Solutions to the 58th William Lowell Putnam Mathematical Competition

Saturday, December 6, 1997

Manjul Bhargava, Kiran Kedlaya, and Lenny Ng

A–1 The centroid G of the triangle is collinear with H and O A–3 Note that the series on the left is simply x exp(−x2 /2).
(Euler line), and the centroid lies two-thirds of the way By integration by parts,
from A to M. Therefore H is also two-thirds of the way Z ∞ Z ∞
2 /2 2 /2
from A to F, so AF = 15. Since the triangles BFH and x2n+1 e−x dx = 2n x2n−1 e−x dx
AFC are similar (they’re right triangles and 0 0

∠HBC = π/2 − ∠C = ∠CAF), and so by induction,

Z ∞
2 /2
we have x2n+1 e−x dx = 2 × 4 × · · · × 2n.
0
BF/FH = AF/FC Thus the desired integral is simply
or ∞
1 √
∑ 2n n! = e.
BF · FC = FH · AF = 75. n=0

Now A–4 In order to have ψ(x) = aφ (x) for all x, we must in par-
ticular have this for x = e, and so we take a = φ (e)−1 .
BC2 = (BF + FC)2 = (BF − FC)2 + 4BF · FC, We first note that
but φ (g)φ (e)φ (g−1 ) = φ (e)φ (g)φ (g−1 )
BF − FC = BM + MF − (MC − MF) = 2MF = 22, and so φ (g) commutes with φ (e) for all g. Next, we
note that
so
p √ φ (x)φ (y)φ (y−1 x−1 ) = φ (e)φ (xy)φ (y−1 x−1 )
BC = 222 + 4 · 75 = 784 = 28.
and using the commutativity of φ (e), we deduce
A–2 We show more precisely that the game terminates with
one player holding all of the pennies if and only if n = φ (e)−1 φ (x)φ (e)−1 φ (y) = φ (e)−1 φ (xy)
2m +1 or n = 2m +2 for some m. First suppose we are in
the following situation for some k ≥ 2. (Note: for us, a or ψ(xy) = ψ(x)ψ(y), as desired.
“move” consists of two turns, starting with a one-penny
pass.) A–5 We may discard any solutions for which a1 6= a2 , since
those come in pairs; so assume a1 = a2 . Similarly, we
– Except for the player to move, each player has k may assume that a3 = a4 , a5 = a6 , a7 = a8 , a9 = a10 .
pennies; Thus we get the equation
– The player to move has at least k pennies. 2/a1 + 2/a3 + 2/a5 + 2/a7 + 2/a9 = 1.
We claim then that the game terminates if and only if
the number of players is a power of 2. First suppose Again, we may assume a1 = a3 and a5 = a7 , so we get
the number of players is even; then after m complete 4/a1 + 4/a5 + 2/a9 = 1; and a1 = a5 , so 8/a1 + 2/a9 =
rounds, every other player, starting with the player who 1. This implies that (a1 − 8)(a9 − 2) = 16, which by
moved first, will have m more pennies than initially, and counting has 5 solutions. Thus N10 is odd.
the others will all have 0. Thus we are reduced to the A–6 Clearly xn+1 is a polynomial in c of degree n, so it suf-
situation with half as many players; by this process, we fices to identify n values of c for which xn+1 = 0. We
eventually reduce to the case where the number of play- claim these are c = n − 1 − 2r for r = 0, 1, . . . , n − 1;
ers is odd. However, if there is more than one player, in this case, xk is the coefficient of t k−1 in the polyno-
after two complete rounds everyone has as many pen- mial f (t) = (1 − t)r (1 + t)n−1−r . This can be verified
nies as they did before (here we need m ≥ 2), so the by noticing that f satisfies the differential equation
game fails to terminate. This verifies the claim.
Returning to the original game, note that after one com- f 0 (t) n − 1 − r r
= −
plete round, b n−1
2 c players remain, each with 2 pennies
f (t) 1+t 1−t
except for the player to move, who has either 3 or 4 pen-
nies. Thus by the above argument, the game terminates
if and only if b n−1
2 c is a power of 2, that is, if and only
if n = 2m + 1 or n = 2m + 2 for some m.
 2

(by logarithmic differentiation) or equivalently, It is immediate (e.g., by induction) that In ≡

