HMMT February 2015

Saturday 21 February 2015

Guts
1. [4] Let R be the rectangle in the Cartesian plane with vertices at (0, 0), (2, 0), (2, 1), and (0, 1). R can
be divided into two unit squares, as shown.

The resulting figure has 7 segments of unit length, connecting neighboring lattice points (those lying
on or inside R). Compute the number of paths from (0, 1) (the upper left corner) to (2, 0) (the lower
right corner) along these 7 segments, where each segment can be used at most once.
Answer: 4 Just count them directly. If the first step is to the right, there are 2 paths. If the first
step is downwards (so the next step must be to the right), there are again 2 paths. This gives a total
of 4 paths.
2. [4] Let ABCDE be a convex pentagon such that ∠ABC = ∠ACD = ∠ADE = 90◦ and AB = BC =
CD = DE = 1. Compute AE.
Answer: 2 By Pythagoras,

AE 2 = AD2 + 1 = AC 2 + 2 = AB 2 + 3 = 4,

so AE = 2.
3. [4] Find the number of pairs of union/intersection operations (1 , 2 ) ∈ {∪, ∩}2 satisfying the fol-
lowing condition: for any sets S, T , function f : S → T , and subsets X, Y, Z of S, we have equality of
sets
f (X)1 (f (Y )2 f (Z)) = f (X1 (Y 2 Z)) ,
where f (X) denotes the image of X: the set {f (x) : x ∈ X}, which is a subset of T . The images f (Y )
(of Y ) and f (Z) (of Z) are similarly defined.
Answer: 1 If and only if 1 = 2 = ∪. See http://math.stackexchange.com/questions/359693/overview-of-b
4. [4] Consider the function z(x, y) describing the paraboloid

z = (2x − y)2 − 2y 2 − 3y.

Archimedes and Brahmagupta are playing a game. Archimedes first chooses x. Afterwards, Brah-
magupta chooses y. Archimedes wishes to minimize z while Brahmagupta wishes to maximize z.
Assuming that Brahmagupta will play optimally, what value of x should Archimedes choose?
Answer: − 83 Viewing x as a constant and completing the square, we find that

z = 4x2 − 4xy + y 2 − 2y 2 − 3y
= −y 2 − (4x + 3)y + 4x2
 2  2
4x + 3 4x + 3
=− y+ + + 4x2 .
2 2

Bhramagupta wishes to maximize z, so regardless of the value of x, he will pick y = − 4x+3

2 . The
expression for z then simplifies to
9
z = 8x2 + 6x + .
4

Guts
 Archimedes knows this and will therefore pick x to minimize the above expression. By completing the
square, we find that x = − 38 minimizes z.
Alternatively, note that z is convex in x and concave in y, so we can use the minimax theorem to switch
the order of moves. If Archimedes goes second, he will set x = y2 to minimize z, so Brahmagupta will
maximize −2y 2 − 3y by setting y = − 43 . Thus Archimedes should pick x = − 83 , as above.

5. [5] Let H be the unit hypercube of dimension 4 with a vertex at (x, y, z, w) for each choice of x, y, z, w ∈
{0, 1}. (Note that H has 24 = 16 vertices.) A bug starts at the vertex (0, 0, 0, 0). In how many ways
can the bug move to (1, 1, 1, 1) (the opposite corner of H) by taking exactly 4 steps along the edges of
H?
Answer: 24 You may think of this as sequentially adding 1 to each coordinate of (0, 0, 0, 0). There
are 4 ways to choose the first coordinate, 3 ways to choose the second, and 2 ways to choose the third.
The product is 24.

6. [5] Let D be a regular ten-sided polygon with edges of length 1. A triangle T is defined by choosing
three vertices of D and connecting them with edges. How many different (non-congruent) triangles T
can be formed?
Answer: 8 The problem is equivalent to finding the number of ways to partition 10 into a sum of
three (unordered) positive integers. These can be computed by hand to be (1, 1, 8), (1, 2, 7), (1, 3, 6),
(1, 4, 5), (2, 2, 6), (2, 3, 5), (2, 4, 4), (3, 3, 4).

7. [5] Let C be a cube of side length 2. We color each of the faces of C blue, then subdivide it into
23 = 8 unit cubes. We then randomly rearrange these cubes (possibly with rotation) to form a new
3-dimensional cube.
What is the probability that its exterior is still completely blue?
1 1 1
Answer: 224 or 88 or 16777216 Each vertex of the original cube must end up as a vertex of the
new cube in order for all the old blue faces to show. There are 8 such vertices, each corresponding to
one unit cube, and each has a probability 18 of being oriented with the old outer vertex as a vertex of
the new length-2 cube. Multiplying gives the answer.

8. [5] Evaluate
sin(arcsin(0.4) + arcsin(0.5)) · sin(arcsin(0.5) − arcsin(0.4)),
where for x ∈ [−1, 1], arcsin(x) denotes the unique real number y ∈ [−π, π] such that sin(y) = x.
9
Answer: 0.09 OR 100 Use the difference of squares identity1 sin(a−b) sin(a+b) = sin(a)2 −sin(b)2
9
to get 0.52 − 0.42 = 0.3 = 0.09 =
2
100 .

9. [6] Let a, b, c be integers. Define f (x) = ax2 + bx + c. Suppose there exist pairwise distinct integers
u, v, w such that f (u) = 0, f (v) = 0, and f (w) = 2. Find the maximum possible value of the
discriminant b2 − 4ac of f .
Answer: 16 By the factor theorem, f (x) = a(x − u)(x − v), so the constraints essentially boil
down to 2 = f (w) = a(w − u)(w − v). (It’s not so important that u 6= v; we merely specified it for a
shorter problem statement.)
We want to maximize the discriminant b2 −4ac = a2 [(u+v)2 −4uv] = a2 (u−v)2 = a2 [(w−v)−(w−u)]2 .
Clearly a | 2. If a > 0, then (w − u)(w − v) = 2/a > 0 means the difference |u − v| is less than 2/a,
whereas if a < 0, since at least one of |w − u| and |w − v| equals 1, the difference |u − v| of factors is
greater than 2/|a|.
So the optimal choice occurs either for a = −1 and |u − v| = 3, or a = −2 and |u − v| = 2. The latter
wins, giving a discriminant of (−2)2 · 22 = 16.
1 proven most easily with complex numbers or the product-to-sum identity on sin(a − b) sin(a + b) (followed by the double

angle formula for cosine)

Guts
10. [6] Let b(x) = x2 + x + 1. The polynomial x2015 + x2014 + · · · + x + 1 has a unique “base b(x)”
representation
N
X
x2015 + x2014 + · · · + x + 1 = ak (x)b(x)k ,
k=0

where

• N is a nonnegative integer;
• each “digit” ak (x) (for 0 ≤ k ≤ N ) is either the zero polynomial (i.e. ak (x) = 0) or a nonzero
polynomial of degree less than deg b = 2; and
• the “leading digit aN (x)” is nonzero (i.e. not the zero polynomial).

Find aN (0) (the “leading digit evaluated at 0”).

