his or her
1
xk −1
has the form (xkP−1)
n (x)
n+1 where Pn (x) is a polyno-
name on a previously unsigned face. The
mial. Find Pn (1). winner is the player who first succeeds in
signing three faces that share a common
A2 Given any five points on a sphere, show that some four vertex.
of them must lie on a closed hemisphere.
A3 Let n ≥ 2 be an integer and Tn be the number of non- Show that the player who signs first will always win by
empty subsets S of {1, 2, 3, . . . , n} with the property playing as well as possible.
that the average of the elements of S is an integer. Prove
that Tn − n is always even. B3 Show that, for all integers n > 1,
A4 In Determinant Tic-Tac-Toe, Player 1 enters a 1 in an n
1 1 1 1
empty 3 × 3 matrix. Player 0 counters with a 0 in a va- < − 1− < .
cant position, and play continues in turn until the 3 × 3 2ne e n ne
matrix is completed with five 1’s and four 0’s. Player
0 wins if the determinant is 0 and player 1 wins other- B4 An integer n, unknown to you, has been randomly
wise. Assuming both players pursue optimal strategies, chosen in the interval [1, 2002] with uniform probabil-
who will win and how? ity. Your objective is to select n in an odd number of
A5 Define a sequence by a0 = 1, together with the rules guesses. After each incorrect guess, you are informed
a2n+1 = an and a2n+2 = an + an+1 for each inte- whether n is higher or lower, and you must guess an
ger n ≥ 0. Prove that every positive rational number integer on your next turn among the numbers that are
appears in the set still feasibly correct. Show that you have a strategy so
that the chance of winning is greater than 2/3.
an−1 1 1 2 1 3
:n≥1 = , , , , ,... .
an 1 2 1 3 2 B5 A palindrome in base b is a positive integer whose base-
b digits read the same backwards and forwards; for ex-
A6 Fix an integer b ≥ 2. Let f (1) = 1, f (2) = 2, and ample, 2002 is a 4-digit palindrome in base 10. Note
for each n ≥ 3, define f (n) = nf (d), where d is the that 200 is not a palindrome in base 10, but it is the 3-
number of base-b digits of n. For which values of b does digit palindrome 242 in base 9, and 404 in base 7. Prove
∞ that there is an integer which is a 3-digit palindrome in
X 1
base b for at least 2002 different values of b.
n=1
f (n)
A2 Let
!!!"$# and % % &!!! % # be nonnegative
there are four ways: 4, 2+2, 1+1+2, 1+1+1+1.
holds identically?
,+ ' % -( ' % ( ' # % #(/. )"# ! h 72i !!! "#
l5#
jm2n by taking the averages of two con-
a new sequence of
Optheo entries,
secutive entries in the first sequence. Repeat the aver-
L
C M
H J #
B3 Show that for each positive integer n,
and
that
0 , 6such % 6 6 9A 0 0 C 6 9E 6 NG 0 ts$ vxw y 8z|{$ o !!!}
~ rP !
u
0 % =O P
A 0 0 E 7O PCQG 0 ! (Here 8-z denotes the least common ~ 6
y
denotes the greatest integer
6 .)
for all real numbers . Show that
multiple, and
' O '
is a lattice path of upsteps ( and B4 Let
downsteps 6 ( that starts at the origin R and never
'a
( M
h O %
g O A
9d
O O
A5 A Dyck -path
is a triangle with side lengths @ % A , and that the area of
' O ( depends only on the distance from to R .
Show that there is a one-to-one correspondence be-
tween the Dyck -paths with no return of even length
'a6 ( be a continuous real-valued function defined
this triangle
BT U ' ( V Y X S
and the Dyck -paths.
S W
' V B
V ( on the interval + J . . Show that
B6 Let
E
, , and
'T_ ( T` ' (
. Is it possible to
partition the nonnegative integers into two sets and C
in such a way that for all ?
A1 Basketball star Shanille O’Keal’s team statistician B1 Let P (x) = cn xn + cn−1 xn−1 + · · · + c0 be a poly-
keeps track of the number, S(N ), of successful free nomial with integer coefficients. Suppose that r is a
throws she has made in her first N attempts of the sea- rational number such that P (r) = 0. Show that the n
son. Early in the season, S(N ) was less than 80% of numbers
N , but by the end of the season, S(N ) was more than
80% of N . Was there necessarily a moment in between cn r, cn r2 + cn−1 r, cn r3 + cn−1 r2 + cn−2 r,
when S(N ) was exactly 80% of N ? . . . , cn rn + cn−1 rn−1 + · · · + c1 r
A2 For i = 1, 2 let Ti be a triangle with side lengths
ai , bi , ci , and area Ai . Suppose that a1 ≤ a2 , b1 ≤ are integers.
b2 , c1 ≤ c2 , and that T2 is an acute triangle. Does it
follow that A1 ≤ A2 ? B2 Let m and n be positive integers. Show that
A1 Show that every positive integer is a sum of one or more B2 Find all positive integers n, k1 , . . . , kn such that k1 +
numbers of the form 2r 3s , where r and s are nonneg- · · · + kn = 5n − 4 and
ative integers and no summand divides another. (For
example, 23 = 9 + 8 + 6.) 1 1
+ ··· + = 1.
k1 kn
A2 Let S = {(a, b)|a = 1, 2, . . . , n, b = 1, 2, 3}. A rook
tour of S is a polygonal path made up of line segments
connecting points p1 , p2 , . . . , p3n in sequence such that B3 Find all differentiable functions f : (0, ∞) → (0, ∞)
for which there is a positive real number a such that
(i) pi ∈ S,
a x
(ii) pi and pi+1 are a unit distance apart, for 1 ≤ i < f0 =
3n, x f (x)
(iii) for each p ∈ S there is a unique i such that pi = p.
for all x > 0.
How many rook tours are there that begin at (1, 1)
B4 For positive integers m and n, let f (m, n) denote the
and end at (n, 1)?
number of n-tuples (x1 , x2 , . . . , xn ) of integers such
(An example of such a rook tour for n = 5 was depicted that |x1 | + |x2 | + · · · + |xn | ≤ m. Show that f (m, n) =
in the original.) f (n, m).
A3 Let p(z) be a polynomial of degree n, all of whose ze- B5 Let P (x1 , . . . , xn ) denote a polynomial with real coef-
ros have absolute value 1 in the complex plane. Put ficients in the variables x1 , . . . , xn , and suppose that
g(z) = p(z)/z n/2 . Show that all zeros of g 0 (z) = 0
have absolute value 1.
∂2 ∂2
+ · · · + P (x1 , . . . , xn ) = 0 (identically)
A4 Let H be an n × n matrix all of whose entries are ±1 ∂x21 ∂x2n
and whose rows are mutually orthogonal. Suppose H
has an a × b submatrix whose entries are all 1. Show and that
that ab ≤ n.
x21 + · · · + x2n divides P (x1 , . . . , xn ).
A5 Evaluate
Z 1
ln(x + 1) Show that P = 0 identically.
dx.
0 x2 + 1
B6 Let Sn denote the set of all permutations of the numbers
1, 2, . . . , n. For π ∈ Sn , let σ(π) = 1 if π is an even
A6 Let n be given, n ≥ 4, and suppose that P1 , P2 , . . . , Pn
permutation and σ(π) = −1 if π is an odd permutation.
are n randomly, independently and uniformly, chosen
Also, let ν(π) denote the number of fixed points of π.
points on a circle. Consider the convex n-gon whose
Show that
vertices are Pi . What is the probability that at least one
of the vertex angles of this polygon is acute? X σ(π) n
= (−1)n+1 .
B1 Find a nonzero polynomial P (x, y) such that ν(π) + 1 n+1
π∈Sn
P (bac, b2ac) = 0 for all real numbers a. (Note:
bνc is the greatest integer less than or equal to ν.)
A1 Find the volume of the region of points (x, y, z) such B1 Show that the curve x3 + 3xy + y 3 = 1 contains only
that one set of three distinct points, A, B, and C, which are
vertices of an equilateral triangle, and find its area.
(x2 + y 2 + z 2 + 8)2 ≤ 36(x2 + y 2 ).
B2 Prove that, for every set X = {x1 , x2 , . . . , xn } of n
A2 Alice and Bob play a game in which they take turns real numbers, there exists a non-empty subset S of X
removing stones from a heap that initially has n stones. and an integer m such that
The number of stones removed at each turn must be one
less than a prime number. The winner is the player who
X 1
m + s ≤ .
takes the last stone. Alice plays first. Prove that there n+1
s∈S
are infinitely many n such that Bob has a winning strat-
egy. (For example, if n = 17, then Alice might take B3 Let S be a finite set of points in the plane. A linear
6 leaving 11; then Bob might take 1 leaving 10; then partition of S is an unordered pair {A, B} of subsets of
Alice can take the remaining stones to win.) S such that A ∪ B = S, A ∩ B = ∅, and A and B
lie on opposite sides of some straight line disjoint from
A3 Let 1, 2, 3, . . . , 2005, 2006, 2007, 2009, 2012, 2016, . . . S (A or B may be empty). Let LS be the number of
be a sequence defined by xk = k for k = 1, 2, . . . , 2006 linear partitions of S. For each positive integer n, find
and xk+1 = xk + xk−2005 for k ≥ 2006. Show that the the maximum of LS over all sets S of n points.
sequence has 2005 consecutive terms each divisible by
2006. B4 Let Z denote the set of points in Rn whose coordinates
are 0 or 1. (Thus Z has 2n elements, which are the
A4 Let S = {1, 2, . . . , n} for some integer n > 1. Say a vertices of a unit hypercube in Rn .) Given a vector sub-
permutation π of S has a local maximum at k ∈ S if space V of Rn , let Z(V ) denote the number of members
of Z that lie in V . Let k be given, 0 ≤ k ≤ n. Find
(i) π(k) > π(k + 1) for k = 1; the maximum, over all vector subspaces V ⊆ Rn of
(ii) π(k − 1) < π(k) and π(k) > π(k + 1) for 1 < dimension k, of the number of points in V ∩ Z.) [Ed-
k < n; itorial note: the proposers probably intended to write
(iii) π(k − 1) < π(k) for k = n. Z(V ) for “the number of points in V ∩ Z”, but this
changes nothing.]
(For example, if n = 5 and π takes values at 1, 2, 3, 4, 5
of 2, 1, 4, 5, 3, then π has a local maximum of 2 at k = B5 For each continuous function f : [0, 1] → R, let I(f ) =
R1 2 R1 2
1, and a local maximum of 5 at k = 4.) What is the 0
x f (x) dx and J(x) = 0 x (f (x)) dx. Find the
average number of local maxima of a permutation of S, maximum value of I(f ) − J(f ) over all such functions
averaging over all permutations of S? f.
A5 Let n be a positive odd integer and let θ be a real number B6 Let k be an integer greater than 1. Suppose a0 > 0, and
such that θ/π is irrational. Set ak = tan(θ + kπ/n), define
k = 1, 2, . . . , n. Prove that
1
a1 + a2 + · · · + an an+1 = an + √
k a
n
a1 a2 · · · an
for n > 0. Evaluate
is an integer, and determine its value.
ak+1
n
A6 Four points are chosen uniformly and independently at lim .
n→∞ nk
random in the interior of a given circle. Find the proba-
bility that they are the vertices of a convex quadrilateral.
A1 Find all values of α for which the curves y = αx2 + f (f (n) + 1) if and only if n = 1. [Editor’s note: one
1 1
αx + 24 and x = αy 2 + αy + 24 are tangent to each must assume f is nonconstant.]
other.
B2 Suppose that f : [0, 1] → R has a continuous derivative
A2 Find the least possible area of a convex set in the plane R1
and that 0 f (x) dx = 0. Prove that for every α ∈
that intersects both branches of the hyperbola xy = 1 (0, 1),
and both branches of the hyperbola xy = −1. (A set S
Z α
in the plane is called convex if for any two points in S
1
max |f ′ (x)|.
the line segment connecting them is contained in S.)
f (x) dx ≤
0 8 0≤x≤1
A3 Let k be a positive integer. Suppose that the integers √
1, 2, 3, . . . , 3k + 1 are written down in random order. B3 Let x0 = 1 and for n ≥ 0, let xn+1 = 3xn + ⌊xn 5⌋.
What is the probability that at no time during this pro- In particular, x1 = 5, x2 = 26, x3 = 136, x4 = 712.
cess, the sum of the integers that have been written up Find a closed-form expression for x2007 . (⌊a⌋ means
to that time is a positive integer divisible by 3? Your the largest integer ≤ a.)
answer should be in closed form, but may include fac- B4 Let n be a positive integer. Find the number of pairs
torials. P, Q of polynomials with real coefficients such that
A4 A repunit is a positive integer whose digits in base 10 (P (X))2 + (Q(X))2 = X 2n + 1
are all ones. Find all polynomials f with real coeffi-
cients such that if n is a repunit, then so is f (n). and deg P > deg Q.
A5 Suppose that a finite group has exactly n elements of B5 Let k be a positive integer. Prove that there exist poly-
order p, where p is a prime. Prove that either n = 0 or nomials P0 (n), P1 (n), . . . , Pk−1 (n) (which may de-
p divides n + 1. pend on k) such that for any integer n,
A6 A triangulation T of a polygon P is a finite collection j n kk jnk j n kk−1
of triangles whose union is P , and such that the inter- = P0 (n) + P1 (n) + · · · + Pk−1 (n) .
section of any two triangles is either empty, or a shared k k k
vertex, or a shared side. Moreover, each side is a side (⌊a⌋ means the largest integer ≤ a.)
of exactly one triangle in T . Say that T is admissible
if every internal vertex is shared by 6 or more triangles. B6 For each positive integer n, let f (n) be the number of
For example, [figure omitted.] Prove that there is an ways to make n! cents using an unordered collection of
integer Mn , depending only on n, such that any admis- coins, each worth k! cents for some k, 1 ≤ k ≤ n.
sible triangulation of a polygon P with n sides has at Prove that for some constant C, independent of n,
most Mn triangles.
2
/2−Cn −n2 /4 2
/2+Cn −n2 /4
nn e ≤ f (n) ≤ nn e .
B1 Let f be a polynomial with positive integer coefficients.
Prove that if n is a positive integer, then f (n) divides
A1 Let f : R2 → R be a function such that f (x, y) + (The elements of G in the sequence are not required to
f (y, z) + f (z, x) = 0 for all real numbers x, y, and z. be distinct. A subsequence of a sequence is obtained
Prove that there exists a function g : R → R such that by selecting some of the terms, not necessarily consec-
f (x, y) = g(x) − g(y) for all real numbers x and y. utive, without reordering them; for example, 4, 4, 2 is a
subsequence of 2, 4, 6, 4, 2, but 2, 2, 4 is not.)