1, −1, 1, 0, 0 (mod 5) for n ≡ 1, 2, 3, 4, 5 (mod 5) respec-
(1 − t 2 ) f 0 (t) = f (t)[(n − 1 − r)(1 − t) − r(1 + t)] tively, and moreover, we have the equality
= f (t)[(n − 1 − 2r) − (n − 1)t]
k
1
Hn = ∑ I m ,
m bn/5 c
and then taking the coefficient of tk on both sides: m=0 5

(k + 1)xk+2 − (k − 1)xk = where k = k(n) denotes the largest integer such that
(n − 1 − 2r)xk+1 − (n − 1)xk . 5k ≤ n. We wish to determine those n such that the
above sum has nonnegative 5–valuation. (By the 5–
n−1
valuation of a number a we mean the largest integer v
In particular, the largest such c is n − 1, and xk = k−1
for k = 1, 2, . . . , n. such that a/5v is an integer.)
Greg Kuperberg has suggested an alternate approach to If bn/5k c ≤ 3, then the last term in the above sum
show directly that c = n − 1 is the largest root, without has 5–valuation −k, since I1 , I2 , I3 each have valua-
computing the others. Note that the condition xn+1 = 0 tion 0; on the other hand, all other terms must have 5–
states that (x1 , . . . , xn ) is an eigenvector of the matrix valuation strictly larger than −k. It follows that Hn has
 5–valuation exactly −k; in particular, Hn has nonneg-
 i j = i+1 ative 5–valuation in this case if and only if k = 0, i.e.,
Ai j = n − j j = i − 1 n = 1, 2, or 3.
 0 otherwise
Suppose now that bn/5k c = 4. Then we must also have
20 ≤ bn/5k−1 c ≤ 24. The former condition implies that
with eigenvalue c. By the Perron-Frobenius theorem, the last term of the above sum is I4 /5k = 1/(12 · 5k−2 ),
A has a unique eigenvector with positive entries, whose which has 5–valuation −(k − 2).
eigenvalue has modulus greater than or equal to that of
any other eigenvalue, which proves the claim. It is clear that I20 ≡ I24 ≡ 0 (mod 25); hence if bn/5k−1 c
equals 20 or 24, then the second–to–last term of the
m m m
B–1 It is trivial to check that 6n = { 6n } ≤ { 3n } for 1 ≤ above sum (if it exists) has valuation at least −(k −
m m m
m ≤ 2n, that 1 − 3n = { 3n } ≤ { 6n } for 2n ≤ m ≤ 3n, 3). The third–to–last term (if it exists) is of the form
m m m
that 3n − 1 = { 3n } ≤ { 6n } for 3n ≤ m ≤ 4n, and that Ir /5k−2 , so that the sum of the last term and the third to
m m m
1 − 6n = { 6n } ≤ { 3n } for 4n ≤ m ≤ 6n. Therefore the last term takes the form (Ir + 1/12)/5k−2 . Since Ir can
desired sum is be congruent only to 0,1, or -1 (mod 5), and 1/12 ≡ 3
(mod 5), we conclude that the sum of the last term
2n−1
m 3n−1  m and third–to–last term has valuation −(k − 2), while
∑ + ∑ 1−
m=1 6n m=2n 3n all other terms have valuation strictly higher. Hence
4n−1   6n−1  Hn has nonnegative 5–valuation in this case only when
m m k ≤ 2, leading to the values n = 4 (arising from k = 0),
+ ∑ −1 + ∑ 1− = n.
m=3n 3n m=4n 6n 20,24 (arising from k = 1 and bn/5k−1 c = 20 and 24
resp.), 101, 102, 103, and 104 (arising from k = 2,
B–2 It suffices to show that | f (x)| is bounded for x ≥ 0, since bn/5k−1 c = 20) and 120, 121, 122, 123, and 124 (aris-
f (−x) satisfies the same equation as f (x). But then ing from k = 2, bn/5k−1 c = 24).
Finally, suppose bn/5k c = 4 and bn/5k−1 c = 21, 22,
d
( f (x))2 + ( f 0 (x))2 = 2 f 0 (x)( f (x) + f 00 (x)) or 23. Then as before, the first condition implies that