Answer: −1006 Comparing degrees easily gives N = 1007. By ignoring terms of degree at most
2013, we see
aN (x)(x2 + x + 1)1007 ∈ x2015 + x2014 + O(x2013 ).
Write aN (x) = ux + v, so

aN (x)(x2 + x + 1)1007 ∈ (ux + v)(x2014 + 1007x2013 + O(x2012 ))

⊆ ux2015 + (v + 1007u)x2014 + O(x2013 ).

Finally, matching terms gives u = 1 and v + 1007u = 1, so v = 1 − 1007 = −1006.

Remark. This problem illustrates the analogy between integers and polynomials, with the nonconstant
(degree ≥ 1) polynomial b(x) = x2 + x + 1 taking the role of a positive integer base b > 1.

11. [6] Find q

 
∞ 1 + 2000000 
X  4k 
 ,
2
k=0

where ⌊x⌋ denotes the largest integer less than or equal to x.

Answer: 1414 The kth floor (for k ≥ 0) counts the number of positive integer solutions to
4k (2x − 1)2 ≤ 2 · 106 . So summing over all k, we want the number of integer solutions to 4k (2x − 1)2 ≤
2 · 106 with k ≥ 0 and x ≥ 1. But each positive integer√can be uniquely represented as a power of 2
times an odd (positive) integer, so there are simply ⌊103 2⌋ = 1414 solutions.

12. [6] For integers a, b, c, d, let f (a, b, c, d) denote the number of ordered pairs of integers (x, y) ∈
{1, 2, 3, 4, 5}2 such that ax + by and cx + dy are both divisible by 5. Find the sum of all possible
values of f (a, b, c, d).
Answer: 31 Standard linear algebra over the field F5 (the integers modulo 5). The dimension of
the solution set is at least 0 and at most 2, and any intermediate value can also be attained. So the
answer is 1 + 5 + 52 = 31.
This also can be easily reformulated in more concrete equation/congruence-solving terms, especially
since there are few variables/equations.

13. [8] Let P (x) = x3 + ax2 + bx + 2015 be a polynomial all of whose roots are integers. Given that
P (x) ≥ 0 for all x ≥ 0, find the sum of all possible values of P (−1).
Answer: 9496 Since all the roots of P (x) are integers, we can factor it as P (x) = (x−r)(x−s)(x−t)
for integers r, s, t. By Viete’s formula, the product of the roots is rst = −2015, so we need three integers
to multiply to −2015.
P (x) cannot have two distinct positive roots u, v since otherwise, P (x) would be negative at least in
some infinitesimal region x < u or x > v, or P (x) < 0 for u < x < v. Thus, in order to have two

Guts
 positive roots, we must have a double root. Since 2015 = 5 × 13 × 31, the only positive double root is
a perfect square factor of 2015, which is at x = 1, giving us a possibility of P (x) = (x − 1)2 (x + 2015).
Now we can consider when P (x) only has negative roots. The possible unordered triplets are (−1, −1, −2015), (−1, −5, −4
(−1, −31, −65), (−5, −13, −31), which yield the polynomials
(x + 1)2 (x + 2015), (x + 1)(x + 5)(x + 403), (x + 1)(x + 13)(x + 155), (x + 1)(x + 31)(x + 65), (x + 5)(x +
13)(x + 31), respectively.
Noticing that P (−1) = 0 for four of these polynomials, we see that the nonzero values are P (−1) =
(−1 − 1)2 (2014), (5 − 1)(13 − 1)(31 − 1), which sum to 8056 + 1440 = 9496.
14. [8] Find the smallest integer n ≥ 5 for which there exists a set of n distinct pairs (x1 , y1 ), . . . , (xn , yn )
of positive integers with 1 ≤ xi , yi ≤ 4 for i = 1, 2, . . . , n, such that for any indices r, s ∈ {1, 2, . . . , n}
(not necessarily distinct), there exists an index t ∈ {1, 2, . . . , n} such that 4 divides xr + xs − xt and
yr + ys − yt .
Answer: 8 In other words, we have a set S of n pairs in (Z/4Z)2 closed under addition. Since
1 + 1 + 1 + 1 ≡ 0 (mod 4) and 1 + 1 + 1 ≡ −1 (mod 4), (0, 0) ∈ S and S is closed under (additive)
inverses. Thus S forms a group under addition (a subgroup of (Z/4Z)2 ). By Lagrange’s theorem
(from basic group theory), n | 42 , so n ≥ 8. To achieve this bound, one possible construction is
{1, 2, 3, 4} × {2, 4}.
Remark. In fact, S is a finite abelian group. Such groups have a very clean classification; this is
clarified by the fact that abelian groups are the same as modules over Z, the ring of integers.
15. [8] Find the maximum possible value of H · M · M · T over all ordered triples (H, M, T ) of integers
such that H · M · M · T = H + M + M + T .
Answer: 8 If any of H, M, T are zero, the product is 0. We can do better (examples below), so
we may now restrict attention to the case when H, M, T =
6 0.
When M ∈ {−2, −1, 1, 2}, a little casework gives all the possible (H, M, T ) = (2, 1, 4), (4, 1, 2), (−1, −2, 1), (1, −2, −1).

• If M = −2, i.e. H − 4 + T = 4HT , then −15 = (4H − 1)(4T − 1), so 4H − 1 ∈ {±1, ±3, ±5, ±15}
(only −1, +3, −5, +15 are possible) corresponding to 4T −1 ∈ {∓15, ∓5, ∓3, ∓1} (only +15, −5, +3, −1
are possible). But H, T are nonzero, we can only have 4H − 1 ∈ {+3, −5}, yielding (−1, −2, 1)
and (1, −2, −1).
• If M = +2, i.e. H + 4 + T = 4HT , then 17 = (4H − 1)(4T − 1), so 4H − 1 ∈ {±1, ±17} (only
−1, −17 are possible) corresponding to 4T − 1 ∈ {±17, ±1} (only −17, −1 are possible). But H, T
are nonzero, so there are no possibilities here.
• If M = −1, i.e. H − 2 + T = HT , then −1 = (H − 1)(T − 1), so we have H − 1 ∈ {±1} and
T − 1 ∈ {∓1}, neither of which is possible (as H, T 6= 0).
• If M = +1, i.e. H + 2 + T = HT , then 3 = (H − 1)(T − 1), so we have H − 1 ∈ {±1, ±3}. Since
H, T 6= 0, H − 1 ∈ {+1, +3}, yielding (2, 1, 4) and (4, 1, 2).