A2 Alan and Barbara play a game in which they take turns
filling entries of an initially empty 2008 × 2008 array. B1 What is the maximum number of rational points that can
Alan plays first. At each turn, a player chooses a real lie on a circle in R2 whose center is not a rational point?
number and places it in a vacant entry. The game ends (A rational point is a point both of whose coordinates
when all the entries are filled. Alan wins if the determi- are rational numbers.)
nant of the resulting matrix is nonzero; Barbara wins if
it is zero. Which player has a winning strategy? B2 RLet F0 (x) = ln x. For n ≥ 0 and x > 0, let Fn+1 (x) =
x
0
Fn (t) dt. Evaluate
A3 Start with a finite sequence a1 , a2 , . . . , an of positive
integers. If possible, choose two indices j < k such
n!Fn (1)
that aj does not divide ak , and replace aj and ak by lim .
gcd(aj , ak ) and lcm(aj , ak ), respectively. Prove that if n→∞ ln n
this process is repeated, it must eventually stop and the B3 What is the largest possible radius of a circle contained
final sequence does not depend on the choices made. in a 4-dimensional hypercube of side length 1?
(Note: gcd means greatest common divisor and lcm
means least common multiple.) B4 Let p be a prime number. Let h(x) be a
polynomial with integer coefficients such that
A4 Define f : R → R by h(0), h(1), . . . , h(p2 − 1) are distinct modulo p2 .
( Show that h(0), h(1), . . . , h(p3 − 1) are distinct
x if x ≤ e modulo p3 .
f (x) =
xf (ln x) if x > e.
B5 Find all continuously differentiable functions f : R →
P∞ 1 R such that for every rational number q, the number
Does converge?
n=1 f (n) f (q) is rational and has the same denominator as q.
(The denominator of a rational number q is the unique
A5 Let n ≥ 3 be an integer. Let f (x) and g(x) be
positive integer b such that q = a/b for some integer a
polynomials with real coefficients such that the points
with gcd(a, b) = 1.) (Note: gcd means greatest com-
(f (1), g(1)), (f (2), g(2)), . . . , (f (n), g(n)) in R2 are
mon divisor.)
the vertices of a regular n-gon in counterclockwise or-
der. Prove that at least one of f (x) and g(x) has degree
B6 Let n and k be positive integers. Say that a permutation
greater than or equal to n − 1.
σ of {1, 2, . . . , n} is k-limited if |σ(i) − i| ≤ k for all
A6 Prove that there exists a constant c > 0 such that in ev- i. Prove that the number of k-limited permutations of
ery nontrivial finite group G there exists a sequence of {1, 2, . . . , n} is odd if and only if n ≡ 0 or 1 (mod
length at most c ln |G| with the property that each el- 2k + 1).
ement of G equals the product of some subsequence.
A–1 The hypothesis
implies
Note: a more sophisticated interpretation of this argu-
for all
(by replacing
!" by ),#and hence
$% ment can be given using Galois theory. Namely, if a
for all (using ). is neither a square nor
lm a lm
twice
a square, then the number
*),+.- fields d and d 3 are distinct quadratic fields,
A–2 Let &#' denote) the desired probability. Then &( ,
so their compositum is a number fieldf of degree 4,
and, for /0 ,
) whose Galois group acts transitively on b d a=b d 3 i .
21 3 / 1 )<; Thus & \ is irreducible.
& ' ) 6 & '879( 4 ):6 & '879(
3 /54 3 / 4 oCTp t
u eo uv
;=) ) A–4 Choose n so that qUr nMsUr r s
21 3 / p
) 6 & '879( 4 )?> rUs , and let w x
yz${ denote the area of triangle
3 /54 3 5/ 4 u u|t
x^yz . Then w sq{ w q{ since the tri-
+CB E-.+.F
The recurrence yields &A@ 3 , &#D , and by a angles
u
have
}
the
+
same
u
altitude 2);
and base. t Also
simple induction, one then checks that for general / one w sUq|{ sq sUr w sUr|{ n , and w~qUr {
+: ), + t<+ u| u );o[
has & ' / 3 /54 . qr sUr r r w sUr{ n (e.g., by the
law of sines). Adding this all up yields
Note: Richard Stanley points out the following
HJI 'KML H +:
nonin-
),
ductive argument. Put G HPO HP ( 4 3.N 3.N 4 ; )< u u t t
w sUq{:4=w s {k4=w~qUr {
then the coefficient of in G is the probability of R)$; R)$;o, ; ; o
getting exactly Q heads. Thus the desired number is 3 n 4n 3 n n
R),S; T;U),R+
G G 3 , and both values of G can be com- R) o,V) oCR) pp,R) )
),V)
puted directly: G , and or n 4 . Similarly 4 4n .
) - ;) ) _ Z
8 _:
T;U),$ 3 / Let H G `
)[w +:) HP w be theRfunction
m given by
G -XW BXWZY,Y[YCW ) ) > G 4 ; then G G G n n . However,
3 /54 3 /54 H H HP
HP^`_ H @
G is strictlyHdecreasing R
in , so G G is increas-
A–3 By the quadratic formula, if &]\ , then ing and H
G G G isHdecreasing.
R H
Thus there is at most
acb 3Cd 3 a one such that G G G ; in fact, since the equa-
egf b
4 3 , and hence the four roots of & \ are C ;) B.+
ahb d 3 i
given by d . If &#\ factors into two tion G has
op
a positive root 4 d 3 , we
nonconstant
polynomials over the integers, then some must have n .
subset of consisting of one or two elements form the u t u|t<+u u
We u
now compute+ w tU s
u
{
t
r
.+
w sUr{ ,
roots of a polynomial with integer coefficients. 3 , analogously
w sU!{ sU s w s {
u .+ 9 u ;
First suppose this subset has a single element, say w~sUr { w~r { 3 , and w~ { w sUr{
ajb d 3 u ; ; u [*);-.C+ ,E 7 D
d ; this element
must be a rational number. w sU!{ w~sUr { w~r { 3 .
Then d ahb d 3 @ 3 4 aEb 3Cd 3 a
is an integer, ) o[^)
@
so a is twice a perfect square, say a 3 / . But then Note: the key relation n 4 can also be de-
ab d 3 V b ), d 3 b )
rived by computing using homogeneous coordinates or
d
/ is only rational if / , i.e.,
if a 3 . vectors.
Next, suppose that the subset contains two elements; '#(;2 )[ ' _._8)
f f A–5 Suppose 4 3
. Notice that
then we can take it to be one of d a*b d 3ki , d 3 b
'#( [
) '
; )
a i f 4=w 4 { is a multiple of ; thus divides
d or b d a 4 d 3 i . In all cases, the sum and the 3
_C_
3
3 W
F
W
)C)
W
) -
[
.
product of the elements of the subset must be a ratio- Vl _C_:) h)
nal number. In the first case, this means 3 d a , Since
-.
3 is divisible by 3, we must
'#( have
) '
,
a , otherwise one of and 4 is a mul-
so is
l
a perfect square. In the second case, we have
, contradiction. In the third case, tiple of 3 and the other is not, so their difference cannot
%l we have
3 d 3 ' #( )X
-.
a @ el
d 4 d 3 , or a 4 3 4 3Cd 3 a , which be divisible
by 3. Now
)[ ' V)X ^-.
, so we must
means that a is twice a perfect square. have 4 , which forces / to be even,
H and in particular at least 2.
We conclude that & \ factors into two nonconstant '#( ;U ), ' ¡;U ), '
¢
polynomials over the integers if and only if a is either If is even, then ;U
4 4
), ' £;U)X
¢
.
a square or twice a square. Since / is even, 4 . Since
B–1 Let (resp. s ) denote the set of red (resp. black) H
- (±
),+ - (R± ;=),R+
o It follows that 4 3 and « 3
squares
o, ino[ such ) a coloring, and for Ás , let o
G /4Â 4 denote the number written in square , is the unique solution satisfying the given equations.
_Ã o, o[Ã ;)
where G
o,
 / . Then it is clear
o
that the ;%),+ @ @ ; )[+M¢ ),+ @
value of G o, depends only on the row of , while o
the B–3 Since N
)[+M¢
3 N N 4
and N 4 3
@
value of  depends only on the column of . Since N 4 N 4 , we have that ÍT/?Î N if and only if
www.adultpdf.com
@ ; )Ã Ã @
NÄ N 4 / N 4Ä N . Hence
Ä injective. Since  is also continuous,  is either strictly
increasing or strictly decreasing.
H
Moreover, Â cannot
'Ð 7 'Ð 'Ð 7 'Ð
½ 3CÏ 4 3 Ï ½ 3CÏ 4 3 Ï tendHPtoa;
finite limit ® as
H<*H
4 , or else we’d have
L
3 ' K L
É L K
3 ' Â Â Â , with the left side bounded and
' ( Ä ( : ' Ñ 'Ä Ð
ÏÉK Ò K K K the right side unbounded.
HÞ;
Similarly, Â cannot tend to a
7 finite limit as . Together with monotonicity,
½ 3 4 3
this yields that  is also surjective.
KÉL
Ä ( '
LK Ò K 3 ' H H eß
7 #(
K K K Ò K K Ò K
Pick ³ arbitrary, H
andHdefine
' for all /
_ H
recur-
½ 7 7 ; 7 7 sively by '#( Â ' for /0 , and ':79(
3 4 3 3 3 7#( H _ Þ
@
¢
R
+
KÉL Â ' for /jà . Let n¹( 4 d 4 3
Ä (
"; @ ¢ R+
KCÓÔK KCÓÔK and n @
H @ ; HX;e¥£_
d 4 3 and n @ be the roots of
_ )
½ 7 7 @Õ ; 7 @Õ , so that n ( 0
KÉL 3 3 gn,ã @ and
0á 0
Ä ( Ä nM( 0 n @ . Then there exist âÉ( â @ such that
H ' ' ^ß
K.ÓK K.ÓÔK ' â ( n ( 4â@,n @ for all / .
½ 7 7 @Õ ; ½ 7 7 @Õ ä
¦ _
KÉL 3 K L
É 3 Suppose  H is strictly H
increasing. If â,@ for some
'
-
( D choice of ³ , then ' is dominated H
by
H
n @ for / suffi-
> ciently negative. But taking ' and ' @ for _
/ suffi-
H
ciently
H negative
H of the
H right parity, we get à '= à
Alternate Ó the sum as solution: rewrite
Ó ' @ but  ' 0% ' @ , contradiction. Thus â @
Ö L ¦ Ö L _ H H H H
'½ ( )3
7 '
. Note that Ͳ/?Î Ï
' Ð Õ
4 '½ ( 3
7 ':7
Ï
'Ð Õ ; sinceH ³ âÉ( and ( âÉ(nM( , we have  n¹(
a @ a a for e all_ . Analogously, if  is strictly decreasing, then
Î if and only if /
Ͳ/Z4 4 for some . H '
;
Thus /4=ÍT/?Î and / Ͳ/?Î each increase by 1 except at â@ or else ' is H dominated H
by n ( for / sufficiently
a @
/ 4
a
, where the former skips from a @ 4 3 a to positive. But taking ' and ' @ _ for / H sufficiently H
pos-
a @ a a @ itive of the right parity, we get à ' @Ëà ' but
4 3 4 3 and the latter repeats the value . Thus H H
theÄ sums are Ä Ä Ä Â ' @ à ' , contradiction. Thus in that case,
HP¾ H H
 n,@ for all .
Ò Ò
½ 7P' ; ½ 7P\ ½ 7P' ½ 7P\ )=-
L 3 L 3 4 L 3 4 L 3 3 4 > B–6 Yes, there must exist infinitely many such
/ . Let
º_
be
³
' ( \ ( ' \ ( the convex hull of the set of points / ' for / .
+.Ø
Geometrically, is the intersection of all convex sets
B–4 For a rational number +.× ØM expressed
in Ø:lowest
terms, (or even all halfplanes)
containing the H#points
«
/ ' ;
define +.its height Ù × to be × 4 . Then for algebraically, is the
set
of points K K which
can
Øj
any × +.ØMR^ Ø expressed in lowest terms, we have be written
K
as â ( / ( :
' å 4 Y,Y[Y 4â /
' æ for some
@ ; @ Ø:
Ù G × × 4 × ; since by assumption â[( >[>,> â which are nonnegative of sum 1.
Ø *
¦ Ø:
× and are nonzero integers with × , we have We prove that for infinitely many / , / ' is a vertex
+CØMR; +CØM¾ Ø @ ; @ Ø:; ; Ø: on the upper boundary of , and that these
/ satisfy the
Ù G × Ù × × 4 × ×
º- : Ø ;= [;= Ø: given condition. The condition
that / ' is a vertex
4 × × on the upper boundary of is equivalent
to the exis-
[;), Ø:;)[ º
× 4 3 3 > tence of a line passing through / ' with all other _
Ó points of below it. That is, there should exist a 0
' ¾
It follows that G Õ consists solely of numbers of such that
height strictly larger than 3 /4 3 , and hence K a ; ç ºe)
Ó à ' 4 N / N > (1)
Ú L ' Õ ¾ Û
'½ ( G > ä)
We
K +
first
`_
show that
ä
/ satisfies (1).
K ; The condition
+: ;m),
_
N as N implies that
f K ;5 R+8 ;),
( N
Note: many choices for the height function are+.possible:
+.ØM
*^ÜMÝ| ÔM Ø: ØM as well. Thus the set ( N i has an upper
one can take Ù × × , or Ù × Ø equal K Ã= ;),
bound a , and now ( 4
a N , as desired.
to the total number of prime factors of × and , and so
on. The key properties of the height function are that Next, we show that given one / satisfying (1), there
on one hand, there are only finitely many rationals with exists a larger
K +
one_ also satisfying
(1). Again, the con-
K ;
height below any finite bound, and on the other hand, dition
R+: ;
N
Z _
as N
implies that
the height function is a sufficiently “algebraic” function ' N
f K ;Ì +: ;
/
K.è
as N . Thus the sequence
+CØ
of its argument that one can relate the heights of × ' N / i ' has a maximum element; sup-
+.ØM
and G × . pose N n is the largest value of N that achieves this
hCé!;¥ +: ;
a
H<
maximum, and put ' n / . Then the
K
« « HR< R é, a
B–5 Note that  H
«
implies that  Â
   line through n of slope lies strictly above N K
and hence from the given equation. That is, Â is for N 0*n and passes through or lies above N for
www.adultpdf.com
a ë). ;h) ;
N àhn . Thus (1) holds for / n with replaced by for Q >,>[> / , the points / Q '87 O and
a ;Ëê ê _
for suitably small 0 . /4mQ ' O lie below the line through / ' of; slope
a a a
. That means ' O à ' 4
Q and '87 O à
' Q;
By induction, we have that (1) holds for infinitely
_
many adding these together gives ':7 O 4 ' O à 3 ' , as de-
a sired.
