Professional Documents
Culture Documents
A1 The answer is 578. the error term O(x1−1/r ) is bounded in absolute value by
Each hop corresponds to adding one of the 12 vectors (r + 1)r(x + 1)r−1 /xr . For x ≥ 1, r ≤ 1 this quantity is
(0, ±5), (±5, 0), (±3, ±4), (±4, ±3) to the position bounded in absolute value by 2(r + 1)r, independently
of the grasshopper. Since (2021, 2021) = 288(3, 4) + of x. This allows us to continue by interchanging the
288(4, 3) + (0, 5) + (5, 0), the grasshopper can reach order of the limits, obtaining
(2021, 2021) in 288 + 288 + 1 + 1 = 578 hops.
lim lim (r + 1 + O(x1−1/r ))1/r
On the other hand, let z = x + y denote the sum of the x r→0 x→∞
and y coordinates of the grasshopper, so that it starts at = lim (r + 1)1/r
z = 0 and ends at z = 4042. Each hop changes the sum r→0
of the x and y coordinates of the grasshopper by at most = lim (1 + 1/s)s = e,
s→∞
7, and 4042 > 577 × 7; it follows immediately that the
grasshopper must take more than 577 hops to get from where in the last step we take s = 1/r.
(0, 0) to (2021, 2021).
Remark. This solution implicitly uses the distance A3 The integers N with this property are those of the form
function 3m2 for some positive integer m.
In one direction, for N = 3m2 , the points
d((x1 , y1 ), (x2 , y2 )) = |x1 − x2 | + |y1 − y2 |
on the plane, variously called the taxicab metric, the (m, m, m), (m, −m, −m), (−m, m, −m), (−m, −m, m)
Manhattan metric, or the L1 -norm (or `1 -norm).
form the vertices of a regular tetrahedron inscribed in
A2 The limit is e. the sphere x2 + y2 + z2 = N.
First solution. By l’Hôpital’s Rule, we have Conversely, suppose that Pi = (xi , yi , zi ) for i = 1, . . . , 4
log((x + 1)r+1 − xr+1 ) are the vertices of an inscribed regular tetrahedron.
lim Then the center of this tetrahedron must equal the cen-
r→0 r
ter of the sphere, namely (0, 0, 0). Consequently, these
d
= lim log((x + 1)r+1 − xr+1 ) four vertices together with Qi = (−xi , −yi , −zi ) for i =
r→0 dr 1, . . . , 4 form the vertices of an inscribed cube in the
(r + 1)((x + 1)r+1 log(x + 1) − xr+1 log x) sphere. For each face of that cube, take the sum of
= lim
r→0 (x + 1)r+1 − xr+1 the vectors corresponding to the four corners of that
= (x + 1) log(x + 1) − x log x, face; this yields six vectors with integer coordinates
with form the vertices ofpan octahedron inscribed in the
where log denotes natural logarithm. It follows that sphere x2 + y2 + z2 = 4 N/3. Choose three mutually
x+1
g(x) = e(x+1) log(x+1)−x log x = (x+1)
xx . Thus perpendicular vectors from these six and let A be the
x matrix with these row vectors; then
g(x) x+1 1
lim = lim · lim 1 + = 1 · e = e. 16N 16N
x→∞ x x→∞ x x→∞ x AAT = I3 =⇒ det(A)2 = .
3 3
Second solution. We first write
Hence N is divisible by 3 and N/3 is a perfect square,
g(x) ((x + 1)r+1 − xr+1 )1/r as claimed.
lim = lim lim √
x→∞ x x→∞ r→0 x 2
A4 The limit exists and equals 2 π log 2.