Now suppose there is such a triple (H, M, T ) for |M | ≥ 3. The equation in the problem gives (M 2 H −
1)(M 2 T − 1) = 2M 3 + 1. Note that since H, T 6= 0, |2M 3 + 1| = |M 2 H − 1| · |M 2 T − 1| ≥ min(M 2 −
1, M 2 + 1)2 = M 4 − 2M 2 + 1 > 2|M |3 + 1 gives a contradiction.
16. [8] Determine the number of unordered triples of distinct points in the 4 × 4 × 4 lattice grid
{0, 1, 2, 3}3 that are collinear in R3 (i.e. there exists a line passing through the three points).
Answer: 376 Define a main plane to be one of the xy, yz, zx planes. Define a space diagonal to
be a set of collinear points not parallel to a main plane. We classify the lines as follows:

(a) Lines parallel to two axes (i.e. orthogonal to a main plane). Notice that given a plane
of the form v = k, where v ∈ {x, y, z}, k ∈ {0, 1, 2, 3}, there are 8 such lines, four in one direction
and four in a perpendicular direction. There are 4 × 3 = 12 such planes. However, each line lies
in two of these (v, k) planes, so there are 8×4×3
2 = 48 such lines. Each of these lines has 4 points,
so there are 4 possible ways to choose 3 collinear points, giving 4 × 48 = 192 triplets.

Guts
 (b) Diagonal lines containing four points parallel to some main plane. Consider a plane of
the form (v, k), as defined above. These each have 2 diagonals that contain 4 collinear points.
Each of these diagonals uniquely determines v, k so these diagonals are each counted once. There
are 12 possible (v, k) pairs, yielding 12 × 2 × 4 = 96 triplets.
(c) Diagonal lines containing three points parallel to some main plane. Again, consider a
plane (v, k). By inspection, there are four such lines and one way to choose the triplet of points
for each of these lines. This yields 4 × 12 = 48 triplets.
(d) Main diagonals. There are four main diagonals, each with 4 collinear points, yielding 4 × 4 = 16
triplets.
(e) Space diagonals containing three points. Choose one of the points in the set {1, 2}3 to be
the midpoint of the line. Since these 8 possibilities are symmetric, say we take the point (1, 1, 1).
There are four space diagonals passing through this point, but one is a main diagonal. So each of
the 8 points has 3 such diagonals with 3 points each, yielding 8 × 3 = 24 ways.

Adding all these yields 192 + 96 + 48 + 16 + 24 = 376.

17. [11] Find the least positive integer N > 1 satisfying the following two properties:

• There exists a positive integer a such that N = a(2a − 1).

• The sum 1 + 2 + · · · + (N − 1) is divisible by k for every integer 1 ≤ k ≤ 10.

Answer: 2016 The second condition implies that 16 divides a(2a − 1)(2a2 − a − 1), which shows
that a ≡ 0 or 1 modulo 16. The case a = 1 would contradict the triviality-avoiding condition N > 1.
a cannot be 16, because 7 does not divide a(2a − 1)(2a2 − a − 1). a cannot be 17, because 9 does not
divide a(2a − 1)(2a2 − a − 1). It can be directly verified that a = 32 is the smallest positive integer
for which 1 + 2 + · · · + (N − 1) = 24 · 32 · 5 · 7 · 13 · 31 which is divisible by 1, 2, . . . , 10. For this a, we
compute N = 32(2 · 32 − 1) = 2016.

18. [11] Let f : Z → Z be a function such that for any integers x, y, we have

f (x2 − 3y 2 ) + f (x2 + y 2 ) = 2(x + y)f (x − y).

Suppose that f (n) > 0 for all n > 0 and that f (2015) · f (2016) is a perfect square. Find the minimum
possible value of f (1) + f (2).
Answer: 246 Plugging in −y in place of y in the equation and comparing the result with the
original equation gives
(x − y)f (x + y) = (x + y)f (x − y)
This shows that whenever a, b ∈ Z − {0} with a ≡ b(mod 2), we have

f (a) f (b)
=
a b
which implies that there are constants α = f (1) ∈ Z>0 , β = f (2) ∈ Z>0 for which f satisfies the
equation (∗): (
n · α when 2 ∤ n
f (n) = n
2 ·β when 2|n

Therefore, f (2015)f (2016) = 2015α · 1008β = 24 · 32 · 5 · 7 · 13 · 31αβ, so αβ = 5 · 7 · 13 · 31 · t2 for

some t ∈ Z>0 . We claim that (α, β, t) = (5 · 31, 7 · 13, 1) is a triple which gives the minimum α + β. In
particular, we claim α + β ≥ 246.
√ √
Consider the case t ≥ 2 first. We have, by AM-GM, α + β ≥ 2 · αβ ≥ 4 · 14105 > 246. Suppose
t = 1. We have α · β = 5 · 7 · 13 · 31. Because (α + β)2 − (α − β)2 = 4αβ is fixed, we want to have α
as close as β as possible. This happens when one of α, β is 5 · 31 and the other is 7 · 13. In this case,
α + β = 91 + 155 = 246.

Guts
 Finally, we note that the equality f (1) + f (2) = 246 can be attained. Consider f : Z → Z such that
f (n) = 91n for every odd n ∈ Z and f (n) = 155 2 n for every even n ∈ Z. It can be verified that f
satisfies the condition in the problem and f (1) + f (2) = 246 as claimed.

19. [11] Find the smallest positive integer n such that the polynomial (x + 1)n − 1 is “divisible by x2 + 1
modulo 3”, or more precisely, either of the following equivalent conditions holds:

• there exist polynomials P, Q with integer coefficients such that (x+1)n −1 = (x2 +1)P (x)+3Q(x);
• or more conceptually, the remainder when (the polynomial) (x + 1)n − 1 is divided by (the
polynomial) x2 + 1 is a polynomial with (integer) coefficients all divisible by 3.

Answer: 8 Solution 1. We have (x + 1)2 = x2 + 2x + 1 ≡ 2x, (x + 1)4 ≡ (2x)2 ≡ −4 ≡ −1, and

(x + 1)8 ≡ (−1)2 = 1. So the order n divides 8, as x + 1 and x2 + 1 are relatively prime polynomials
modulo 3 (or more conceptually, in F3 [x]), but cannot be smaller by our computations of the 2nd and
4th powers.
Remark. This problem illustrates the analogy between integers and polynomials (specifically here,
polynomials over the finite field of integers modulo 3), with x2 + 1 (mod 3) taking the role of a
deg(x2 +1)
prime number. Indeed, we can prove analogously to Fermat’s little theorem that f (x)3 ≡ f (x)
(mod x2 + 1, 3) for any polynomial f (x): try to work out the proof yourself! (This kind of problem,
with 3 replaced by any prime p and x2 + 1 by any irreducible polynomial modulo p, is closely related
to the theory of (extensions of ) finite fields.)
Solution 2. Here’s a solution avoiding the terminology of groups and fields. Let R(x) = (x + 1)n − 1 −
P (x)(x2 + 1) = 3Q(x) be the remainder (here P is the quotient), whose coefficients (upon expansion)
must all be divisible by 3. Now consider R(i). The coefficients of the even and odd powers of x are
divisible by 3, so R(i) must have real and imaginary parts divisible by 3. Notice that (1 + i)2 = 2i,
(1 + i)4 = −4, (1 + i)6 = −8i, (1 + i)8 = 16, and that (1 + i)8 − 1 = 15, which has real and imaginary
parts divisible by 3. Since any even power of (1 + i) (for n ≤ 6) yields a purely real or purely imaginary
number with a coefficient not divisible by 3, multiplying it by 1 + i will yield an imaginary part not
divisible by 3. To see that n = 8 works, notice that, when taken modulo 3,