/ . For any such / there exists 0 such that
www.adultpdf.com
Solutions to the 63rd William Lowell Putnam Mathematical Competition
Saturday, December 7, 2002
A1 By differentiating Pn (x)/(xk − 1)n+1 , we find that A4 (partly due to David Savitt) Player 0 wins with opti-
Pn+1 (x) = (xk − 1)Pn0 (x) − (n + 1)kxk−1 Pn (x); sub- mal play. In fact, we prove that Player 1 cannot prevent
stituting x = 1 yields Pn+1 (1) = −(n + 1)kPn (1). Player 0 from creating a row of all zeroes, a column of
Since P0 (1) = 1, an easy induction gives Pn (1) = all zeroes, or a 2 × 2 submatrix of all zeroes. Each of
(−k)n n! for all n ≥ 0. these forces the determinant of the matrix to be zero.
Note: one can also argue by expanding in Taylor series For i, j = 1, 2, 3, let Aij denote the position in row i
around 1. Namely, we have and column j. Without loss of generality, we may as-
sume that Player 1’s first move is at A11 . Player 0 then
1 1 1 plays at A22 :
= = (x − 1)−1 + · · · ,
xk − 1 k(x − 1) + · · · k
1 ∗ ∗
so ∗ 0 ∗
dn 1 (−1)n n! ∗ ∗ ∗
=
n k
dx x − 1 k(x − 1)−n−1 After Player 1’s second move, at least one of A23 and
and A32 remains vacant. Without loss of generality, assume
A23 remains vacant; Player 0 then plays there.
dn 1 After Player 1’s third move, Player 0 wins by playing at
Pn (x) = (xk − 1)n+1
dxn xk − 1 A21 if that position is unoccupied. So assume instead
= (k(x − 1) + · · · )n+1 that Player 1 has played there. Thus of Player 1’s three
(−1)n n!
moves so far, two are at A11 and A21 . Hence for i equal
(x − 1)−n−1 + · · · to one of 1 or 3, and for j equal to one of 2 or 3, the
k
following are both true:
= (−k)n n! + · · · .
(a) The 2 × 2 submatrix formed by rows 2 and i and
A2 Draw a great circle through two of the points. There are by columns 2 and 3 contains two zeroes and two
two closed hemispheres with this great circle as bound- empty positions.
ary, and each of the other three points lies in one of (b) Column j contains one zero and two empty posi-
them. By the pigeonhole principle, two of those three tions.
points lie in the same hemisphere, and that hemisphere
thus contains four of the five given points. Player 0 next plays at Aij . To prevent a zero column,
Player 1 must play in column j, upon which Player 0
Note: by a similar argument, one can prove that among completes the 2 × 2 submatrix in (a) for the win.
any n+3 points on an n-dimensional sphere, some n+2
of them lie on a closed hemisphere. (One cannot get by Note: one can also solve this problem directly by mak-
with only n + 2 points: put them at the vertices of a reg- ing a tree of possible play sequences. This tree can be
ular simplex.) Namely, any n of the points lie on a great considerably collapsed using symmetries: the symme-
sphere, which forms the boundary of two hemispheres; try between rows and columns, the invariance of the
of the remaining three points, some two lie in the same outcome under reordering of rows or columns, and the
hemisphere. fact that the scenario after a sequence of moves does
not depend on the order of the moves (sometimes called
A3 Note that each of the sets {1}, {2}, . . . , {n} has the “transposition invariance”).
desired property. Moreover, for each set S with in- Note (due to Paul Cheng): one can reduce Determi-
teger average m that does not contain m, S ∪ {m} nant Tic-Tac-Toe to a variant of ordinary tic-tac-toe.
also has average m, while for each set T of more than Namely, consider a tic-tac-toe grid labeled as follows:
one element with integer average m that contains m,
T \{m} also has average m. Thus the subsets other than A11 A22 A33
{1}, {2}, . . . , {n} can be grouped in pairs, so Tn − n is A23 A31 A12
even. A32 A13 A21
A5 It suffices to prove that for any relatively prime positive Again comparing an integral to a Riemann sum, we see
integers r, s, there exists an integer n with an = r and that for d ≥ 3,
an+1 = s. We prove this by induction on r + s, the case d
2X −1 2d
r + s = 2 following from the fact that a0 = a1 = 1.
Z
1 1 1 dx
Given r and s not both 1 with gcd(r, s) = 1, we must < d−1 − d +
n 2 2 2d−1 x
have r 6= s. If r > s, then by the induction hypothesis n=2d−1
we have an = r − s and an+1 = s for some n; then 1
= + log 2
a2n+2 = r and a2n+3 = s. If r < s, then we have 2d
an = r and an+1 = s − r for some n; then a2n+1 = r 1
≤ + log 2 < 0.125 + 0.7 < 1.
and a2n+2 = s. 8
Note: a related problem is as follows. Starting with the 1
Put c = +log 2 and L = 1+ 12 + 6(1−c)
8
1
. Then we can
sequence i
P2 −1 1
prove that n=1 f (n) < L for all i ≥ 2 by induction
0 1 on i. The case i = 2 is clear. For the induction, note
, ,
1 0 that by (2),
repeat the following operation: insert between each pair i
2X −1 i
a c a+c 1 1 1 X 1
b and d the pair b+d . Prove that each positive rational <1+ + +c
number eventually appears. f (n) 2 6 f (d)
n=1 d=3
Observe that by induction, if ab and dc are consecutive 1 1 1
terms in the sequence, then bc − ad = 1. The same <1+ + +c
2 6 6(1 − c)
holds for consecutive terms of the n-th Farey sequence, 1 1
the sequence of rational numbers in [0, 1] with denomi- =1+ + = L,
2 6(1 − c)
nator (in lowest terms) at most n.
P∞ 1
A6 The sum converges for b = 2 and diverges for b ≥ 3. as desired. We conclude that n=1 f (n) converges to a
We first consider b ≥ 3. Suppose the sum converges; limit less than or equal to L.
then the fact that f (n) = nf (d) whenever bd−1 ≤ n ≤ Note: the above argument proves that the sum for b = 2
bd − 1 yields is at most L < 2.417. One can also obtain a lower
d
bound by the same technique, namely 1 + 12 + 6(1−c 1
0)
∞ ∞ −1
bX
X 1 X 1 1 with c0 = log 2. This bound exceeds 2.043. (By con-
= . (1)
f (n) f (d) n trast, summing the first 100000 terms of the series only
n=1 d−1
d=1 n=b
yields a lower bound of 1.906.) Repeating the same ar-
However, by comparing the integral of 1/x with a Rie- guments with d ≥ 4 as the cutoff yields the upper bound
mann sum, we see that 2.185 and the lower bound 2.079.
d
bX −1 Z bd
B1 The probability is 1/99. In fact, we show by induc-
1 dx tion on n that after n shots, the probability of having
>
n bd−1 x made any number of shots from 1 to n − 1 is equal to
n=bd−1
1/(n − 1). This is evident for n = 2. Given the result
= log(bd ) − log(bd−1 ) = log b, for n, we see that the probability of making i shots after
n + 1 attempts is
where log denotes the natural logarithm. Thus (1) yields
∞ ∞
i−1 1 i 1 (i − 1) + (n − i)
1 1 + 1− =
n n−1 n n−1 n(n − 1)
X X
> (log b) ,
n=1
f (n) n=1
f (n) 1
= ,
n
a contradiction since log b > 1 for b ≥ 3. Therefore the
sum diverges. as claimed.
By taking logarithms, we can rewrite the desired in- We now prove by induction that am = b(2m + 1)/3c
equalities as and bm = b2m/3c for m ≥ 0. The inductive step relies
on the inequality bxc + byc ≤ bx + yc, with equal-
ity when one of x, y is an integer. Now suppose that
1 1
− log 1 − < −1 − n log 1 − ai = b(2i + 1)/3c and bi = b2i/3c for i < m. Then
2n n
1
2(k − 1)
2(m − k)
< − log 1 − . 1 + bk−1 + bm−k = 1 + +
n 3 3
Rewriting these in terms of the Taylor expansion of 2m
≤
− log(1−x), we see that the desired result is also equiv- 3
alent to
∞ ∞ ∞ and similarly ak−1 + am−k ≤ b(2m + 1)/3c, with
X 1 X 1 X 1
equality in both cases attained, e.g., when k = 1. The
i i
< i
< ,
i=1
i2 n i=1
(i + 1)n i=1
ini inductive formula for am and bm follows.
Note (due to Russ Mann): one can also give a “noncon- − (y p − xp−1 y)p−1 (z p − xp−1 z))
structive” argument. Let N be a large positive integer. 2 2
−p p
≡ (xy p − xp y)(z p − xp z )
For b ∈ (N 2 , N 3 ), the number of 3-digit base-b palin-
p p−1
dromes in the range [b2 , N 6 − 1] is at least − x(y − x y) (z − xp−1 z)
p p
2 2 2 2
−p+1 p p
6 ≡ xy p z p − xp yz p − xp y z + xp yz p
N − b2 N6
− 1 ≥ 2 − b − 2, 2 2
−p+1 p p 2 2
b b − xy p z p + xp y z + xp y p z − xp y p z
2 2 2
p−1 p−1
coefficient of z p y p x (since this term only appears in
Y Y the expansion of (3) when you choose the first term in
x (y + ix) (z + ix + jy). (3)
each factor). Thus the determinant is congruent to (3),
i=0 i,j=0
as desired.
We first check that Either argument can be used to generalize to a corre-
sponding n × n determinant, called a Moore determi-
p−1
Y nant; we leave the precise formulation to the reader.
(y + ix) ≡ y p − xp−1 y (mod p). (4) Note the similarity with the classical Vandermonde de-
i=0
terminant: if A is the n × n matrix with Aij = xji for
Since both sides are homogeneous as polynomials in x i, j = 0, . . . , n − 1, then
and y, it suffices to check (4) for x = 1, as a congru- Y
ence between polynomials. Now note that the right side det(A) = (xj − xi ).
has 0, 1, . . . , p − 1 as roots modulo p, as does the left 1≤i<j≤n
A1 There are n such sums. More precisely, there is exactly the left. Thus AM-GM shows that the terms under con-
one such sum with k terms for each of k = 1, . . . , n sideration on the left exceed those on the right; adding
(and clearly no others). To see this, note that if n = these inequalities over all k yields the desired result.
a1 + a2 + · · · + ak with a1 ≤ a2 ≤ · · · ≤ ak ≤ a1 + 1, Third solution: Since both sides are continuous in each
then ai , it is sufficient to prove the claim with a1 , . . . , an all
positive (the general case follows by taking limits as
ka1 = a1 + a1 + · · · + a1 some of the ai tend to zero). Put ri = bi /ai ; then the
≤ n ≤ a1 + (a1 + 1) + · · · + (a1 + 1) given inequality is equivalent to
= ka1 + k − 1.
(1 + r1 )1/n · · · (1 + rn )1/n ≥ 1 + (r1 · · · rn )1/n .
However, there is a unique integer a1 satisfying these
inequalities, namely a1 = bn/kc. Moreover, once a1 In terms of the function
is fixed, there are k different possibilities for the sum
a1 + a2 + · · · + ak : if i is the last integer such that f (x) = log(1 + ex )
ai = a1 , then the sum equals ka1 + (i − 1). The pos-
and the quantities si = log ri , we can rewrite the de-
sible values of i are 1, . . . , k, and exactly one of these
sired inequality as
sums comes out equal to n, proving our claim.
Note: In summary, there is a unique partition of n with 1 s1 + · · · + sn
(f (s1 ) + · · · + f (sn )) ≥ f .
k terms that is “as equally spaced as possible”. One n n
can also obtain essentially the same construction induc-
tively: except for the all-ones sum, each partition of n This will follow from Jensen’s inequality if we can ver-
is obtained by “augmenting” a unique partition of n−1. ify that f is a convex function; it is enough to check that
f 00 (x) > 0 for all x. In fact,
A2 First solution: Assume without loss of generality that
ai + bi > 0 for each i (otherwise both sides of the de- ex 1
f 0 (x) = =1−
sired inequality are zero). Then the AM-GM inequality 1 + ex 1 + ex
gives
is an increasing function of x, so f 00 (x) > 0 and
a1 · · · an
1/n Jensen’s inequality thus yields the desired result. (As
long as the ai are all positive, equality holds when
(a1 + b1 ) · · · (an + bn ) s1 = · · · = sn , i.e., when the vectors (a1 , . . . , an ) and
1 a1 an (b1 , . . . , bn ). Of course other equality cases crop up if
≤ + ··· + ,
n a1 + b1 an + bn some of the ai vanish, i.e., if a1 = b1 = 0.)
Fourth solution: We apply induction on n, the case
and likewise with the roles of a and b reversed. Adding
n = 1 being evident. First we verify the auxiliary in-
these two inequalities and clearing denominators yields
equality
the desired result.
Second solution: Write the desired inequality in the (an + bn )(cn + dn )n−1 ≥ (acn−1 + bdn−1 )n
form
for a, b, c, d ≥ 0. The left side can be written as
(a1 +b1 ) · · · (an +bn ) ≥ [(a1 · · · an )1/n +(b1 · · · bn )1/n ]n ,
an cn(n−1) + bn dn(n−1)
expand both sides, and compare the terms on both n−1
X n − 1
sides in which k of the terms are among the + an cni dn(n−1−i)
ai . On the left, one has the product of each k- i=1
i
element subset of {1, . . . , n}; on the right, one has n−1
X n − 1
n
bn cn(n−i) dn(i−1) .
k/n
k (a1 · · · an ) · · · (b1 . . . bn )(n−k)/n , which is pre- +
n i − 1
cisely k times the geometric mean of the terms on i=1
to obtain
implies that there is a triangle whose sides, as vec- Z 1Z 1 Z 1
tors, correspond to the complex numbers x − 1, ω(x − |f (x) + f (y)| dx dy − |f (x)| dx
ω), ω 2 (x − ω 2 ); this triangle has sides of length a, b, c. 0 0 0
≥ max{(µ+ − µ− )(I+ + I− ) + 2µ− (I− − I+ ),
To calculate the area of this triangle, we first note a more
general formula. If a triangle in the plane has vertices (µ+ − µ− )(I+ − I− ),
at 0, v1 = s1 + it1 , v2 = s2 + it2 , then it is well (µ− − µ+ )(I+ + I− ) + 2µ+ (I+ − I− )}.
known that the area of the triangle is |s1 t2 − s2 t1 |/2 =
|v1 v2 − v2 v1 |/4. In our case, we have v1 = x − 1 and Now simply note that for each of the possible compar-
v2 = ω(x − ω); then isons between µ+ and µ− , and between I+ and I− , one
of the three terms above is manifestly nonnegative. This
√ yields the desired result.
v1 v2 − v2 v1 = (ω 2 − ω)(xx − 1) = i 3(|x|2 − 1).