dx the last term of the sum in (*) has valuation −(k − 2),
= −2xg(x)( f 0 (x))2 ≤ 0, while the second condition implies that the second–to–
last term in the same sum has valuation −(k −1). Hence
so that ( f (x))2 ≤ ( f (0))2 + ( f 0 (0))2 for x ≥ 0. all terms in the sum (*) have 5–valuation strictly higher
than −(k − 1), except for the second–to–last term, and
B–3 The only such n are the numbers 1–4, 20–24, 100–104, therefore Hn has 5–valuation −(k − 1) in this case. In
and 120–124. For the proof let particular, Hn is integral (mod 5) in this case if and only
n if k ≤ 1, which gives the additional values n = 21, 22,
1
Hn = ∑ and 23.
m=1 m
B–4 Let sk = ∑i (−1)i ak−1,i be the given sum (note that ak−1,i
and introduce the auxiliary function is nonzero precisely for i = 0, . . . , b 2k
3 c). Since
1 am+1,n = am,n + am,n−1 + am,n−2 ,
In = ∑ .
1≤m≤n,(m,5)=1
m
 3

we have and that AD < CD. In any dissection of the triangle

into four parts, some two of A, B,C, D, E must belong to
sk − sk−1 + sk+2 = ∑(−1)i (an−i,i + an−i,i+1 + an−i,i+2 ) the same part, forcing the least diameter to be at least
i 25/13.
= ∑(−1)i an−i+1,i+2 = sk+3 . We now exhibit a dissection with least diameter 25/13.
i
(Some variations of this dissection are possible.) Put
By computing s0 = 1, s1 = 1, s2 = 0, we may easily F = (15/13, 19/13), G = (15/13, 0), H = (0, 19/13),
verify by induction that s4 j = s4 j+1 = 1 and s4 j+2 = J = (32/15, 15/13), and divide ABC into the convex
s4 j+3 = 0 for all j ≥ 0. (Alternate solution suggested by polygonal regions ADFH, BEJ, CGFH, DFGEJ. To
John Rickert: write S(x, y) = ∑∞ 2 2 3 i check that this dissection has least diameter 25/13, it
i=0 (y + xy + x y ) , and
note note that sk is the coefficient of yk in S(−1, y) = suffices (by the following remark) to check that the dis-
(1 + y)/(1 − y4 ).) tances

B–5 Define the sequence x1 = 2, xn = 2xn−1 for n > 1. It suf- AD, AF, AH, BE, BJ, DE,CF,CG,CH,
fices to show that for every n, xm ≡ xm+1 ≡ · · · (mod n) DF, DG, DH, DJ, EF, EG, EJ, FG, FH, FJ, GJ
for some m < n. We do this by induction on n, with
n = 2 being obvious. are all at most 25/13. This can be checked by a long
Write n = 2a b,where b is odd. It suffices to show that numerical calculation, which we omit in favor of some
xm ≡ · · · modulo 2a and modulo b, for some m < n. shortcuts: note that ADFH and BEJ are contained in
For the former, we only need xn−1 ≥ a, but clearly circular sectors centered at A and B, respectively, of ra-
xn−1 ≥ n by induction on n. For the latter, note that dius 25/13 and angle less than π/3, while CGFH is a
xm ≡ xm+1 ≡ · · · (mod b) as long as xm−1 ≡ xm ≡ · · · rectangle with diameter CF < 25/13.
(mod φ (b)), where φ (n) is the Euler totient function. Remark. The preceding argument uses implicitly the
By hypothesis, this occurs for some m < φ (b) + 1 ≤ n. fact that for P a simple closed polygon in the plane, if
(Thanks to Anoop Kulkarni for catching a lethal typo in we let S denote the set of points on or within P, then the
an earlier version.) maximum distance between two points of S occurs be-
tween some pair of vertices of P. This is an immediate
B–6 The answer is 25/13. Place the triangle on the cartesian consequence of the compactness of S (which guarantees
plane so that its vertices are at C = (0, 0), A = (0, 3), B = the existence of a maximum) and the convexity of the
(4, 0). Define also the points D = (20/13, 24/13), and function taking (x, y) ∈ S × S to the squared distance be-
E = (27/13, 0). We then compute that tween x and y (which is obvious in terms of Cartesian
25 coordinates).
= AD = BE = DE
13
27
= BC −CE = BE < BC
13
39 p
= AC < AC2 +CE 2 = AE
13
40
= AB − AD = BD < AB
13