(x + 1)8 − 1 = x8 + 8x7 + 28x6 + 56x5 + 70x4 + 56x3 + 28x2 + 8x

≡ x8 − x7 + x6 − x5 + x4 − x3 + x2 − x
= (x2 + 1)(x6 − x5 + x2 − x) (mod 3),

as desired.
20. [11] What is the largest real number θ less than π (i.e. θ < π) such that
10
Y
cos(2k θ) 6= 0
k=0

and
10  
Y 1
1+ = 1?
cos(2k θ)
k=0

2046π
Answer: 2047 For equality to hold, note that θ cannot be an integer multiple of π (or else sin = 0
and cos = ±1).
Let z = eiθ/2 6= ±1. Then in terms of complex numbers, we want
10 10 k k
Y 2 Y (z 2 + z −2 )2
(1 + )= ,
k=0
z 2k+1 + z −2k+1 k=0
z 2k+1 + z −2k+1

Guts
 which partially telescopes to
10
z + z −1 Y 2k k

2 11 −2 11 (z + z −2 ).
z +z
k=0

Using a classical telescoping argument (or looking at binary representation; if you wish we may note
that z − z −1 6= 0, so the ultimate telescoping identity holds2 ), this simplifies to
11 11
z + z −1 z 2 − z −2 tan(210 θ)
= .
z 2 + z −211 z − z −1
11
tan(θ/2)

Since tan x is injective modulo π (i.e. π-periodic and injective on any given period), θ works if and
only if θ2 + ℓπ = 1024θ for some integer ℓ, so θ = 2047
2ℓπ
. The largest value for ℓ such that θ < π is at
ℓ = 1023, which gives θ = 2046π
2047 . 3

Remark. It’s also possible to do this without complex numbers, but it’s less systematic. The steps
2 k−1
2k θ
are the same, though, first note that 1 + sec 2k θ = 1+cos
cos 2k θ
= 2 cos 2
cos 2k θ
θ
using the identity cos 2x =
2
2 cos x − 1 (what does this correspond to in complex numbers?). hen we telescope using the identity
2 cos x = sin 2x
sin x (again, what does this correspond to in complex numbers?).

21. [14] Define a sequence ai,j of integers such that a1,n = nn for n ≥ 1 and ai,j = ai−1,j + ai−1,j+1 for
all i, j ≥ 1. Find the last (decimal) digit of a128,1 .
Answer: 4 By applying the recursion multiple times, we find that a1,1 = 1, a2,n = nn + (n + 1)n+1 ,
and a3,n = nn + 2(n + 1)n+1 + (n + 2)n+2 . At this point, we can conjecture and prove by induction
that
m−1
X m − 1 X m − 1
n+k
am,n = (n + k) = (n + k)n+k .
k k
k=0 k≥0

(The second expression is convenient for dealing with boundary cases. The induction relies on m


0 =
m−1 m m−1 m−1





0
on the k = 0 boundary, as well as k =
k + k−1 for k ≥ 1.) We fix m = 128. Note that
127 127
k ≡ 1 (mod 2) for all 1 ≤ k ≤ 127 and k ≡ 0 (mod 5) for 3 ≤ k ≤ 124, by Lucas’ theorem on
binomial coefficients. Therefore, we find that
127   127
X 127 X
a128,1 = (k + 1)k+1 ≡ (k + 1)k+1 ≡ 0 (mod 2)
k
k=0 k=0

and  
X 127
a128,1 ≡ (k + 1)k+1 ≡ 4 (mod 5).
k
k∈[0,2]∪[125,127]

Therefore, a128,1 ≡ 4 (mod 10).

22. [14] Let A1 ,A2 , . . .,A2015 be distinct points on the unit circle with center O. For every two distinct
integers i, j, let Pij be the midpoint of Ai and Aj . Find the smallest possible value of
X
OPij2 .
1≤i<j≤2015

2015·2013
OR 4056195 |ai + aj |2 /4 = (2 + 2ai · aj )/4 = 12 2015
P P

Answer: 4 4 Use vectors. 2 +
1 2014 2015 2015·2013
P 2 P 2
P
4 (| ai | − |ai | ) ≥ 2015 · 4 − 4 = 4 , with equality if and only if ai = 0, which
occurs for instance for a regular 2015-gon.
k k
2 The identity still holds even if z 2 − z −2 = 0 for some k ≥ 1 used in the telescoping argument: why?
Q
3 This indeed works, since 10 k
k=0 cos(2 θ) 6= 0: why?

Guts
23. [14] Let S = {1, 2, 4, 8, 16, 32, 64, 128, 256}. A subset P of S is called squarely if it is nonempty and
the sum of its elements is a perfect square. A squarely set Q is called super squarely if it is not a proper
subset of any squarely set. Find the number of super squarely sets.
(A set A is said to be a proper subset of a set B if A is a subset of B and A 6= B.)
Answer: 5 Clearly we may biject squarely sets with binary representations of perfect squares
between 1 and 20 + · · · + 28 = 29 − 1 = 511, so there are 22 squarely sets, corresponding to n2 for
n = 1, 2, . . . , 22. For convenience, we say N is (super) squarely if and only if the set corresponding to
N is (super) squarely.
The general strategy is to rule out lots of squares at time, by searching for squares with few missing
digits (and ideally most 1’s consecutive, for simplicity). We can restrict ourselves (for now) to odds;
(2k)2 is just k 2 with two additional zeros at the end. 1, 9, 25, 49, 81 are ineffective, but 121 = 27 − 7 =
26 + 25 + 24 + 23 + 20 immediately rules out all odd squares up to 92 , as they must be 1 (mod 8).
Fortunately, 222 = 4 · 112 is in our range (i.e. less than 512), ruling out all even squares up to 202 as
well.
This leaves us with 112 , 132 , 152 , 172 , 192 , 212 , 222 , with binary representations 001111001, 010101001,
011100001, 100100001, 101101001 (kills 172 ), 110111001 (kills 132 ), 111100100 (kills nothing by parity).
Thus 112 , 152 , 192 , 212 , 222 are the only super squarely numbers, for a total of 5.

24. [14] ABCD is a cyclic quadrilateral with sides AB = 10, BC = 8, CD = 25, and DA = 12. A circle
ω is tangent to segments DA, AB, and BC. Find the radius of ω.
q √
1209 8463
Answer: 7 OR 7 Denote E an intersection point of AD and BC. Let x = EA and
y = EB. Because ABCD is a cyclic quadrilateral, △EAB is similar to △ECD. Therefore, y+8 25
x = 10
x+12 25 128 152
and y = 10 . We get x = 21 and y = 21 . Note that ω is the E-excircle of △EAB, so we may
finish by standard calculations.
Indeed, first we compute the semiperimeter s = EA+AB+BE
2 = x+y+10
2 = 35 3 . Now the radius of ω is
(by Heron’s formula for area)
r r √
[EAB] s(s − x)(s − y) 1209 8463
rE = = = = .
s − AB s − 10 7 7

25. [17] Let r1 , . . . , rn be the distinct real zeroes of the equation

x8 − 14x4 − 8x3 − x2 + 1 = 0.