Second solution: We will show at the end that it is
√
Hence the area of the triangle is 3(1 − |x|2 )/4, which enough to prove a discrete analogue: if x1 , . . . , xn are
depends only on the distance |x| from P to O. real numbers, then
n n
1 X 1X
B6 First solution: (composite of solutions by Feng Xie |xi + xj | ≥ |xi |.
and David Pritchard) Let µ denote Lebesgue measure n2 i,j=1 n i=1
A1 Yes. Suppose otherwise. Then there would be an N is acute), we have sin ∠P1 ≤ sin ∠P2 . By the Law of
such that S(N ) < 80% and S(N + 1) > 80%; that is, Sines,
O’Keal’s free throw percentage is under 80% at some 1 1
point, and after one subsequent free throw (necessarily A1 = b1 c1 sin ∠P1 ≤ b2 c2 sin ∠P2 = A2 .
2 2
made), her percentage is over 80%. If she makes m of
her first N free throws, then m/N < 4/5 and (m + Remark: Many other solutions are possible; for in-
1)/(N + 1) > 4/5. This means that 5m < 4n < stance, one uses Heron’s formula for the area of a tri-
5m + 1, which is impossible since then 4n is an integer angle in terms of its side lengths.
between the consecutive integers 5m and 5m + 1.
A3 Define a sequence vn by vn = (n − 1)(n − 3) · · · (4)(2)
Remark: This same argument works for any fraction if n is odd and vn = (n − 1)(n − 3) · · · (3)(1) if n is
of the form (n − 1)/n for some integer n > 1, but not even; it suffices to prove that un = vn for all n ≥ 2.
for any other real number between 0 and 1. Now vn+3 vn = (n + 2)(n)(n − 1)! and vn+2 vn+1 =
A2 First solution: (partly due to Ravi Vakil) Yes, it does (n + 1)!, and so vn+3 vn − vn+2 vn+1 = n!. Since we
follow. For i = 1, 2, let Pi , Qi , Ri be the vertices of Ti can check that un = vn for n = 2, 3, 4, and un and vn
opposide the sides of length ai , bi , ci , respectively. satisfy the same recurrence, it follows by induction that
un = vn for all n ≥ 2, as desired.
We first check the case where a1 = a2 (or b1 = b2 or
c1 = c2 , by the same argument after relabeling). Imag- A4 It suffices to verify that
ine T2 as being drawn with the base Q2 R2 horizontal x1 · · · xn
and the point P2 above the line Q2 R2 . We may then
position T1 so that Q1 = Q2 , R1 = R2 , and P1 lies 1 X
= n (e1 · · · en )(e1 x1 + · · · + en xn )n .
above the line Q1 R1 = Q2 R2 . Then P1 also lies inside 2 n!
ei ∈{−1,1}
the region bounded by the circles through P2 centered
at Q2 and R2 . Since ∠Q2 and ∠R2 are acute, the part To check this, first note that the right side vanishes iden-
of this region above the line Q2 R2 lies within T2 . In tically for x1 = 0, because each term cancels the corre-
particular, the distance from P1 to the line Q2 R2 is less sponding term with e1 flipped. Hence the right side, as
than or equal to the distance from P2 to the line Q2 R2 ; a polynomial, is divisible by x1 ; similarly it is divisible
hence A1 ≤ A2 . by x2 , . . . , xn . Thus the right side is equal to x1 · · · xn
times a scalar. (Another way to see this: the right side is
To deduce the general case, put
clearly odd as a polynomial in each individual variable,
r = max{a1 /a2 , b1 /b2 , c1 /c2 }. but the only degree n monomial in x1 , . . . , xn with that
property is x1 · · · xn .) Since each summand contributes
1
Let T3 be the triangle with sides ra2 , rb2 , rc2 , which 2n x1 · · · xn to the sum, the scalar factor is 1 and we are
has area r2 A2 . Applying the special case to T1 and T3 , done.
we deduce that A1 ≤ r2 A2 ; since r ≤ 1 by hypothesis, Remark: Several variants on the above construction are
we have A1 ≤ A2 as desired. possible; for instance,
Remark: Another geometric argument in the case a1 = x1 · · · xn
a2 is that since angles ∠Q2 and ∠R2 are acute, the per-
1 X
pendicular to Q2 R2 through P2 separates Q2 from R2 . = (−1)n−e1 −···−en (e1 x1 + · · · + en xn )n
If A1 > A2 , then P1 lies above the parallel to Q2 R2 n!
ei ∈{0,1}
through P2 ; if then it lies on or to the left of the vertical
line through P2 , we have c1 > c2 because the inequality by the same argument as above.
holds for both horizontal and vertical components (pos- Remark: These construction work over any field of
sibly with equality for one, but not both). Similarly, if characteristic greater than n (at least for n > 1). On the
P1 lies to the right of the vertical, then b1 > b2 . other hand, no construction is possible over a field of
Second solution: (attribution unknown) Retain nota- characteristic p ≤ n, since the coefficient of x1 · · · xn
tion as in the first paragraph of the first solution. Since in the expansion of (e1 x1 + · · · + en xn )n is zero for any
the angle measures in any triangle add up to π, some ei .
angle of T1 must have measure less than or equal to its Remark: Richard Stanley asks whether one can use
counterpart in T2 . Without loss of generality assume fewer than 2n terms, and what the smallest possible
that ∠P1 ≤ ∠P2 . Since the latter is acute (because T2 number is.
A5 First solution: First recall that any graph with n ver- – With probability 1/4, the i-th square is opposite in
tices and e edges has at least n − e connected com- color from the adjacent squares directly above and
ponents (add each edge one at a time, and note that to the left of it. In this case adding the i-th square
it reduces the number of components by at most 1). adds one component.
Now imagine the squares of the checkerboard as a
– With probability 1/8, the i-th square is the same
graph, whose vertices are connected if the correspond-
in color as the adjacent squares directly above and
ing squares share a side and are the same color. Let A
to the left of it, but opposite in color from its diag-
be the number of edges in the graph, and let B be the
onal neighbor above and to the left. In this case,
number of 4-cycles (formed by monochromatic 2 × 2
adding the i-th square either removes a component
squares). If we remove the bottom edge of each 4-cycle,
or leaves the number unchanged.
the resulting graph has the same number of connected
components as the original one; hence this number is at – In all other cases, the number of components re-
least mains unchanged upon adding the i-th square.
Suppose the i-th square does not abut the left edge or for
the top row of the board. Then we may divide into three
cases. F (x, y, z, w) = f (x, y) + f (z, w) − f (x, w) − f (z, y).
Remark: (by Po-Ning Chen) The discrete inequality (m + n)!/(m + n)m+n . On the other hand, if p is the
can be arrived at more systematically by repeatedly ap- probability of picking exactly m red balls, then p < 1
plying the following identity: for any real a1 , . . . , an , and the probability of picking each ball exactly once is
!2 p(mm /m!)(nn /n!).
n n
X X X Second solution: (by David Savitt) Define
(xi − xj )2 = n x2i − xi .
1≤i<j≤n i=1 i=1
Sk = {i/k : i = 1, . . . , k}
Remark: (by David Savitt) The discrete inequality can and rewrite the desired inequality as
also be interpreted as follows. For c, d ∈ {1, . . . , n−1}
and ζn = e2πi/n , put
Y Y Y
x y> z.
X x∈Sm y∈Sn z∈Sm+n
zc,d = ζnci+dj xij .
i,j To prove this, it suffices to check that if we sort the
multiplicands on both sides into increasing order, the i-
Then the given inequality is equivalent to th term on the left side is greater than or equal to the i-th
n−1 term on the right side. (The equality is strict already for
i = 1, so you do get a strict inequality above.)
X
|zc,d |2 ≥ 0.
c,d=1 Another way to say this is that for any i, the number of
factors on the left side which are less than i/(m + n) is
B1 Let k be an integer, 0 ≤ k ≤ n−1. Since P (r)/rk = 0, less than i. But since j/m < i/(m + n) is equivalent
we have to j < im/(m + n), that number is
cn rn−k + cn−1 rn−k+1 + · · · + ck+1 r im in
−1+ −1
= −(ck + ck−1 r−1 + · · · + c0 r−k ). m+n m+n
im in
≤ + − 1 = i − 1.
Write r = p/q where p and q are relatively prime. Then m+n m+n
the left hand side of the above equation can be written as
a fraction with denominator q n−k , while the right hand Third solution: Put f (x) = x(log(x+1)−log x); then
side is a fraction with denominator pk . Since p and q for x > 0,
are relatively prime, both sides of the equation must be
an integer, and the result follows. 1
f 0 (x) = log(1 + 1/x) −
Remark: If we write r = a/b in lowest terms, then x+1
P (x) factors as (bx − a)Q(x), where the polynomial 1
f 00 (x) = − .
Q has integer coefficients because you can either do the x(x + 1)2
long division from the left and get denominators divis-
ible only by primes dividing b, or do it from the right Hence f 00 (x) < 0 for all x; since f 0 (x) → 0 as x → ∞,
and get denominators divisible only by primes dividing we have f 0 (x) > 0 for x > 0, so f is strictly increasing.
a. The numbers given in the problem are none other Put g(m) = m log m − log(m!); then g(m + 1) −
than a times the coefficients of Q. More generally, if g(m) = f (m), so g(m + 1) − g(m) increases with
P (x) is divisible, as a polynomial over the rationals, by m. By induction, g(m + n) − g(m) increases with n
a polynomial R(x) with integer coefficients, then P/R for any positive integer n, so in particular
also has integer coefficients; this is known as “Gauss’s
lemma” and holds in any unique factorization domain. g(m + n) − g(m) > g(n) − g(1) + f (m)
≥ g(n)
B2 First solution: We have
m+n
since g(1) = 0. Exponentiating yields the desired in-
m+n
(m + n) > mm n n equality.
m
B3 The answer is {a | a > 2}. If a > 2, then the func-
because the binomial expansion of (m + n)m+n in- tion f (x) = 2a/(a − 2) has the desired property; both
cludes the term on the right as well as some others. Re- perimeter and area of R in this case are 2a2 /(a − 2).
arranging this inequality yields the claim. Now suppose that a ≤ 2, and let f (x) be a nonnegative
Remark: One can also interpret this argument com- continuous function on [0, a]. Let P = (x0 , y0 ) be a
binatorially. Suppose that we choose m + n times point on the graph of f (x) with maximal y-coordinate;
(with replacement) uniformly randomly from a set of then the area of R is at most ay0 since it lies below the
m + n balls, of which m are red and n are blue. Then line y = y0 . On the other hand, the points (0, 0), (a, 0),
the probability of picking each ball exactly once is and P divide the boundary of R into three sections. The
length of the section between (0, 0) and P is at least the since limN →∞ (xN ln(1 + xN +1 )) = 0 for 0 < x < 1,
distance between (0, 0) and P , which is at least y0 ; the we conclude that L = − ln 2 + limx→1− f (x), where
length of the section between P and (a, 0) is similarly ∞
at least y0 ; and the length of the section between (0, 0)
X
f (x) = (1/x − 1) xn ln(1 + xn )
and (a, 0) is a. Since a ≤ 2, we have 2y0 + a > ay0 n=1
and hence the perimeter of R is strictly greater than the X∞ X∞
area of R. = (1/x − 1) (−1)m+1 xn+mn /m.
n=1 m=1
B4 First solution: Identify the xy-plane with the complex
plane C, so that Pk is the real number k. If z is sent to This final double sum converges absolutely when 0 <
z 0 by a counterclockwise rotation by θ about Pk , then x < 1, since
z 0 − k = eiθ (z − k); hence the rotation Rk sends z to ∞ X
∞ ∞
ζz + k(1 − ζ), where ζ = e2πi/n . It follows that R1
X X
xn+mn /m = xn (− ln(1 − xn ))
followed by R2 sends z to ζ(ζz + (1 − ζ)) + 2(1 − ζ) = n=1 m=1 n=1
ζ 2 z + (1 − ζ)(ζ + 2), and so forth; an easy induction X∞
shows that R sends z to < xn (− ln(1 − x)),
n=1
ζ n z + (1 − ζ)(ζ n−1 + 2ζ n−2 + · · · + (n − 1)ζ + n).
which converges. (Note that − ln(1 − x) and − ln(1 −
Expanding the product (1 − ζ)(ζ n−1 + 2ζ n−2 + · · · + xn ) are positive.) Hence we may interchange the sum-
(n − 1)ζ + n) yields −ζ n − ζ n−1 − · · · − ζ + n = mations in f (x) to obtain
n. Thus R sends z to z + n; in cartesian coordinates, ∞ X ∞
R(x, y) = (x + n, y).
X (−1)m+1 x(m+1)n
f (x) = (1/x − 1)
m
Second solution: (by Andy Lutomirski, via Ravi Vakil) m=1 n=1
∞
Imagine a regular n-gon of side length 1 placed with its (−1)m+1 xm (1 − x)
X
top edge on the x-axis and the left endpoint of that edge = (1/x − 1) .
m=1
m 1 − xm+1
at the origin. Then the rotations correspond to rolling
this n-gon along the x-axis; after the n rotations, it This last sum converges absolutely uniformly in x, so
clearly ends up in its original rotation and translated n it is legitimate to take limits term by term. Since
units to the right. Hence the whole plane must do so as xm 1−x 1
limx→1− 1−x m+1 = m+1 for fixed m, we have
well.
∞
Third solution: (attribution unknown) Viewing each
X (−1)m+1
lim f (x) =
Rk as a function of a complex number z as in the first x→1−
m=1
m(m + 1)
solution, the function Rn ◦ Rn−1 ◦ · · · ◦ R1 (z) is linear ∞
1 1
in z with slope ζ n = 1. It thus equals z + T for some
X
m+1
= (−1) −
T ∈ C. Since f1 (1) = 1, we can write 1 + T = Rn ◦ m=1
m m+1
· · · ◦ R2 (1). However, we also have ∞
!
X (−1)m+1
=2 −1
Rn ◦ · · · ◦ R2 (1) = Rn−1 ◦ R1 (0) + 1 m=1
m
B6 First solution: (based on a solution of Dan Bernstein) of (b) with n replaced by n + 1 holds for en+1 6= 0; of
Note that for any b, the condition that b ∈ / B already course it holds for en+1 = 0 because (b) was already
forces lim sup N (x)/x to be at most 1/2: pair off 2mb+ known. Since the analogue of (a) holds by construction,
n with (2m + 1)b + n for n = 1, . . . , b, and note that we have completed this step of the construction and the
at most one member of each pair may belong to A. The recursion may continue.
idea of the proof is to do something similar with pairs Since we can construct b0 , . . . , bn satisfying (a) and (b)
replaced by larger clumps, using long runs of excluded for any n, we have lim sup N (x)/x ≤ 1/2n for any n,
elements of B. yielding lim N (x)/x = 0 as desired.
Suppose we have positive integers b0 = 1, b1 , . . . , bn
Second solution: (by Paul Pollack) Let S be the set
with the following properties:
of possible values of lim sup N (x)/x; since S ⊆ [0, 1]
(a) For i = 1, . . . , n, ci = bi /(2bi−1 ) is an integer. is bounded, it has a least upper bound L. Suppose by
way of contradiction that L > 0; we can then choose
(b) For ei ∈ {−1, 0, 1}, |e1 b1 + · · · + en bn | ∈
/ B.