((r + 1)x + O(xr−1 ))1/r
r
= lim lim . We first note that we can interchange x and y to obtain
x→∞ r→0 x
We would like to interchange the order of the limits, but ZZ
1 + 2y2 1 + x2
this requires some justification. Using Taylor’s theorem I(R) = 4 2 2 4
− dx dy.
x2 +y2 ≤R2 1 + x + 6x y + y 2 + x 4 + y4
with remainder, for x ≥ 1 we can bound the error term
O(xr−1 ) in absolute value by (r + 1)r(x + 1)r−1 . This Averaging the two expressions for I(R) yields
means that if we continue to rewrite the orginial limit as
ZZ
lim lim (r + 1 + O(x1−1/r ))1/r , I(R) = ( f (x, y) − g(x, y)) dx dy
r→0 x→∞ x2 +y2 ≤R2
2
3 + cos 4θ
dθ = 2
3 + cos θ
dθ . If α 6= 0 and Re(α) ≥ 0, then Re(1 + an−1 α −1 ) ≥ 1 and
0 0
– If the argument of α belongs to [−π/4, π/4], then dropped square is at distance less than 1/2 from that
Re(α −1 ), Re(α −2 ) ≥ 0, so side of Sθ . To check the converse, note that there are
two ways to dissect the square Sθ into the square Sθ0 plus
|α|−3 four sin θ × cos θ rectangles. If θ 6= 0, π/4, then one of
1 ≤ |α|−3 + · · · + |α|−n < .
1 − |α|−1 these dissections has the property that each corner P of
S appears as an interior point of a side (not a corner) of
Hence |α|−1 is greater than the unique positive one of the rectangles R. It will suffice to check that if the
root of x3 + x − 1, which is greater than 2/3. center of the dropped square is in R, then the dropped
square covers P; this follows from the fact that sin θ and
– Otherwise, α has argument in (−π/2, √ π/4) ∪ cos θ are both at most 1.
(π/4, π/2), so the bound
√ |α| <√(1 + 5)/2 im-
plies that Re(α) < (1 + 5)/(2 2) < 3/2. It follows that the conditional probability, given that the
angle of rotation is chosen to be θ , that the dropped
By hypothesis, there exists a factorization P(x) = square does not cover any corners of S is (sin θ +
Q(x)R(x) into two nonconstant integer polynomials, cos θ − 1)2 . We then compute the original probability
which we may assume are monic. Q(x + 3/2) is a prod- as the integral
uct of polynomials, each of the form x − α where α is a
2
Z π/2
real root of P or of the form (sin θ + cos θ − 1)2 dθ
π 0
3 3
x+ −α x+ −α 2
Z π/2
2 2 = (2 + sin 2θ − 2 sin θ − 2 cos θ ) dθ
2 π 0
3 3 π/2
= x2 + 2Re
− α x + − α 2 1
2 2 = 2θ − cos 2θ + 2 cos θ − 2 sin θ
π 2 0
where α is a nonreal root of P. It follows that Q(x + 2 6
= (π + 1 − 2 − 2) = 2 − .
3/2) has positive coefficients; comparing its values at π π
x = 1/2 and x = −1/2 yields Q(2) > Q(1). We can-
not have Q(1) ≤ 0, as otherwise the intermediate value B2 The answer is 2/3.
theorem would imply that Q has a real root in [1, ∞); By AM-GM, we have
hence Q(1) ≥ 1 and so Q(2) ≥ 2. Similarly R(2) ≥ 2,
1/n
so P(2) = Q(2)R(2) is composite. 2n+1 (a1 · · · an )1/n = (4a1 )(42 a2 ) · · · (4n an )
Remark. A theorem of Brillhart, Filaseta, and Odlyzko ∑nk=1 (4k ak )
from 1981 states that if a prime p is written as ∑i ai bi ≤ .
n
in any base b ≥ 2, the polynomial ∑i ai xi is irreducible.