Evaluate r12 + · · · + rn2 .

Answer: 8 Observe that

x8 − 14x4 − 8x3 − x2 + 1 = (x8 + 2x4 + 1) − (16x4 + 8x3 + x2 )

= (x4 + 4x2 + x + 1)(x4 − 4x2 − x + 1).

The polynomial x4 + 4x2 + x + 1 = x4 + 15 2 x 2

4 x + ( 2 + 1) has no real roots. On the other hand,
4 2
let P (x) = x − 4x − x + 1. Observe that P (−∞) = +∞ > 0, P (−1) = −1 < 0, P (0) = 1 > 0,
P (1) = −3 < 0, P (+∞) = +∞ > 0, so by the intermediate value theorem, P (x) = 0 has four distinct
real roots, which are precisely the real roots of the original degree 8 equation. By Vieta’s formula on
P (x),
X
r12 + r22 + r32 + r42 = (r1 + r2 + r3 + r4 )2 − 2 · ( ri rj )
i<j
2
= 0 − 2(−4) = 8.

Guts
 √ √ √
26. [17] Let a = 17 and b = i 19, where i = −1. Find the maximum possible value of the ratio
|a − z|/|b − z| over all complex numbers z of magnitude 1 (i.e. over the unit circle |z| = 1).
Answer: 43 Let |a − z|/|b − z| = k. We wish to determine the minimum and maximum value of k.
Squaring and expansion give:

|a − z|2 = |b − z|2 · k 2
|a|2 − 2a · z + 1 = (|b|2 − 2b · z + 1)k 2
|a|2 + 1 − (|b|2 + 1)k 2 = 2(a − bk 2 ) · z,

where · is a dot product of complex numbers, i.e., the dot product of vectors corresponding to the
complex numbers in the complex plane. Now, since z has modulus 1 but can assume any direction,
the only constraint on the value of k is

||a|2 + 1 − (|b|2 + 1)k 2 | ≤ 2|a − bk 2 |.

Squaring again and completing the square, the inequality reduces to:

(|a|2 − 1)2 + (|b|2 − 1)2 k 4 + 2(4a · b − (|a|2 + 1)(|b|2 + 1))k 2 ≤ 0

((|a|2 − 1) − (|b|2 − 1)k 2 )2 − 4|a − b|2 k 2 ≤ 0
|(|a|2 − 1) − (|b|2 − 1)k 2 | ≤ 2|a − b|k.

At this stage all the relevant expressions are constant real numbers. Denote, for simplicity, A =
|a|2 − 1, B = |b|2 − 1, and C = |a − b|. Then, we are looking for k such that |A − Bk 2 | ≤ 2Ck. If B = 0,
A A
then k ≥ | 2C |, so the minimum value is | 2C | and the maximum value is +∞. Otherwise, consider

C 2 + AB = (|a|2 − 2a · b + |b|2 ) + (|a|2 − 1)(|b|2 − 1)

= |ab|2 − 2a · b + 1
= |ab|2 − 2ℜ(ab) + 1
= |ab − 1|2
√
So let D = |ab−1| = C 2 + AB. We may assume B > 0 (the another case is analogous: just substitute
A, B with −A, −B). Then, k is determined by the following inequalities:

Bk 2 + 2Ck − A ≥ 0
Bk 2 − 2Ck − A ≤ 0

The first inequality gives k ≤ −C−D

B or k ≥ −C+D
B , and the second gives C−D
B ≤k≤ C+D
B . Combining,
this gives | C−D
B | ≤ k ≤ | C+D
B |, as claimed.
To summarize the general answer, let A = |a|2 − 1, B = |b|2 − 1, C = |a − b|, D = |ab − 1|. Then, if
A
|b| = 1, min is | 2C | and max is +∞; otherwise, min is | C−D C+D
B | and max is | B |.
√ √ √ √ √
In the special case a = 17 and b = 19i, we have A = 16, B = 18, C = 17 − 19i = 36 = 6,
√
and D = 17 · 19 + 1 = 18. Thus the answer is C+D B = 6+18 4
18 = 3 .

27. [17] Let a, b be integers chosen independently and uniformly at random from the

set {0, 1, 2, . . . , 80}.
Compute the expected value of the remainder when the binomial coefficient ab = b!(a−b)!
a!
is divided
0 a



by 3. (Here 0 = 1 and b = 0 whenever a < b.)
1816
Answer: 6561 By Lucas’ Theorem we’re looking at

4  
Y ai
i=1
bi

Guts
 where the ai and bi are the digits of a and b in base 3. If any ai < bi , then the product is zero modulo
3.
Otherwise, the potential residues are 20 = 1, 21 = 2, 22 = 1, 10 = 1, 11 = 1, 00 = 1.

So each term in the product has a 13 chance of being zero; given that everything is nonzero, each term
has a 16 chance of being 2 and a 65 chance of being 1. The probability that an even number of terms
are 1 given that none are zero is then given by the roots of unity filter

( 56 + 1
6 · (1))4 + ( 56 + 1
6 · (−1))4 81 + 16 97
= = .
2 162 162

Thus the expected value is

 4  
2 97 1816
2− = .
3 162 6561

28. [17] Let w, x, y, and z be positive real numbers such that

0 6= cos w cos x cos y cos z

2π = w + x + y + z
3 tan w = k(1 + sec w)
4 tan x = k(1 + sec x)
5 tan y = k(1 + sec y)
6 tan z = k(1 + sec z).
1
(Here sec t denotes cos t when cos t 6= 0.) Find k.
√
Answer: 19 From the identity tan u2 = 1+cos
sin u w x
u , the conditions work out to 3 tan 2 = 4 tan 2 =
y y
5 tan 2 = 6 tan z2 = k. Let a = tan w2 , b = tan x2 , c = tan 2 , and d = tan z2 . Using the identity
tan M +tan N
tan(M + N ) = 1−tan M tan N , we obtain

w+x
+ tan y+z



tan
 
w+x y+z 2 2
tan + =
− tan w+x tan y+z



2 2 1 2 2
a+b c+d
1−ab + 1−cd
=   
a+b c+d
1− 1−ab 1−cd

a + b + c + d − abc − abd − bcd − acd

= .
1 + abcd − ab − ac − ad − bc − bd − cd

Because x + y + z + w = π, we get that tan x+y+z+w

2 = 0 and thus a+b+c+d = abc+abd+bcd+acd.
Substituting
√ a, b, c,
√ d corresponding to the variable k, we obtain that k 3 − 19k = 0. Therefore,√k can
be only 0, 19, − 19. However, k = 0 is impossible√ as w, x, y, z will all be 0. Also, k = − 19 is
impossible as w, x, y, z will exceed π. Therefore, k = 19.