A, B satisfying the conditions of the problem such that
Each nonnegative integer a has a unique “base expan- lim sup N (x)/x > 3L/4.
sion” To begin with, we can certainly find some positive inte-
ger m ∈ / B, so that A is disjoint from A+m = {a+m :
a = a0 b0 + · · · + an−1 bn−1 + mbn (0 ≤ ai < 2ci ); a ∈ A}. Put A0 = A∪(A+m) and let N 0 (x) be the size
of A0 ∩{1, . . . , x}; then lim sup N 0 (x)/x = 3L/2 > L,
if two integers have expansions with the same value
so A0 cannot obey the conditions of the problem state-
of m, and values of ai differing by at most 1 for
ment. That is, if we let B 0 be the set of positive integers
i = 0, . . . , n − 1, then their difference is not in B, so
that occur as differences between elements of A0 , then
at most one of them lies in A. In particular, for any
there exists an integer n such that among any n consec-
di ∈ {0, . . . , ci − 1}, any m0 ∈ {0, 2c0 − 1} and any
utive integers, at least one lies in B 0 . But
mn , the set
B 0 ⊆ {b + em : b ∈ B, e ∈ {−1, 0, 1}},
{m0 b0 + (2d1 + e1 )b0 + · · ·
+ (2dn−1 + en−1 )bn−1 + (2mn + en )bn }, so among any n + 2m consecutive integers, at least one
lies in B. This contradicts the condition of the problem
where each ei runs over {0, 1}, contains at most one statement.
element of A; consequently, lim sup N (x)/x ≤ 1/2n .
We conclude that it is impossible to have L > 0, so
We now produce such bi recursively, starting with b0 = L = 0 and lim N (x)/x = 0 as desired.
1 (and both (a) and (b) holding vacuously). Given
b0 , . . . , bn satisfying (a) and (b), note that b0 + · · · + Remark: A hybrid between these two arguments is
bn−1 < bn by induction on n. By the hypotheses to note that if we can produce c1 , . . . , cn such that
of the problem, we can find a set Sn of 6bn consec- |ci − cj | ∈ / B for i, j = 1, . . . , n, then the translates A +
utive integers, none of which belongs to B. Let bn+1 c1 , . . . , A + cn are disjoint and so lim sup N (x)/x ≤
be the second-smallest multiple of 2bn in Sn ; then 1/n. Given c1 ≤ · · · ≤ cn as above, we can then choose
bn+1 + x ∈ Sn for −2bn ≤ x ≤ 0 clearly, and also cn+1 to be the largest element of a run of cn + 1 con-
for 0 ≤ x ≤ 2bn because there are most 4bn − 1 ele- secutive integers, none of which lie in B.
ments of Sn preceding bn+1 . In particular, the analogue
We first check that g 0 (z) 6= 0 whenever z is real and On the other hand, the term on the left is the dot product
z > 1. In this case, for rj = eiθj , we have z − rj = of r1 +· · ·+ra with itself, i.e., its squared length. Since
1 1
(z − cos(θj )) + sin(θj )i, so the real part of z−r j
− 2z this vector has a in each of its first b coordinates, the
is dot product is at least a2 b. Hence an ≥ a2 b, whence
n ≥ ab as desired.
z − cos(θj ) 1
− Second solution: (by Richard Stanley) Suppose with-
z 2 − 2z cos(θj ) + 1 2z out loss of generality that the a × b submatrix occupies
z2 − 1 the first a rows and the first b columns. Let M be the
= > 0. submatrix occupying the first a rows and the last n − b
2z(z 2 − 2z cos(θj ) + 1)
columns. Then the hypothesis implies that the matrix
Hence g 0 (z)/g(z) has positive real part, so g 0 (z)/g(z) M M T has n − b’s on the main diagonal and −b’s else-
and hence g(z) are nonzero. where. Hence the column vector v of length a consist-
ing of all 1’s satisfies M M T v = (n − ab)v, so n − ab is
Applying the same argument after replacing p(z) by an eigenvalue of M M T . But M M T is semidefinite, so
p(eiθ z), we deduce that g 0 cannot have any roots out- its eigenvalues are all nonnegative real numbers. Hence
side the unit circle. Applying the same argument after n − ab ≥ 0.
replacing p(z) by z n p(1/z), we also deduce that g 0 can-
not have any roots inside the unit circle. Hence all roots Remarks: A matrix as in the problem is called
of g 0 have absolute value 1, as desired. a Hadamard matrix, because it meets the equality
Qn condition of Hadamard’s inequality: any n × n matrix
Third solution: Write p(z) = c j=1 (z − rj ) and put with ±1 entries has absolute determinant at most nn/2 ,
rj = e2iθj . Note that g(e2iθ ) is equal to a nonzero with equality if and only if the rows are mutually
constant times orthogonal (from the interpretation of the determinant
n as the volume of a paralellepiped whose edges are
Y ei(θ+θj ) − e−i(θ+θj ) parallel to the row vectors). Note that this implies
h(θ) =
j=1
2i that the columns are also mutually orthogonal. A
n generalization of this problem, with a similar proof,
is known as Lindsey’s lemma: the sum of the entries
Y
= sin(θ + θj ).
j=1
in any√a × b submatrix of a Hadamard matrix is at
most abn. Stanley notes that Ryser (1981) asked
Since h has at least 2n roots (counting multiplicity) in for the smallest size of a Hadamard matrix containing
the interval [0, 2π), h0 does also by repeated applica- an r × s submatrix of all 1’s, and refers to the URL
tion of Rolle’s theorem. Since g 0 (e2iθ ) = 2ie2iθ h0 (θ), www3.interscience.wiley.com/cgi-bin/
g 0 (z 2 ) has at least 2n roots on the unit circle. Since abstract/110550861/ABSTRACT for more
g 0 (z 2 ) is equal to z −n−1 times a polynomial of degree information.
2n, g 0 (z 2 ) has all roots on the unit circle, as then does A5 First solution: We make the substitution x = tan θ,
g 0 (z). rewriting the desired integral as
Remarks: The second solution imitates the proof of Z π/4
the Gauss-Lucas theorem: the roots of the derivative of log(tan(θ) + 1) dθ.
a complex polynomial lie in the convex hull of the roots 0
of the original polynomial. The second solution is close
to problem B3 from the 2000 Putnam. A hybrid be- Write
tween the first and third solutions is to check that on the log(tan(θ) + 1)
unit circle, Re(zg 0 (z)/g(z)) = 0 while between any
two roots of p, Im(zg 0 (z)/g(z)) runs from +∞ to −∞ = log(sin(θ) + cos(θ)) − log(cos(θ))
and so must have a zero crossing. (This only works √
when p has distinct roots, but the general case follows and then note that sin(θ) + cos(θ) = 2 cos(π/4 − θ).
by the continuity of the roots of a polynomial as func- We may thus rewrite the integrand as
tions of the coefficients.) One can also construct a solu- 1
tion using Rouché’s theorem. log(2) + log(cos(π/4 − θ)) − log(cos(θ)).
2
A4 First solution: Choose a set of a rows r1 , . . . , ra con- But over the interval [0, π/4], the integrals of
taining an a × b submatrix whose entries are all 1. Then log(cos(θ)) and log(cos(π/4 − θ)) are equal, so their
for i, j ∈ {1, . . . , a}, we have ri · rj = n if i = j and 0 contributions cancel out. The desired integral is then
otherwise. Hence just the integral of 12 log(2) over the interval [0, π/4],
a
which is π log(2)/8.
X
ri · rj = an. Second solution: (by Roger Nelsen) Let I denote the
i,j=1 desired integral. We make the substitution x = (1 −
at Q2 and proceeding counterclockwise must contain If n = 1, we must have k1 = 1, which works. Note that
all of Q3 , . . . , Qn , while the open semicircle starting hereafter we cannot have k1 = 1.
at Qi and proceeding counterclockwise must contain If n = 2, we have (k1 , k2 ) ∈ {(2, 4), (3, 3)}, neither of
Qi+1 , . . . , Qn , Q1 , . . . , Qi−1 . Thus two open semicir- which work.
cles cover the entire circle, contradiction.
If n = 3, we have k1 + k2 + k3 = 11,
It follows that if the polygon has at least one acute an- so 2 ≤ k1 ≤ 3. Hence (k1 , k2 , k3 ) ∈
gle, then it has either one acute angle or two acute an- {(2, 2, 7), (2, 3, 6), (2, 4, 5), (3, 3, 5), (3, 4, 4)}, and
gles occurring consecutively. In particular, there is a only (2, 3, 6) works.
unique pair of consecutive vertices Q1 , Q2 in counter-
clockwise order for which ∠Q2 is acute and ∠Q1 is not If n = 4, we must have equality in the AM-HM inequal-
acute. Then the remaining points all lie in the arc from ity, which only happens when k1 = k2 = k3 = k4 = 4.
the antipode of Q1 to Q1 , but Q2 cannot lie in the arc, Hence the solutions are n = 1 and k1 = 1, n = 3 and
and the remaining points cannot all lie in the arc from (k1 , k2 , k3 ) is a permutation of (2, 3, 6), and n = 4 and
the antipode of Q1 to the antipode of Q2 . Given the (k1 , k2 , k3 , k4 ) = (4, 4, 4, 4).
choice of Q1 , Q2 , let x be the measure of the counter- Remark: In the cases n = 2, 3, Greg Kuperberg sug-
clockwise arc from Q1 to Q2 ; then the probability that gests the alternate approach of enumerating the solu-
the other points fall into position is 2−n+2 − xn−2 if tions of 1/k1 +· · ·+1/kn = 1 with k1 ≤ · · · ≤ kn . This
x ≤ 1/2 and 0 otherwise. is easily done by proceeding in lexicographic order: one
Hence the probability that the polygon has at least one obtains (2, 2) for n = 2, and (2, 3, 6), (2, 4, 4), (3, 3, 3)
acute angle with a given choice of which two points will for n = 3, and only (2, 3, 6) contributes to the final an-
act as Q1 and Q2 is swer.
Z 1/2
n − 2 −n+1 B3 First solution: The functions are precisely f (x) = cxd
(2−n+2 − xn−2 ) dx = 2 . for c, d > 0 arbitrary except that we must take c = 1
0 n−1
in case d = 1. To see that these work, note that
Since there are n(n − 1) choices for which two points f 0 (a/x) = dc(a/x)d−1 and x/f (x) = 1/(cxd−1 ), so
act as Q1 and Q2 , the probability of at least one acute the given equation holds if and only if dc2 ad−1 = 1.
angle is n(n − 2)2−n+1 . If d 6= 1, we may solve for a no matter what c is; if
d = 1, we must have c = 1. (Thanks to Brad Rodgers
Second solution: (by Calvin Lin) As in the first solu-
for pointing out the d = 1 restriction.)
tion, we may compute the probability that for a particu-
lar one of the points Q1 , the angle at Q1 is not acute but To check that these are all solutions, put b = log(a) and
the following angle is, and then multiply by n. Imagine y = log(a/x); rewrite the given equation as
picking the points by first choosing Q1 , then picking
f (eb−y )f 0 (ey ) = eb−y .
n − 1 pairs of antipodal points and then picking one
member of each pair. Let R2 , . . . , Rn be the points of Put
the pairs which lie in the semicircle, taken in order away
from Q1 , and let S2 , . . . , Sn be the antipodes of these. g(y) = log f (ey );
Then to get the desired situation, we must choose from then the given equation rewrites as
the pairs to end up with all but one of the Si , and we
cannot take Rn and the other Si or else ∠Q1 will be g(b − y) + log g 0 (y) + g(y) − y = b − y,
acute. That gives us (n − 2) good choices out of 2n−1 ;
since we could have chosen Q1 to be any of the n points, or
the probability is again n(n − 2)2−n+1 . log g 0 (y) = b − g(y) − g(b − y).
B1 Take P (x, y) = (y − 2x)(y − 2x − 1). To see that By the symmetry of the right side, we have g 0 (b − y) =
this works, first note that if m = bac, then 2m is an g 0 (y). Hence the function g(y) + g(b − y) has zero
integer less than or equal to 2a, so 2m ≤ b2ac. On derivative and so is constant, as then is g 0 (y). From
the other hand, m + 1 is an integer strictly greater than this we deduce that f (x) = cxd for some c, d, both
a, so 2m + 2 is an integer strictly greater than 2a, so necessarily positive since f 0 (x) > 0 for all x.
b2ac ≤ 2m + 1.
Second solution: (suggested by several people) Substi-
B2 By the arithmetic-harmonic mean inequality or the tute a/x for x in the given equation:
Cauchy-Schwarz inequality, a
f 0 (x) = .
xf (a/x)
1 1
(k1 + · · · + kn ) + ··· + ≥ n2 .
k1 kn Differentiate:
2
We must thus have 5n − 4 ≥ n , so n ≤ 4. Without a a2 f 0 (a/x)
f 00 (x) = − + .
loss of generality, we may suppose that k1 ≤ · · · ≤ kn . x2 f (a/x) x3 f (a/x)2
Now substitute to eliminate evaluations at a/x: Third solution: (by Greg Martin) As in the second so-
lution, it is convenient to allow f (m, 0) = f (0, n) = 1.
f 0 (x) f 0 (x)2 Define the generating function
f 00 (x) = − + .
x f (x)
∞ X
X ∞
Clear denominators: G(x, y) = f (m, n)xm y n .
m=0 n=0
xf (x)f 00 (x) + f (x)f 0 (x) = xf 0 (x)2 .
As equalities of formal power series (or convergent se-
Divide through by f (x)2 and rearrange: ries on, say, the region |x|, |y| < 31 ), we have
f 0 (x) xf 00 (x) xf 0 (x)2 XX X
0= + − . G(x, y) = xm y n 1
f (x) f (x) f (x)2
m≥0 n≥0 k1 , ..., kn ∈Z
0 |k1 |+···+|kn |≤m
The right side is the derivative of xf (x)/f (x), so that X X X
quantity is constant. That is, for some d, = yn xm
n≥0 k1 , ..., kn ∈Z m≥|k1 |+···+|kn |
f 0 (x) d
= . x|k1 |+···+|kn |
f (x) x
X X
= yn
1−x
n≥0 k1 , ..., kn ∈Z
Integrating yields f (x) = cxd , as desired. n
1 X
n
X
|k|
B4 First solution: Define f (m, n, k) as the number of n- = y x
1−x
tuples (x1 , x2 , . . . , xn ) of integers such that |x1 |+· · ·+ n≥0 k∈Z
n
|xn | ≤ m and exactly k of x1 , . . . , xn are nonzero. To
1 X n 1+x
choose such a tuple, we may choose the k nonzero posi- = y
1−x 1−x
tions, the signs of those k numbers, and then an ordered n≥0
polynomials of degree d, and let Hd be the subset of Integrating both sides from 0 to 1 (and substituting y =
Pd of polynomials killed by ∇2 , which has dimension 1 − x) yields
≥ dim(Pd ) − dim(Pd−2 ); the given problem amounts Z 1
to showing that this inequality is actually an equality. X sgn(π)
= (x + n − 1)(x − 1)n−1 dx
Consider the operator Q∇2 (i.e., apply ∇2 then multi- ν(π) + 1 0
π∈Sn
ply by Q) on Pd ; its zero eigenspace is precisely Hd . Z 1
By the calculation from the first solution, if R ∈ Pd , = (−1)n+1 (n − y)y n−1 dy
then 0
n
= (−1)n+1 ,
2 2
∇ (QR) − Q∇ R = (2n + 4d)R. n+1
as desired.