(The case b = 10 is an older result of Cohn.) The solu- Thus
tion given above is taken from: Ram Murty, Prime num-
bers and irreducible polynomials, Amer. Math. Monthly ∑nk=1 (4k ak )
∞
2S ≤ ∑
109 (2002), 452–458). The final step is due to Pólya and n=1 4n
Szegő. ∞ n ∞ ∞
= ∑ ∑ (4k−n ak ) = ∑ ∑ (4k−n ak )
B1 The probability is 1/π. n=1 k=1 k=1 n=k
∞
Set coordinates so that the original tiling includes the 4ak 4
= ∑ =
(filled) square S = {(x, y) : 0 ≤ x, y ≤ 1}. It is then k=1 3 3
equivalent to choose the second square by first choosing
a point uniformly at random in S to be the center of the and S ≤ 2/3. Equality is achieved when ak = 43k for all
square, then choosing an angle of rotation uniformly at k, since in this case 4a1 = 42 a2 = · · · = 4n an for all n.
random from the interval [0, π/2].
B3 We prove the given statement.
For each θ ∈ [0, π/2], circumscribe a square Sθ around
S with angle of rotation θ relative to S; this square has For any circle S of radius r whose center is at distance
side length sin θ + cos θ . Inside Sθ , draw the smaller d from the origin, express the integral in polar coordi-
square Sθ0 consisting of points at distance greater than nates s, θ :
1/2 from each side of Sθ ; this square has side length ZZ Z s2 Z θ2 (s)
sin θ + cos θ − 1. ρ= (yhx − xhy )(s sin θ , s cos θ )s dθ ds.
S s1 θ1 (s)
We now verify that a unit square with angle of rotation
θ fails to cover any corners of S if and only if its center For fixed s, the integral over θ is a line integral of grad h,
lies in the interior of Sθ0 . In one direction, if one of the which evaluates to h(P2 ) − h(P1 ) where P1 , P2 are the
corners of S is covered, then that corner lies on a side of endpoints of the endpoints of the arc of the circle of ra-
Sθ which meets the dropped square, so the center of the dius s centered at the origin lying within S . If we now
4
The expected value of |Xk − 21 | equals (This can also be interpreted as an instance of the rear-
rangement inequality.)
Z 1/2 We now see that
1
µk = 2 − x fk (x) dx
0 2 Z 1/2
Z 1/2
1
µk ≥ 2 xgk (x) dx
=2 x fk − x dx 0
0 2 k Z (1/2)(2/3)k
3
≥2 x dx
2 0
Define the function k
1 2 (1/2)(2/3)
k
3
h i =2 x
3 k 2 k 2 2 0
(
x ∈ 0, 21
gk (x) = 2 3 k 2k
1 2 k
0 otherwise. 3 1 2
=2 =
2 8 3 4 3
Note that for x ∈ [0, 1/2] we have as desired.
Z x Remark. For comparison, if we instead take the me-
(gk (t) − fk (1/2 − t)) dt ≥ 0 dian of a list of n numbers, the probability distribution
0 is given by
with equality at x = 0 or x = 1/2. (On the in- P2n+1 (x) = (2n + 1)!/(n!n!)xn (1 − x)n .
terval [0, (1/2)(2/3)k ] the integrand is nonnegative,
so the function increases from 0; on the interval The expected value of the absolute difference between
[(1/2)(2/3)k , 1/2] the integrand is nonpositive, so the 1/2 and the median is
function decreases to 0.) Hence by integration by parts, Z 1/2
2n + 1
2 (1/2 − x)P2n+1 (x)dx = 2−2n−2 .
Z 1/2 0 n
µk − 2 xgk (x) dx
0 For n = 32021 , using Stirling’s approximation this can be
1/2 estimated as 1.13(0.577)2021 < 0.25(0.667)2021 . This
1
Z
= 2x( fk − x − gk (x)) dx
0 2 shows that the trimming procedure produces a quantity
Z 1/2 Z x Z x that is on average further away from 1/2 than the me-
2 1
= x gk (t) − fk − t dt dt dx ≥ 0. dian.
0 0 0 2