29. [20] Let ABC be a triangle whose incircle has center I and is tangent to BC, CA, AB, at D, E, F .
\ of the circumcircle of ABC. Suppose P is a point on
Denote by X the midpoint of major arc BAC
line XI such that DP ⊥ EF .
Given that AB = 14, AC = 15, and BC = 13, compute DP .
√
Answer: 4 5 5 Let H be the orthocenter of triangle DEF . We claim that P is the midpoint of DH.
Indeed, consider an inversion at the incicrle of ABC, denoting the inverse of a point with an asterik.
It maps ABC to the nine-point circle of △DEF . According to ∠IAX = 90◦ , we have ∠A∗ X ∗ I = 90◦ .
Hence line XI passes through the point diametrically opposite to A∗ , which is the midpoint of DH,
as claimed.

Guts
 The rest is a straightforward computation. The inradius of △ABC is r = 4. The length of EF is given
by EF = 2 AF ·r
AI =
√16
5
. Then,
 2
1 1 64 16
DP 2 = 4r2 − EF 2 = 42 −


DH = = .
2 4 5 5
√
4 5
Hence DP = 5 .
Remark. This is also not too bad of a coordinate bash.
30. [20] Find the sum of squares of all distinct complex numbers x satisfying the equation

0 = 4x10 − 7x9 + 5x8 − 8x7 + 12x6 − 12x5 + 12x4 − 8x3 + 5x2 − 7x + 4.

7
Answer: − 16 For convenience denote the polynomial by P (x). Notice 4 + 8 = 7 + 5 = 12 and
that the consecutive terms 12x6 − 12x5 + 12x4 are the leading terms of 12Φ14 (x), which is suggestive.
Indeed, consider ω a primitive 14-th root of unity; since ω 7 = −1, we have 4ω 10 = −4ω 3 , −7ω 9 = 7ω 2 ,
and so on, so that
P (ω) = 12(ω 6 − ω 5 + · · · + 1) = 12Φ14 (ω) = 0.
Dividing, we find
P (x) = Φ14 (x)(4x4 − 3x3 − 2x2 − 3x + 4).
This second polynomial is symmetric; since 0 is clearly not a root, we have
1 2 1
4x4 − 3x3 − 2x2 − 3x + 4 = 0 ⇐⇒ 4(x + ) − 3(x + ) − 10 = 0.
x x
Setting y = x + 1/x and solving the quadratic gives y = 2 and y = −5/4 as solutions; replacing
y with √x + 1/x and solving the two resulting quadratics give the double root x = 1 and the roots
(−5 ± i 39)/8 respectively. Together with the primitive fourteenth roots of unity, these are all the
roots of our polynomial.
Explicitly, the roots are
√
eπi/7 , e3πi/7 , e5πi/7 , e9πi/7 , e11πi/7 , e13πi/7 , 1, (−5 ± i 39)/8.

The sum of squares of the roots of unity (including 1) is just 0 by symmetry (or a number of other
2
methods). The sum of the squares of the final conjugate pair is 2(5 8−39)
2 = − 14 7
32 = − 16 .

31. [20] Define a power cycle to be a set S consisting of the nonnegative integer powers of an integer a,
i.e. S = {1, a, a2 , . . . } for some integer a. What is the minimum number of power cycles required such
that given any odd integer n, there exists some integer k in one of the power cycles such that n ≡ k
(mod 1024)?
Answer: 10 Solution 1. Partition the odd residues mod 1024 into 10 classes:

• Class 1: 1 (mod 4).

• Class n (2 ≤ n ≤ 9): 2n − 1 (mod 2n+1 ).
• Class 10: −1 (mod 1024).

Let Sa be the power cycle generated by a. If a is in class 1, all of Sa is in class 1. If a is in class n

(2 ≤ n ≤ 9), then Sa is in the union of class n and the residues 1 (mod 2n+1 ). If a is in class 10, then
Sa is in the union of class n and the residues 1 (mod 1024). Therefore, Sa cannot contain two of the
following residues: 5, 22 − 1, 23 − 1, . . . 210 − 1, and that at least 10 cycles are needed.
Note that 5128 −1 = (5−1)(5+1)(52 +1) · · · (564 +1) has exactly 9 factors of 2 in its prime factorization,
while 5256 − 1 = (5128 − 1)(5128 + 1) is divisible by 1024 so the order of 5 modulo 1024, the smallest

Guts
 positive power of 5 that is congruent to 1, is 256. Observe that among 50 , 51 , . . . 5255 , the ratio between
any two is a positive power of 5 smaller than 5256 , so the ratio is not congruent to 1 and any two terms
are not congruent mod 1024. In addition, all terms are in class 1, and class 1 has 256 members, so S5
contains members congruent to each element of class 1.
Similarly, let 2 ≤ n ≤ 9. Then the order of a, where a = 2n − 1, is 210−n . The 29−n terms
10−n
a1 , a3 , . . . a2 −1
are pairwise not congruent and all in class n. Class n only has 29−n members,
so Sa contains members congruent to each element of class n.
Finally, S−1 contains members congruent to the element of class 10.
The cycles S5 , S−1 , and 8 cycles Sa cover all the residues mod 1024, so the answer is 10.
Solution 2. Lemma. Given a positive integer n ≥ 3, there exists an odd integer x such that the
order of x modulo 2n is 2n−2 .
Proof. We apply induction on n. The base cases of n = 3, 4 are clearly true with x = 3, so suppose we
have the statement holds for n − 1 and we wish to show it for n where n ≥ 5. Suppose no such integer
n−3 n−4 n−4
x exists, so we have x2 ≡ 1 (mod 2n ) for all odd x. But then remark that (x2 − 1)(x2 + 1) ≡ 0
n−4
n 2
(mod 2 ). As n − 4 ≥ 1 we have x + 1 ≡ 2 (mod 4) as all squares are 1 (mod 4), so it follows for
n−4
the above relation to be true we require 2n−1 divides x2 − 1 for all odd x. However, by taking x to
have order 2n−3 modulo 2n−1 (which exists by the inductive hypothesis) we get a contradiction so we
are done. 
Now, let x have order 28 modulo 210 . Remark that if for some integer k we had xk ≡ −1 (mod 210 ), we
would have x2k ≡ 1 (mod 210 ) so 27 | k. In that case k is even so as x is odd we have xk ≡ 1 (mod 4)
but −1 + 1024m is never 1 (mod 4) for any integer m so it follows −1 is not equal to xk modulo 1024
for any integer k. Thus it follows that when we let S to be the set of powers of x, then no two elements
in S and −S are congruent modulo 1024. As the order of x is 256 and there are 512 possible odd
remainders upon dividing by 1024, it immediately follows that every integer x is congruent modulo
1024 to ±xk for some 1 ≤ k ≤ 256 and some choice of sign.
7 8
Now, it is easy so see that x, −x, −x2 , −x4 , . . . , −x2 , −x2 generating 10 power cycles works (xa for
k k
any a is in the first, and then −x2 a for odd a is in the power cycle generated by −x2 ). To show
that we cannot do any better suppose there exist 9 power cycles which include all odd integers modulo
k
1024. Then remark that −x2 is congruent to a number in a power cycle modulo 1024 if and only if
k
the power cycle is generated by an integer congruent to x−2 ·a (mod 1024) where a is an odd integer.
It follows that 9 integers must be congruent to −xa1 , −x2a2 , . . . , −x256a9 (mod 1024) for some odd
integers a1 , a2 , . . . , a9 . But then 3 is clearly not in any of the power cycles, contradiction so it follows
we must have at least 10 power cycles so we are done.
32. [20] A wealthy king has his blacksmith fashion him a large cup, whose inside is a cone of height 9
inches and base diameter 6 inches (that is, the opening at the top of the cup is 6 inches in diameter).
At one of his many feasts, he orders the mug to be filled to the brim with cranberry juice.
For each positive integer n, the king stirs his drink vigorously and takes a sip such that the height
of fluid left in his cup after the sip goes down by n12 inches. Shortly afterwards, while the king is
distracted, the court jester adds pure Soylent to the cup until it’s once again full. The king takes sips
precisely every minute, and his first sip is exactly one minute after the feast begins.
As time progresses, the amount of juice consumed by the king (in cubic inches) approaches a number
r. Find r.
3
√
Answer: 216π 8π −2187 3
2 First, we find the total amount of juice consumed. We can simply subtract
the amount of juice remaining at infinity from the initial amount of juice in the cup, which of course
is simply the volume of the cup; we’ll denote this value by V .
Since volume in the cup varies as the cube of height, the amount of juice remaining in the cup after m
minutes is
m  3 m  ! 3
Y 9 − n12 Y 1
V · =V · 1− 2 .
n=1
9 n=1
9n