Consequently, Qj Hd−2j is contained in the eigenspace
of Q∇2 on Pd of eigenvalue Second solution: We start by recalling a form of the
principle of inclusion-exclusion: if f is a function on
(2n + 4(d − 2j)) + · · · + (2n + 4(d − 2)). the power set of {1, . . . , n}, then
A standard argument in enumerative combinatorics (the by induction on n, which for x = 1 implies the desired
Exponential Formula) gives result. (This can also be deduced as in the other solu-
tions, but in this argument it is necessary to formulate
∞ ∞
X tn X tk the strong induction hypothesis.)
Zn (x1 , . . . , xn ) = exp xk ,
n=0
n! k
k=1
Let R(n, x) be the right hand side of the above equa-
yielding tion. It is easy to verify that
∞ 1
tn t2 t3
X Z
f (n) = exp xt − + − · · · dx
n=0
n! 0 2 3 (−1)n+1
R(x, n) = R(x + 1, n − 1) + (n − 1)!
Z 1 x
= e(x−1)t+log(1+t) dx n−1
X (n − 1)!
0 + (−1)l−1 R(x, n − l),
Z 1 (n − l)!
l=2
= (1 + t)e(x−1)t dx
0
1
= (1 − e−t )(1 + t). since the sum telescopes. To prove the desired equality,
t
it suffices to show that the left hand side satisfies the
Expanding the right side as a Taylor series and compar- same recurrence. This follows because we can classify
ing coefficients yields the desired result. each π ∈ Sn as either fixing n, being an n-cycle, or hav-
Fourth solution (sketch): (by David Savitt) We prove ing n in an l-cycle for one of l = 2, . . . , n − 1; writing
the identity of rational functions the sum over these classes gives the desired recurrence.
i occurring as a local maximum is i(i − 1)/(n(n − 1)), expected value of a random variable over all choices of
and the average number of local maxima is P, Q, R. Write [XY Z] for the area of triangle XY Z.
n n If P, Q, R, S are the four points, we may ignore the case
X i(i − 1) 2 X i
= where three of them are collinear, as this occurs with
i=1
n(n − 1) n(n − 1) i=1 2 probability zero. Then the only way they can fail to
2
n+1
form the vertices of a convex quadrilateral is if one of
= them lies inside the triangle formed by the other three.
n(n − 1) 3
There are four such configurations, depending on which
n+1 point lies inside the triangle, and they are mutually ex-
= .
3 clusive. Hence the desired probability is 1 minus four
One can obtain a similar (if slightly more intricate) so- times the probability that S lies inside triangle P QR.
lution inductively, by removing the known local maxi- That latter probability is simply E([P QR]) divided by
mum n and splitting into two shorter sequences. the area of the disc.
Remark: The usual term for a local maximum in this Let O denote the center of the circle, and let P ′ , Q′ , R′
sense is a peak. The complete distribution for the num- be the projections of P, Q, R onto the circle from O.
ber of peaks is known; Richard Stanley suggests the ref- We can write
erence: F. N. David and D. E. Barton, Combinatorial
[P QR] = ±[OP Q] ± [OQR] ± [ORP ]
Chance, Hafner, New York, 1962, p. 162 and subse-
quent. for a suitable choice of signs, determined as follows. If
A5 Since the desired expression involves symmetric func- the points P ′ , Q′ , R′ lie on no semicircle, then all of the
tions of a1 , . . . , an , we start by finding a polynomial signs are positive. If P ′ , Q′ , R′ lie on a semicircle in
with a1 , . . . , an as roots. Note that that order and Q lies inside the triangle OP R, then the
sign on [OP R] is positive and the others are negative.
1 ± i tan θ = e±iθ sec θ If P ′ , Q′ , R′ lie on a semicircle in that order and Q lies
outside the triangle OP R, then the sign on [OP R] is
so that negative and the others are positive.
1 + i tan θ = e2iθ (1 − i tan θ). We first calculate
Consequently, if we put ω = e2inθ , then the polynomial E([OP Q] + [OQR] + [ORP ]) = 3E([OP Q]).
cations. For simplicity, we assume without loss of gen- In particular, the distribution on θ = θ1 + θ2 is π2θ2 on
erality that the circle has radius 1. Let E denote the [0, π]. Put rP = OP, rQ = OQ, rR = OR. Again, the
distribution on rP is given by 2rP on [0, 1], and simi- the lines P Q, QR, RP , which with probability 1 di-
larly for rQ , rR ; these are independent from each other vide the interior of the circle into seven regions. Put
and from the joint distribution of θ1 , θ2 . Write E ′ (X) a = [P QR], let b1 , b2 , b3 denote the areas of the three
for the expectation of a random variable X restricted to other regions sharing a side with the triangle, and let
this part of the domain. c1 , c2 , c3 denote the areas of the other three regions.
Let χ be the random variable with value 1 if Q is inside Put A = E(a), B = E(b1 ), C = E(c1 ), so that
triangle OP R and 0 otherwise. We now compute A + 3B + 3C = π.
Note that c1 + c2 + c3 + a is the area of the region
E ′ ([OP R]) in which we can choose a fourth point S so that the
Z 1 2 Z π quadrilateral P QRS fails to be convex. By comparing
1 2 2θ
= 2r dr 2
sin(θ) dθ expectations, we have 3C + A = 4A, so A = C and
2 0 0 π 4A + 3B = π.
4
= We will compute B + 2A = B + 2C, which is the ex-
9π pected area of the part of the circle cut off by a chord
E ′ (χ[OP R]) through two random points D, E, on the side of the
= E ′ (2[OP R]2 /θ) chord not containing a third random point F . Let h be
Z 1 2 Z π the distance from the center O of the circle to the line
1 2θ −1 2 DE. We now determine the distribution of h.
= 2r3 dr 2
θ sin (θ) dθ
2 0 0 π Put r = OD; the distribution of r is 2r on [0, 1]. With-
1 out loss of generality, suppose O is the origin and D
= .
8π lies on the positive x-axis. For fixed r, the distribution
Also recall that given any triangle XY Z, if T is chosen of h runs over [0, r], and can be computed as the area
uniformly at random inside XY Z, the expectation of of the infinitesimal region in which E can be chosen so
[T XY ] is the area of triangle bounded by XY and the the chord through DE has distance to O between h and
centroid of XY Z, namely 31 [XY Z]. h + dh, divided by π. This region splits into two sym-
metric pieces, one of which lies between chords making
Let χ be the random variable with value 1 if Q is inside angles of arcsin(h/r) and arcsin((h + dh)/r) with the
triangle OP R and 0 otherwise. Then x-axis. The angle between these is dθ = dh/(r2 − h2 ).
Draw the chord through D at distance h to O, and let
E ′ ([OP Q] + [OQR] + [ORP ] − [P QR]) L1 , L2 be the lengths of the parts on opposite sides of
= 2E ′ (χ([OP Q] + [OQR]) + 2E ′ ((1 − χ)[OP R]) D; then the area we are looking for is 21 (L21 + L22 )dθ.
2 Since
= 2E ′ ( χ[OP R]) + 2E ′ ([OP R]) − 2E ′ (χ[OP R])
3
p p
{L1 , L2 } = 1 − h2 ± r2 − h2 ,
2 29
= 2E ([OP R]) − E ′ (χ[OP R]) =
′
. the area we are seeking (after doubling) is
3 36π
Finally, note that the case when P ′ , Q′ , R′ lie on a semi- 1 + r2 − 2h2
2 √ .
circle in some order occurs with probability 3/4. (The r 2 − h2
case where they lie on a semicircle proceeding clock-
Dividing by π, then integrating over r, we compute the
wise from P ′ to its antipode has probability 1/4; this
distribution of h to be
case and its two analogues are exclusive and exhaus-
1 1 1 + r2 − 2h2
Z
tive.) Hence
2 √ 2r dr
π h r 2 − h2
E([P QR])
16
= E([OP Q] + [OQR] + [ORP ]) = (1 − h2 )3/2 .
3π
3
− E ′ ([OP Q] + [OQR] + [ORP ] − [P QR]) We now return to computing B + 2A. Let A(h) de-
4
4 29 35 note the smaller of the two areas of the disc cut off by
= − = , a chord at distance h. The chance that the third point
3π 48π 48π
is in the smaller (resp. larger) portion is A(h)/π (resp.
so the original probability is 1−A(h)/π), and then the area we are trying to compute
is π − A(h) (resp. A(h)). Using the distribution on h,
4E([P QR]) 35 and the fact that
1− =1− .
π 12π 2 Z 1p
A(h) = 2 1 − h2 dh
Second solution: (by David Savitt) As in the first so- h
lution, it suffices to check that for P, Q, R chosen uni- π p
35 = − arcsin(h) − h 1 − h2 ,
formly at random in the disc, E([P QR]) = 48π . Draw 2
Let P be the line at infinity in the real projective plane; the segment O′ P . If we consider the dual affine plane
i.e., P is the set of possible directions of lines in the as being divided into regions by the lines of S ′ , then the
plane, viewed as a circle. Remove the directions corre- lines of S ′ crossing the segment O′ P are determined by
sponding to lines through two points of S; this leaves which region P lies in.
behind n2 intervals. Thus our original maximum is equal to the maximum
Given a direction in one of the intervals, consider the number of regions into which n−1 lines divide an affine
set of linear partitions achieved by lines parallel to that plane. Byinduction on n, this number is easily seen to
direction. Note that the resulting collection of partitions be 1 + n2 .
depends only on the interval. Then note that the collec- Remark: Given a finite set S of points in Rn , a non-
tions associated to adjacent intervals differ in only one Radon partition of S is a pair (A, B) of complementary
element. subsets that can be separated by a hyperplane. Radon’s
The trivial partition that puts all of S on one side is theorem states that if #S ≥ n+2, then not every (A, B)
in every such collection. We now observe that for any is a non-Radon partition. The result of this problem has
other linear partition {A, B}, the set of intervals to been greatly extended, especially within the context of
which {A, B} is: matroid theory and oriented matroid theory. Richard
Stanley suggests the following references: T. H. Bry-
(a) a consecutive block of intervals, but lawski, A combinatorial perspective on the Radon con-
(b) not all of them. vexity theorem, Geom. Ded. 5 (1976), 459-466; and T.
Zaslavsky, Extremal arrangements of hyperplanes, Ann.
For (a), note that if ℓ1 , ℓ2 are nonparallel lines achieving N. Y. Acad. Sci. 440 (1985), 69-87.
the same partition, then we can rotate around their point
of intersection to achieve all of the intermediate direc- B4 The maximum is 2k , achieved for instance by the sub-
tions on one side or the other. For (b), the case n = 3 space
is evident; to reduce the general case to this case, take
points P, Q, R such that P lies on the opposite side of {(x1 , . . . , xn ) ∈ Rn : x1 = · · · = xn−k = 0}.
the partition from Q and R.
First solution: More generally, we show that any affine
It follows now that that each linear partition, except for
k-dimensional plane in Rn can contain at most 2k
the trivial one, occurs in exactly one place as the parti-
points in Z. The proof is by induction on k + n; the
tion associated to some interval but not to its immediate
case k = n = 0 is clearly true.
counterclockwise neighbor. In other words, the num-
ber of linear partitions is one more than the number of Suppose that V is a k-plane in Rn . Denote the hyper-
intervals, or n2 + 1 as desired. planes {xn = 0} and {xn = 1} by V0 and V1 , respec-
tively. If V ∩ V0 and V ∩ V1 are each at most (k − 1)-
Second solution: We prove the upper bound by induc-
dimensional, then V ∩ V0 ∩ Z and V ∩ V1 ∩ Z each have
tion on n. Choose a point P in the convex hull of S. cardinality at most 2k−1 by the induction assumption,
Put S ′ = S \ {P }; by the induction hypothesis, there
and hence V ∩ Z has at most 2k elements. Otherwise,
are at most n−12 + 1 linear partitions of S ′ . Note that if V ∩ V0 or V ∩ V1 is k-dimensional, then V ⊂ V0
each linear partition of S restricts to a linear partition
or V ⊂ V1 ; now apply the induction hypothesis on V ,
of S ′ . Moreover, if two linear partitions of S restrict to viewed as a subset of Rn−1 by dropping the last coor-
the same linear partition of S ′ , then that partition of S ′
dinate.
is achieved by a line through P .
Second solution: Let S be a subset of Z contained in
By rotating a line through P , we see that there are at a k-dimensional subspace of V . This is equivalent to
most n − 1 partitions of S ′ achieved by lines through asking that any t1 , . . . , tk+1 ∈ S satisfy a nontrivial
P : namely, the partition only changes when the rotating linear dependence c1 t1 + · · · + ck+1 tk+1 = 0 with
line passes through one of the points of S. This yields c1 , . . . , ck+1 ∈ R. Since t1 , . . . , tk+1 ∈ Qn , given
the desired result. such a dependence we can always find another one with
Third solution: (by Noam Elkies) We enlarge the plane c1 , . . . , ck+1 ∈ Q; then by clearing denominators, we
to a projective plane by adding a line at infinity, then can find one with c1 , . . . , ck+1 ∈ Z and not all having a
apply the polar duality map centered at one of the points common factor.
O ∈ S. This turns the rest of S into a set S ′ of n − 1 Let F2 denote the field of two elements, and let S ⊆ Fn2
lines in the dual projective plane. Let O′ be the point be the reductions modulo 2 of the points of S. Then any
in the dual plane corresponding to the original line at t1 , . . . , tk+1 ∈ S satisfy a nontrivial linear dependence,
infinity; it does not lie on any of the lines in S ′ . because we can take the dependence from the end of
Let ℓ be a line in the original plane, corresponding to a the previous paragraph and reduce modulo 2. Hence S
point P in the dual plane. If we form the linear partition is contained in a k-dimensional subspace of F2n , and
induced by ℓ, then the points of S \ {O} lying in the the latter has cardinality exactly 2k . Thus S has at most
same part as O correspond to the lines of S ′ which cross 2k elements, as does S.
nates. Since each coordinate of a point in Z can only for n large, the right side is smaller than ǫ.
take two values, V ∩ Z can have at most 2k elements.
In our case, we deduce that
Remark: The proposers probably did not
realize that this problem appeared online (k+1)/k
an k+1
about three months before the exam, at lim =
http://www.artofproblemsolving.com/ n→∞ n k
Forum/viewtopic.php?t=105991. (It may and so
very well have also appeared even earlier.)
k
ak+1
n k+1
B5 The answer is 1/16. We have lim = ,
n→∞ nk k
Z 1 Z 1
2
x f (x) dx − xf (x)2 dx as desired.