Guts
 We can now factor the term inside the product to find
m
!3  3
Y (3n + 1)(3n − 1) (3m + 1)!
V =V .
n=1
9n2 33m (m!)3

If remains to evaluate the limit of this expression as m goes to infinity.

However, by Stirling’s approximation, we have

( 3n+1 3n+1
p
(3m + 1)! e ) · 2π(3n + 1)
lim = lim
m→∞ 33m (m!)3 ( 3n
p
m→∞ 3n (2πn)3
e )
√  3n
(3n + 1) 3 3n + 1
= lim
m→∞ 2πne 3n
√
3 3
= .
2π
Therefore the total amount of juice the king consumes is
√ !3  2 √ ! √
8π 3 − 81 3 216π 3 − 2187 3

3 3 3 ·π·9
V −V = = .
2π 3 8π 3 8π 2

(3m+1)!
Remark. We present another way to calculate the limit at m → ∞ of f (m) = 33m (m!)3 . We have

(3m + 4)! (m + 32 )(m + 34 )

f (m + 1) = = f (m) ,
33m+3 (m
+ 1)! 3 (m + 1)2
whence we can write
cΓ(m + 23 )Γ(m + 34 )
f (m) =
Γ(m + 1)2
for some constant c. We can find c by equating the expressions at m = 0; we have
cΓ( 32 )Γ( 43 )
1 = f (0) = ,
Γ(1)2

so that c = Γ(1)2 /Γ( 23 )Γ( 43 ).

Of course, Γ(1) = 0! = 1. We can evaluate the other product as follows:
       
2 4 1 2 1 1 π 2π
Γ Γ = Γ Γ = · = √ .
3 3 3 3 3 3 sin π/3 3 3
Here the first step follows
√ from Γ(n + 1) = nΓ(n), while the second follows from Euler’s reflection
formula. Thus c = 3 3/2π. We can now compute
√
cΓ(m + 32 )Γ(m + 43 ) 3 3 Γ(m + 32 )Γ(m + 43 )
lim f (m) = lim 2
= lim .
m→∞ m→∞ Γ(m + 1) 2π m→∞ Γ(m + 1)2
√
Since limn→∞ Γ(n + α)/[Γ(n)nα ] = 1, this final limit is 1 and f (m) → 3 3/2π as m → ∞.
33. [25] Let N denote the sum of the decimal digits of 1000


100 . Estimate the value of N . If your answer is
a positive integer A written fully in decimal notation (for example, 521495223), your score will be the
|A−N |
greatest integer not exceeding 25 · (0.99) . Otherwise, your score will be zero.
Answer: 621 http://www.wolframalpha.com/input/?i=sum+of+digits+of+nCr(1000,100)
To see this, one can estimate there are about 150 digits, and we expect the digits to be roughly random,
for 150 · 4.5 ≈ 675, which is already very close to the actual answer. The actual number of digits is
140, and here 140 · 4.5 = 630 is within 9 of the actual answer.

Guts
34. [25] For an integer n, let f (n) denote the number of pairs (x, y) of integers such that x2 + xy + y 2 = n.
Compute the sum
106
X
nf (n).
n=1
b
Write your answer in the form a · 10 , where b is an integer and 1 ≤ a < 10 is a decimal number.
If your answer is written in this form, your score will be max{0, 25 − ⌊100|log10 (A/N )|⌋)}, where
N = a · 10b is your answer to this problem and A is the actual answer. Otherwise, your score will be
zero.
Answer: 1.813759629294 · 1012 Rewrite the sum as
X
(x2 + xy + y 2 ),
x2 +xy+y 2 ≤106

where the sum is over all pairs (x, y) of integers with x2 + xy + y 2 ≤ 106 . We can find a crude upper
bound for this sum by noting that
3 x 2 3
x2 + xy + y 2 = x2 + + y ≥ x2 ,
4 2 4
so each term of this sum has |x| ≤ √2 103 . Similarly, |y| ≤ √2 103 . Therefore, the number of terms in
3 3
the sum is at most  2
4
√ 103 + 1 ≈ 106 .
3
(We are throwing away “small” factors like 16 3 in the approximation.) Furthermore, each term in the
sum is at most 106 , so the total sum is less than about 1012 . The answer 1 · 1012 would unfortunately
still get a score of 0.
For a better answer, we can approximate the sum by an integral:
X ZZ
2 2
(x + xy + y ) ≈ (x2 + xy + y 2 ) dy dx.
x2 +xy+y 2 ≤106 x2 +xy+y 2 ≤106

√ 
Performing the change of variables (u, v) = 23 x, 21 x + y and then switching to polar coordinates
√
(r, θ) = ( u2 + v 2 , tan−1 (v/u)) yields
2
ZZ ZZ
(x2 + xy + y 2 ) dy dx = √ (u2 + v 2 ) dv du
2 2
x +xy+y ≤10 6 3 2 2
u +v ≤10 6

Z 2π Z 103
2
=√ r3 dr dθ
3 0 0
Z 3
4π 10 3
=√ r dr
3 0
π
= √ · 1012 .
3
This is approximately 1.8138 · 1012 , which is much closer to the actual answer. (An answer of 1.8 · 1012
is good enough for full credit.)
The answer can also be computed exactly by the Common Lisp code:

(defconstant +MAX+ 1e6)

(defvar +lower+ -2000)
(defvar +upper+ 2000)
(princ

Guts
 (loop for x from +lower+ to +upper+ sum
(loop for y from +lower+ to +upper+
sum
(let ((S (+ (* x x) (* x y) (* y y))))
(if (and (<= S +MAX+) (> S 0)) S 0)))))

35. [25] Let P denote the set of all subsets of {1, . . . , 23}. A subset S ⊆ P is called good if whenever A, B
are sets in S, the set (A \ B) ∪ (B \ A) is also in S. (Here, A \ B denotes the set of all elements in A
that are not in B, and B \ A denotes the set of all elements in B that are not in A.) What fraction
of the good subsets of P have between 2015 and 3015 elements, inclusive?
If your answer is a decimal number or a fraction (of the form m/n, where m and n are positive integers),
then your score on this problem will be equal to max{0, 25 − ⌊1000|A − N |⌋}, where N is your answer
and A is the actual answer. Otherwise, your score will be zero.
18839183877670041942218307147122500601235
Answer: 47691684840486192422095701784512492731212 ≈ 0.3950203047068107 Let n = 23, and ℓ =
⌊n/2⌋ = 11.
We use the well-known rephrasing of the symmetric difference ((A \ B) ∪ (B \ A)) in
terms of addition
modulo 2 of “indicators/characteristic vectors”. So we simply want the number nℓ 2 of dimension ℓ
subspaces of the F := F2 -vector space V := Fn2 . Indeed, good subsets of 2d elements simply correspond
to dimension d subspaces (in particular, good subsets can only have sizes equal to powers of |F | = 2,
and 2ℓ is the only power between 2015 and 3015, inclusive).
To do this, it’s easier to first count the number of (ordered) tuples of ℓ linearly independent elements of
V , and divide (to get the subspace count) by the number of (ordered) tuples of ℓ linearly independent
elements of any ℓ-dimensional subspace of V (a well-defined number independent of the choice of
subspace).
In general, if we want to count tuples of m linearly independent elements in an n-dimensional space
(with n ≥ m), just note that we are building on top of (0) (the zero-dimensional subspace), and once
we’ve chosen r ≤ m elements (with 0 ≤ r ≤ m − 1), there are 2n − 2r elements linearly independent
to the previous r elements (which span a subspace of dimension r, hence of 2r “bad” elements). Thus
the number of m-dimensional subspaces of an n-dimensional space is

(2n − 20 )(2n − 21 ) · · · (2n − 2m−1 )

 
n
:= m ,
m 2 (2 − 20 )(2m − 21 ) · · · (2m − 2m−1 )

a “Gaussian binomial coefficient.”

We want to estimate
n 23



ℓ 2 1 11 2
Pn n

= P11
.
m=0 m 2
2 m=0 23
m 2

n n



To do this, note that m 2 = n−m 2
, so we may restrict our attention to the lower half. Intuitively, the
Gaussian binomial coefficients should decay exponentially (or similarly quickly) away from the center;
indeed, if m ≤ n/2, then

(2m − 20 ) · 2m−1
   
n n
/ = ≈ 22m−1−n .
m−1 2 m 2 (2n − 2m−1 )

So in fact, the decay is super-exponential, starting (for n = 23 odd and m ≤ ℓ = 11) at an ≈ 41 rate.
So most of the terms (past the first 2 to 4, say) are negligible in our estimation. If we use the first two
terms, we get an approximation of 12 · 1+1 1 = 52 = 0.4, which is enough for 20 points. (Including the
4
32
next term gives an approximation of 81 ≈ 0.3950617, which is good enough to get full credit.)
To compute the exact answer, we used the following python3 code:

Guts
 from functools import lru_cache
from fractions import Fraction

@lru_cache(maxsize=None)
def gauss_binom(n, k, e):
if k < 0 or k > n:
return 0
if k == 0 or k == n:
return 1
return e ** k * gauss_binom(n - 1, k, e) + \
gauss_binom(n - 1, k - 1, e)

N = 23
K = 11
good = gauss_binom(N, K, 2)
total = sum(gauss_binom(N, i, 2) for i in range(N + 1))
print(Fraction(good, total))
print(float(good/total))

36. [25] A prime number p is twin if at least one of p + 2 or p − 2 is prime and sexy if at least one of p + 6
and p − 6 is prime.
How many sexy twin primes (i.e. primes that are both twin and sexy) are there less than 109 ? Express
your answer as a positive integer N in decimal notation; for example, 521495223. If your answer is
1
 
in this form, your score for this problem will be max 0, 25 − ⌊ 10000 |A − N | }, where A is the actual
answer to this problem. Otherwise, your score will be zero.
Answer: 1462105 The Hardy-Littlewood conjecture states that given a set A of integers, the
number of integers x such that x + a is a prime for all a ∈ A is

x Y 1 − w(p;A)
p
(1 + o(1))
(ln x)|A| p (1 − p1 )k

where w(p; A) is the number of distinct residues of A modulo p and the o(1) term goes to 0 as x goes to
infinity. Note that for the 4 tuples of the form (0, ±2, ±6), w(p; A) = 3, and using the approximation
k
1−k/p
≈ 1 − k2 /p2 ≈ (1 − p12 )(2) , we have


(1−1/p)k

 k  (k2)  (k2)  (k2)

6 (2)
Y 1− k 
p 4 9 9
1 k ≈ · ≈
p>3
(1 − p)
π2 3 8 10

Applying this for the four sets A = (0, ±2, ±6), x = 109 (and approximating ln x = 20 and just taking
the p = 2 and p = 3 terms, we get the approximate answer
 3
109 1 − 12 1 − 23 9
4· 3 1 3 1 3 = 1640250
20 ( 2 ) ( 3 ) 10
One improvement we can make is to remove the double-counted tuples, in particular, integers x such
that x, x + 6, x − 6, and one of x ± 2 is prime. Again by the Hardy-Littlewood conjecture, the number
of such x is approximately (using the same approximations)
 6
109 1 − 12 1 − 23 9
2· 4 1 4 1 4 ≈ 90000
20 ( 2 ) ( 3 ) 10

Subtracting gives an estimate of about 1550000. Note that this is still an overestimate, as ln 109 is
k
1−k/p 1 (2)
actually about 20.7 and (1−1/p) k < (1 − p2 ) .
Here is the C++ code that we used to generate the answer:

Guts
#include<iostream>
#include<cstring> // memset
using namespace std;

const int MAXN = 1e9;

bool is_prime[MAXN + 6];

int main(){
// Sieve of Eratosthenes
memset(is_prime, true, sizeof(is_prime));
is_prime[0] = is_prime[1] = false;
for (int i=2; i<MAXN + 6; i++){
if (is_prime[i]){
for (int j=2 * i; j < MAXN + 6; j += i){
is_prime[j] = false;
}
}
}

// Count twin sexy primes.

int ans = 1; // 5 is the only twin sexy prime < 6.
for (int i=6; i<MAXN; i++){
if (is_prime[i]
&& (is_prime[i-6] || is_prime[i+6])
&& (is_prime[i-2] || is_prime[i+2])) {
ans++;
}
}

cout << ans << endl;

return 0;
}

Guts