0 0
Z 1 Remark: The use of Cesaro’s lemma above is the spe-
= (x3 /4 − x(f (x) − x/2)2 ) dx cial case bn = n of the Cesaro-Stolz theorem: if an , bn
0 are sequences such that bn is positive, strictly increas-
Z 1
≤ x3 /4 dx = 1/16, ing, and unbounded, and
0
an+1 − an
with equality when f (x) = x/2. lim = L,
n→∞ bn+1 − bn
B6 First solution: We start with some easy upper and then
lower bounds on an . We write O(f (n)) and Ω(f (n))
for functions g(n) such that f (n)/g(n) and g(n)/f (n), an
lim = L.
respectively, are bounded above. Since an is a non- n→∞ bn
decreasing sequence, an+1 − an is bounded above, so
−1/k Second solution: In this solution, rather than applying
an = O(n). That means an = Ω(n−1/k ), so
Taylor’s theorem with remainder to (1 + x)m for 1 <
n
!
X m < 2 and x > 0, we only apply convexity to deduce
an = Ω i−1/k = Ω(n(k−1)/k ). that (1 + x)m ≥ 1 + mx. This gives
i=1
But turning this around, the fact that Note that both bounds have sign opposite to bn ; more-
over, by the bound an = Ω(n(k−1)/k ), both bounds
an+1 − an have absolutely value strictly less than that of bn for n
= a−1/k sufficiently large. Consequently, for n large,
n
−1/(k+1)
k+1 |e1 | ≤ |bn |.
≤ n−1/(k+1) (1 + o(1)),
k We now work on e2 . By Taylor’s theorem with remain-
der applied to (1 + x)m for x > 0 and 0 < m < 1,
where o(1) denotes a function tending to 0 as n → ∞,
yields 1 + mx ≥ (1 + x)m
m(m − 1) 2
an ≥ 1 + mx + x .
2
−1/(k+1) Xn
k+1
≤ i−1/(k+1) (1 + o(1)) The “main term” of L((n + 1)k/(k+1) − nk/(k+1) )
k k
i=1 is L k+1 n−1/(k+1) . To make this coincide with
−1/(k+1) −1/k −1/(k+1)
k+1 k+1 L n , we take
= nk/(k+1) (1 + o(1))
k k k/(k+1)
k/(k+1) k+1
k+1 L= .
= nk/(k+1) (1 + o(1)), k
k
so We then find that
an
k+1
k/(k+1) |e2 | = O(n−2 ),
lim sup ≤
n→∞ nk/(k+1) k
and because bn+1 = e1 + e2 , we have |bn+1 | ≤ |bn | +
and this completes the proof. |e2 |. Hence
Third solution: We argue that an → ∞ as in the first n
!
solution. Write bn = an − Lnk/(k+1) , for a value of L
X
−2
|bn | = O i = O(1),
to be determined later. We have i=1
bn+1 and so
k/(k+1) k/(k+1)
= bn + a−1/k
n − L((n + 1) −n ) ak+1
k+1
k
n
= e1 + e2 , lim = Lk+1 = .
n→∞ nk k
where
Remark: The case k = 2 appeared on the 2004 Roma-
e1 = bn + a−1/k
n − L−1/k n−1/(k+1) nian Olympiad (district level).
e2 = L((n + 1)k/(k+1) − nk/(k+1) ) Remark: One can make a similar argument for any se-
quence given by an+1 = an + f (an ), when f is a de-
− L−1/k n−1/(k+1) . creasing function.
We first estimate e1 . For −1 < m < 0, by the convexity Remark: Richard Stanley suggests a heuristic for de-
of (1 + x)m and (1 + x)1−m , we have termining the asymptotic behavior of sequences of this
type: replace the given recursion
1 + mx ≤ (1 + x)m
≤ 1 + mx(1 + x)m−1 . an+1 − an = a−1/k
n
1 y ′ = y −1/k
− L−(k+1)/k n−1 bn ≤ e1 − bn
k
1 and determine the asymptotics of the latter.
≤ − bn an−(k+1)/k .
k
√
A1 The only such α are 2/3, 3/2, (13 ± 601)/12. Second solution: For any nonzero value of α, the two
First solution: Let C1 and C2 be the curves y = αx2 + conics will intersect in four points in the complex pro-
αx + 241 1
and x = αy 2 + αy + 24 , respectively, and let jective plane P2 (C). To determine the y-coordinates of
L be the line y = x. We consider three cases. these intersection points, subtract the two equations to
obtain
If C1 is tangent to L, then the point of tangency (x, x)
satisfies (y − x) = α(x − y)(x + y) + α(x − y).
1 Therefore, at a point of intersection we have either
2αx + α = 1, x = αx2 + αx + ;
24 x = y, or x = −1/α − (y + 1). Substituting these
two possible linear conditions into the second equation
by symmetry, C2 is tangent to L there, so C1 and C2 are shows that the y-coordinate of a point of intersection is
tangent. Writing α = 1/(2x + 1) in the first equation a root of either Q1 (y) = αy 2 + (α − 1)y + 1/24 or
and substituting into the second, we must have Q2 (y) = αy 2 + (α + 1)y + 25/24 + 1/α.
x2 + x 1 If two curves are tangent, then the y-coordinates of at
x= + , least two of the intersection points will coincide; the
2x + 1 24
converse is also true because one of the curves is the
which simplifies to 0 = 24x2 − 2x− 1 = (6x+ 1)(4x− graph of a function in x. The coincidence occurs pre-
1), or x ∈ {1/4, −1/6}. This yields α = 1/(2x + 1) ∈ cisely when either the discriminant of at least one of
{2/3, 3/2}. Q1 or Q2 is zero, or there is a common root of Q1 and
Q2 . Computing the discriminants of Q1 and Q2 yields
If C1 does not intersect L, then C1 and C2 are separated
(up to constant factors) f1 (α) = 6α2 − 13α + 6 and
by L and so cannot be tangent.
f2 (α) = 6α2 − 13α − 18, respectively. If on the other
If C1 intersects L in two distinct points P1 , P2 , then it hand Q1 and Q2 have a common root, it must be also
is not tangent to L at either point. Suppose at one of a root of Q2 (y) − Q1 (y) = 2y + 1 + 1/α, yielding
these points, say P1 , the tangent to C1 is perpendicular y = −(1 + α)/(2α) and 0 = Q1 (y) = −f2 (α)/(24α).
to L; then by symmetry, the same will be true of C2 , so Thus the values of α for which the two curves are tan-
C1 and C2 will be tangent at P1 . In this case, the point gent must be contained in the√set of zeros of f1 and f2 ,
P1 = (x, x) satisfies namely 2/3, 3/2, and (13 ± 601)/12.
1 Remark: The fact that the two conics in P2 (C) meet in
2αx + α = −1, x = αx2 + αx + ; four points, counted with multiplicities, is a special case
24
of Bézout’s theorem: two curves in P2 (C) of degrees
writing α = −1/(2x + 1) in the first equation and sub- m, n and not sharing any common component meet in
stituting into the second, we have exactly mn points when counted with multiplicity.
Many solvers were surprised that the proposers chose
x2 + x 1
x=− + , the parameter 1/24 to give two rational roots and two
2x + 1 24 nonrational roots. In fact, they had no choice in the
√ matter: attempting to make all four roots rational by
or x = (−23√± 601)/72. This yields α = −1/(2x +
replacing 1/24 by β amounts to asking for β 2 + β and
1) = (13 ± 601)/12. β 2 + β + 1 to be perfect squares. This cannot happen
If instead the tangents to C1 at P1 , P2 are not perpen- outside of trivial cases (β = 0, −1) ultimately because
dicular to L, then we claim there cannot be any point the elliptic curve 24A1 (in Cremona’s notation) over Q
where C1 and C2 are tangent. Indeed, if we count inter- has rank 0. (Thanks to Noam Elkies for providing this
sections of C1 and C2 (by using C1 to substitute for y computation.)
in C2 , then solving for y), we get at most four solutions However, there are choices that make the radical milder,
counting multiplicity. Two of these are P1 and P2 , and e.g., β = 1/3 gives β 2 + β = 4/9 and β 2 + β + 1 =
any point of tangency counts for two more. However, 13/9, while β = 3/5 gives β 2 + β = 24/25 and β 2 +
off of L, any point of tangency would have a mirror im- β + 1 = 49/25.
age which is also a point of tangency, and there cannot
be six solutions. Hence we have now found all possible A2 The minimum is 4, achieved by the square with vertices
α. (±1, ±1).
First solution: To prove that 4 is a lower bound, mod 3 than to −1, the sequence mod 3 must look like
let S be a convex set of the desired form. Choose 1, 1, −1, 1, −1, . . ..
A, B, C, D ∈ S lying on the branches of the two hyper- It follows that the ordering satisfies the given condition
bolas, with A in the upper right quadrant, B in the upper if and only if the following two conditions hold: the
left, C in the lower left, D in the lower right. Then the first element in the ordering is not divisible by 3, and
area of the quadrilateral ABCD is a lower bound for the sequence mod 3 (ignoring zeroes) is of the form
the area of S. 1, 1, −1, 1, −1, . . .. The two conditions are indepen-
Write A = (a, 1/a), B = (b, −1/b), C = (−c, −1/c), dent, and the probability of the first is (2k +1)/(3k
+1)
D = (−d, 1/d) with a, b, c, d > 0. Then the area of the while the probability of the second is 1/ 2k+1 k , since
quadrilateral ABCD is 2k+1
there are k ways to order (k + 1) 1’s and k −1’s.
1 Hence the desired probability is the product of these
(a/b + b/c + c/d + d/a + b/a + c/b + d/c + a/d), k!(k+1)!
2 two, or (3k+1)(2k)! .
which by the arithmetic-geometric mean inequality is at A4 Note that n is a repunit if and only if 9n + 1 = 10m for
least 4. some power of 10 greater than 1. Consequently, if we
put
Second solution: Choose A, B, C, D as in the first so-
lution. Note that both the hyperbolas and the area of
n−1
the convex hull of ABCD are invariant under the trans- g(n) = 9f + 1,
formation (x, y) 7→ (xm, y/m) for any m > 0. For 9
m small, the counterclockwise angle from the line AC
then f takes repunits to repunits if and only if g takes
to the line BD approaches 0; for m large, this angle
powers of 10 greater than 1 to powers of 10 greater than
approaches π. By continuity, for some m this angle be-
1. We will show that the only such functions g are those
comes π/2, that is, AC and BD become perpendicular.
of the form g(n) = 10cnd for d ≥ 0, c ≥ 1 − d (all of
The area of ABCD is then AC · BD.
√ which clearly work), which will mean that the desired
It thus suffices to note that AC ≥ 2 2 (and similarly polynomials f are those of the form
for BD). This holds because if we draw the tangent
lines to the hyperbola xy = 1 at the points (1, 1) and 1
f (n) = (10c (9n + 1)d − 1)
(−1, −1), then A and C lie outside the region between 9
these lines. If we project the segment AC orthogonally
onto the line x √ = y = 1, the resulting projection has for the same c, d.
length at least 2 2, so AC must as well. It is convenient to allow “powers of 10” to be of the
Third solution: (by Richard Stanley) Choose form 10k for any integer k. With this convention, it
A, B, C, D as in the first solution. Now fixing A and suffices to check that the polynomials g taking powers
C, move B and D to the points at which the tangents of 10 greater than 1 to powers of 10 are of the form
to the curve are parallel to the line AC. This does not 10c nd for any integers c, d with d ≥ 0.
increase the area of the quadrilateral ABCD (even if First solution: Suppose that the leading term of g(x)
this quadrilateral is not convex). is axd , and note that a > 0. As x → ∞, we have
Note that B and D are now diametrically opposite; g(x)/xd → a; however, for x a power of 10 greater
write B = (−x, 1/x) and D = (x, −1/x). If we thus than 1, g(x)/xd is a power of 10. The set of powers of
repeat the procedure, fixing B and D and moving A and 10 has no positive limit point, so g(x)/xd must be equal
C to the points where the tangents are parallel to BD, to a for x = 10k with k sufficiently large, and we must
then A and C must move to (x, 1/x) and (−x, −1/x), have a = 10c for some c. The polynomial g(x) − 10c xd
respectively, forming a rectangle of area 4. has infinitely many roots, so must be identically zero.
Remark: Many geometric solutions are possible. An Second solution: We proceed by induction on d =
example suggested by David Savitt (due to Chris deg(g). If d = 0, we have g(n) = 10c for some c.
Brewer): note that AD and BC cross the positive and Otherwise, g has rational coefficients by Lagrange’s in-
negative x-axes, respectively, so the convex hull of terpolation formula (this applies to any polynomial of
ABCD contains O. Then check that the area of tri- degree d taking at least d + 1 different rational numbers
angle OAB is at least 1, et cetera. to rational numbers), so g(0) = t is rational. More-
over, g takes each value only finitely many times, so the
A3 Assume that we have an ordering of 1, 2, . . . , 3k + sequence g(100 ), g(101 ), . . . includes arbitrarily large
1 such that no initial subsequence sums to 0 mod powers of 10. Suppose that t 6= 0; then we can choose
3. If we omit the multiples of 3 from this order- a positive integer h such that the numerator of t is not
ing, then the remaining sequence mod 3 must look divisible by 10h . But for c large enough, g(10c) − t has
like 1, 1, −1, 1, −1, . . . or −1, −1, 1, −1, 1, . . .. Since numerator divisible by 10b for some b > h, contradic-
there is one more integer in the ordering congruent to 1 tion.
Consequently, t = 0, and we may apply the induction We first claim that a1 + · · · + an ≤ 4n − 6. Let V, E, F
hypothesis to g(n)/n to deduce the claim. denote the number of vertices, edges, and faces in T .
Remark: The second solution amounts to the fact that By Euler’s Formula, (F +1)−E+V = 2 (one must add
g, being a polynomial with rational coefficients, is con- 1 to the face count for the region exterior to P ). Each
tinuous for the 2-adic and 5-adic topologies on Q. By face has three edges, and each edge but the n outside
contrast, the first solution uses the “∞-adic” topology, edges belongs to two faces; hence F = 2E − n. On the
i.e., the usual real topology. other hand, each edge has two endpoints, and each of
the V − n internal vertices is an endpoint of at least 6
A5 In all solutions, let G be a finite group of order m. edges; hence a1 +· · ·+an +6(V −n) ≤ 2E. Combining
this inequality with the previous two equations gives
First solution: By Lagrange’s theorem, if m is not
divisible by p, then n = 0. Otherwise, let S be
the set of p-tuples (a0 , . . . , ap−1 ) ∈ Gp such that a1 + · · · + an ≤ 2E + 6n − 6(1 − F + E)
a0 · · · ap−1 = e; then S has cardinality mp−1 , which = 4n − 6,
is divisible by p. Note that this set is invariant under
cyclic permutation, that is, if (a0 , . . . , ap−1 ) ∈ S, then
as claimed.
(a1 , . . . , ap−1 , a0 ) ∈ S also. The fixed points under
this operation are the tuples (a, . . . , a) with ap = e; all Now set A3 = 1 and An = An−1 + 2n − 3 for n ≥ 4;
other tuples can be grouped into orbits under cyclic per- we will prove by induction on n that T has at most An
mutation, each of which has size p. Consequently, the triangles. For n = 3, since a1 + a2 + a3 = 6, a1 =
number of a ∈ G with ap = e is divisible by p; since a2 = a3 = 2 and hence T consists of just one triangle.
that number is n + 1 (only e has order 1), this proves
the claim. Next assume that an admissible triangulation of an
Second solution: (by Anand Deopurkar) Assume that (n − 1)-gon has at most An−1 triangles, and let T
n > 0, and let H be any subgroup of G of order p. Let be an admissible triangulation of an n-gon. If any
S be the set of all elements of G\H of order dividing p, ai = 2, then we can remove the triangle of T contain-
and let H act on G by conjugation. Each orbit has size ing vertex vi to obtain an admissible triangulation of an
p except for those which consist of individual elements (n−1)-gon; then the number of triangles in T is at most
g which commute with H. For each such g, g and H An−1 + 1 < An by induction. Otherwise, all ai ≥ 3.
generate an elementary abelian subgroup of G of order Now the average of a1 , . . . , an is less than 4, and thus
p2 . However, we can group these g into sets of size there are more ai = 3 than ai ≥ 5. It follows that
p2 − p based on which subgroup they generate together there is a sequence of k consecutive vertices in P whose
with H. Hence the cardinality of S is divisible by p; degrees are 3, 4, 4, . . . , 4, 3 in order, for some k with
adding the p− 1 nontrivial elements of H gives n ≡ −1 2 ≤ k ≤ n − 1 (possibly k = 2, in which case there are
(mod p) as desired. no degree 4 vertices separating the degree 3 vertices). If
we remove from T the 2k − 1 triangles which contain at
Third solution: Let S be the set of elements in G hav-
least one of these vertices, then we are left with an ad-
ing order dividing p, and let H be an elementary abelian
missible triangulation of an (n − 1)-gon. It follows that
p-group of maximal order in G. If |H| = 1, then we
there are at most An−1 +2k−1 ≤ An−1 +2n−3 = An
are done. So assume |H| = pk for some k ≥ 1, and
triangles in T . This completes the induction step and
let H act on S by conjugation. Let T ⊂ S denote
the proof.
the set of fixed points of this action. Then the size
of every H-orbit on S divides pk , and so |S| ≡ |T | Remark: We can refine the bound An somewhat. Sup-
(mod p). On the other hand, H ⊂ T , and if T con- posing that ai ≥ 3 for all i, the fact that a1 + · · · + an ≤
tained an element not in H, then that would contradict 4n − 6 implies that there are at least six more indices
the maximality of H. It follows that H = T , and so i with ai = 3 than with ai ≥ 5. Thus there exist six
|S| ≡ |T | = |H| = pk ≡ 0 (mod p), i.e., |S| = n + 1 sequences with degrees 3, 4, . . . , 4, 3, of total length at
is a multiple of p. most n + 6. We may thus choose a sequence of length
Remark: This result is a theorem of Cauchy; the first k ≤ ⌊ n6 ⌋ + 1, so we may improve the upper bound to
solution above is due to McKay. A more general (and An = An−1 + 2⌊ n6 ⌋ + 1, or asymptotically 61 n2 .
more difficult) result was proved by Frobenius: for any
positive integer m, if G is a finite group of order divis- However (as noted by Noam Elkies), a hexagonal
ible by m, then the number of elements of G of order swatch of a triangular lattice, with the boundary as close
dividing m is a multiple of m. to regular as possible, achieves asymptotically 16 n2 tri-
angles.
A6 For an admissible triangulation T , number the vertices
of P consecutively v1 , . . . , vn , and let ai be the number
of edges in T emanating from vi ; note that ai ≥ 2 for B1 The problem fails if f is allowed to be constant, e.g.,
all i. take f (n) = 1. We thus assume that f is nonconstant.
Remark: Variants of this solution are possible that con- For an upper bound on f (n), we use the inequalities
struct the Pi less explicitly, using Lagrange interpola- 0 ≤ ai ≤ n!/i! to deduce that there are at most n!/i! +
tion or Vandermonde determinants. 1 ≤ 2(n!/i!) choices for ai . Hence
B6 (Suggested by Oleg Golberg) Assume n ≥ 2, or else n! n!
the problem is trivially false. Throughout this proof, f (n) ≤ 2n ···
1! n!
any Ci will be a positive constant whose exact value is
= 2n 21 32 · · · nn−1
immaterial. As in the proof of Stirling’s approximation, 2
/2+C3 n −n2 /4
we estimate for any fixed c ∈ R, ≤ nn e .
n
X 1 2 1 For a lower bound on f (n), we note that if 0 ≤ ai <
(i + c) log i = n log n − n2 + O(n log n)
2 4 (n − 1)!/i! for i = 2, . . . , n − 1 and an = 0, then 0 ≤
i=1
a2 2!+· · ·+an n! ≤ n!, so there is a unique choice of a1
by comparing the sum to an integral. This gives to complete this to a solution of a1 1! + · · · + an n! = n!.
Hence
2
/2−C1 n −n2 /4
nn e ≤ 11+c 22+c · · · nn+c
(n − 1)! (n − 1)!
2
/2+C2 n −n2 /4 f (n) ≥ ···
≤ nn e . 2! (n − 1)!
We now interpret f (n) as counting the number of n- = 31 42 · · · (n − 1)n−3
2
/2+C4 n −n2 /4
tuples (a1 , . . . , an ) of nonnegative integers such that ≥ nn e .
a1 1! + · · · + an n! = n!.
A1 The function g(x) = f (x, 0) works. Substituting bers a′h+1 , . . . , a′n are divisible by pm while the num-
(x, y, z) = (0, 0, 0) into the given functional equation bers a′1 , . . . , a′h are not. Repeating this argument for
yields f (0, 0) = 0, whence substituting (x, y, z) = each pair (p, m) such that pm divides the initial product
(x, 0, 0) yields f (x, 0) + f (0, x) = 0. Finally, substi- a1 , . . . , an , we can determine the exact prime factoriza-
tuting (x, y, z) = (x, y, 0) yields f (x, y) = −f (y, 0) − tion of each of a′1 , . . . , a′n . This proves that the final
f (0, x) = g(x) − g(y). sequence is unique.
Remark: A similar argument shows that the possible Remark: (by David Savitt and Noam Elkies) Here are
functions g are precisely those of the form f (x, 0) + c two other ways to prove the termination. One is to ob-
for some c. serve that j ajj is strictly increasing at each step, and
Q
A2 Barbara wins using one of the following strategies. bounded above by (a1 · · · an )n . The other is to notice
that a1 is nonincreasing but always positive, so even-
First solution: Pair each entry of the first row with the
tually becomes constant; then a2 is nonincreasing but
entry directly below it in the second row. If Alan ever
always positive, and so on.
writes a number in one of the first two rows, Barbara
writes the same number in the other entry in the pair. If Reinterpretation: For each p, consider the sequence
Alan writes a number anywhere other than the first two consisting of the exponents of p in the prime factoriza-
rows, Barbara does likewise. At the end, the resulting tions of a1 , . . . , an . At each step, we pick two positions
matrix will have two identical rows, so its determinant i and j such that the exponents of some prime p are in
will be zero. the wrong order at positions i and j. We then sort these
Second solution: (by Manjul Bhargava) Whenever two position into the correct order for every prime p si-
Alan writes a number x in an entry in some row, Bar- multaneously.
bara writes −x in some other entry in the same row. At It is clear that this can only terminate with all se-
the end, the resulting matrix will have all rows summing quences being sorted into the correct order. We must
to zero, so it cannot have full rank. still check that the process terminates; however, since
all but finitely many of the exponent sequences consist
A3 We first prove that the process stops. Note first that
of all zeroes, and each step makes a nontrivial switch
the product a1 · · · an remains constant, because aj ak =
in at least one of the other exponent sequences, it is
gcd(aj , ak ) lcm(aj , ak ). Moreover, the last number in
enough to check the case of a single exponent sequence.
the sequence can never decrease, because it is always
This can be done as in the first solution.
replaced by its least common multiple with another
number. Since it is bounded above (by the product of Remark: Abhinav Kumar suggests the following proof
all of the numbers), the last number must eventually that the process always terminates in at most n2 steps.
reach its maximum value, after which it remains con- (This is a variant of the worst-case analysis of the bub-
stant throughout. After this happens, the next-to-last ble sort algorithm.)
number will never decrease, so it eventually becomes Consider the number of pairs (k, l) with 1 ≤ k < l ≤ n
constant, and so on. After finitely many steps, all of the such that ak does not divide al (call these bad pairs). At
numbers will achieve their final values, so no more steps each step, we find one bad pair (i, j) and eliminate it,
will be possible. This only happens when aj divides ak and we do not touch any pairs that do not involve either
for all pairs j < k. i or j. If i < k < j, then neither of the pairs (i, k)
We next check that there is only one possible final and (k, j) can become bad, because ai is replaced by a
sequence. For p a prime and m a nonnegative in- divisor of itself, while aj is replaced by a multiple of
teger, we claim that the number of integers in the itself. If k < i, then (k, i) can only become a bad pair
list divisible by pm never changes. To see this, sup- if ak divided ai but not aj , in which case (k, j) stops
pose we replace aj , ak by gcd(aj , ak ), lcm(aj , ak ). If being bad. Similarly, if k > j, then (i, k) and (j, k)
neither of aj , ak is divisible by pm , then neither of either stay the same or switch status. Hence the number
gcd(aj , ak ), lcm(aj , ak ) is either. If exactly one aj , ak of bad pairs goes down by at least 1 each time; since it
is divisible by pm , then lcm(aj , ak ) is divisible by pm is at most n2 to begin with, this is an upper bound for
but gcd(aj , ak ) is not. the number of steps.
gcd(aj , ak ), lcm(aj , ak ) are as well. Remark: This problem is closely related to the classi-
If we started out with exactly h numbers not divisible fication theorem for finite abelian groups. Namely, if
by pm , then in the final sequence a′1 , . . . , a′n , the num- a1 , . . . , an and a′1 , . . . , a′n are the sequences obtained
A6 For notational convenience, we will interpret the prob- as then we have n − m < 1 and hence H = G.
lem as allowing the empty subsequence, whose product To obtain this last inequality, put
is the identity element of the group. To solve the prob-
lem in the interpretation where the empty subsequence k = ⌊2 log2 n⌋ < (2/ ln 2) ln n,
is not allowed, simply append the identity element to
so that 2k+1 ≥ n2 . From the facts that ln n ≤ ln 2 +
the sequence given by one of the following solutions.
(n − 2)/2 ≤ n/2 and ln(1 − 1/n) < −1/n for all
First solution: Put n = |G|. We will say that a se- n ≥ 2, we have
quence S produces an element g ∈ G if g occurs as the
n2
product of some subsequence of S. Let H be the set of 1 n
2k ln 1 − <− = − < − ln n,
elements produced by the sequence S. n 2n 2
Start with S equal to the empty sequence. If at any point yielding the desired inequality.
the set H −1 H = {h1 h2 : h−11 , h2 ∈ H} fails to be all
Remark: An alternate approach in the second solution
of G, extend S by appending an element g of G not
is to distinguish betwen the cases of H small (i.e., m <
in H −1 H. Then Hg ∩ H must be empty, otherwise
n1/2 , in which case m can be replaced by a value no
there would be an equation of the form h1 g = h2 with
less than 2m − 1) and H large. This strategy is used in
h1 , h2 ∈ G, or g = h−1 1 h2 , a contradiction. Thus we
a number of recent results of Bourgain, Tao, Helfgott,
can extend S by one element and double the size of H.
and others on small doubling or small tripling of subsets
After k ≤ log2 n steps, we must obtain a sequence of finite groups.
S = a1 , . . . , ak for which H −1 H = G. Then the se-
In the second solution, if we avoid the rather weak in-
quence a−1 −1
k , . . . , a1 , a1 , . . . , ak produces all of G and equality ln n ≤ n/2, we instead get sequences of length
has length at most (2/ ln 2) ln n.
log2 (n ln n) = log2 (n) + log2 (ln n). This is close to
Second solution: optimal: one cannot use fewer than log2 n terms be-
Put m = |H|. We will show that we can append one cause the number of subsequences must be at least n.
element g to S so that the resulting sequence of k + 1
B1 There are at most two such points. For example,
elements will produce at least 2m − m2 /n elements of
the points (0, 0) and (1, 0) lie on a circle with center
G. To see this, we compute
(1/2, x) for any real number x, not necessarily rational.
On the other hand, suppose P = (a, b), Q =
X X
|H ∪ Hg| = (|H| + |Hg| − |H ∩ Hg|)
g∈G g∈G (c, d), R = (e, f ) are three rational points that lie on
X a circle. The midpoint M of the side P Q is ((a +
= 2mn − |H ∩ Hg| c)/2, (b + d)/2), which is again rational. Moreover, the
g∈G slope of the line P Q is (d−b)/(c−a), so the slope of the
= 2mn − |{(g, h) ∈ G2 : h ∈ H ∩ Hg}| line through M perpendicular to P Q is (a − c)/(b − d),
X which is rational or infinite.
= 2mn − |{g ∈ G : h ∈ Hg}|
h∈H
Similarly, if N is the midpoint of QR, then N is a ratio-
X nal point and the line through N perpendicular to QR
= 2mn − |H −1 h| has rational slope. The center of the circle lies on both
h∈H of these lines, so its coordinates (g, h) satisfy two linear
2 equations with rational coefficients, say Ag + Bh = C
= 2mn − m .
and Dg + Eh = F . Moreover, these equations have a
By the pigeonhole principle, we have |H ∪Hg| ≥ 2m− unique solution. That solution must then be
m2 /n for some choice of g, as claimed.
g = (CE − BD)/(AE − BD)
In other words, by extending the sequence by one el-
ement, we can replace the ratio s = 1 − m/n (i.e., h = (AF − BC)/(AE − BD)
and as a1 + · · · + a2k+1 = 0
from the (k + 1)-st coordinate, we deduce that all of the
j−1
X ai must be zero, and so Mn,k must be invertible.
rk+1−b+(2k+1)i .
i=0 Remark: The matrices Mn,k are examples of banded
matrices, which occur frequently in numerical appli-
Hence Mn,k is not invertible. cations of linear algebra. They are also examples of
We next check that if n ≡ 0, 1 (mod 2k + 1), then Toeplitz matrices.
Mn,k is invertible. Suppose that a1 , . . . , an are scalars