HMMT February 2016

February 20, 2016

Team
1. [25] Let a and b be integers (not necessarily positive). Prove that a3 + 5b3 6= 2016.
Proposed by: Evan Chen
Since cubes are 0 or ±1 modulo 9, by inspection we see that we must have a3 ≡ b3 ≡ 0 (mod 3) for
this to be possible. Thus a, b are divisible by 3. But then we get 33 | 2016, which is a contradiction.
One can also solve the problem in the same manner by taking modulo 7 exist, since all cubes are 0 or
±1 modulo 7. The proof can be copied literally, noting that 7 | 2016 but 73 - 2016.

2. [25] For positive integers n, let cn be the smallest positive integer for which ncn − 1 is divisible by 210,
if such a positive integer exists, and cn = 0 otherwise. What is c1 + c2 + · · · + c210 ?
Proposed by: Joy Zheng
Answer: 329
In order for cn 6= 0, we must have gcd(n, 210) = 1, so we need only consider such n. The number
ncn − 1 is divisible by 210 iff it is divisible by each of 2, 3, 5, and 7, and we can consider the order
of n modulo each modulus separately; cn will simply be the LCM of these orders. We can ignore the
modulus 2 because order is always 1. For the other moduli, the sets of orders are

a ∈ {1, 2} mod 3
b ∈ {1, 2, 4, 4} mod 5
c ∈ {1, 2, 3, 3, 6, 6} mod 7.

By the Chinese Remainder Theorem, each triplet of choices from these three multisets occurs for
exactly one n in the range {1, 2, . . . , 210}, so the answer we seek is the sum of lcm(a, b, c) over a, b, c
in the Cartesian product of these multisets. For a = 1 this table of LCMs is as follows:

1 2 3 3 6 6
1 1 2 3 3 6 6
2 2 2 6 6 6 6
4 4 4 12 12 12 12
4 4 4 12 12 12 12

which has a sum of 21 + 56 + 28 + 56 = 161. The table for a = 2 is identical except for the top row,
where 1, 3, 3 are replaced by 2, 6, 6, and thus has a total sum of 7 more, or 168. So our answer is
161 + 168 = 329 .
This can also be computed by counting how many times each LCM occurs:

• 12 appears 16 times when b = 4 and c ∈ {3, 6}, for a contribution of 12 × 16 = 192;

• 6 appears 14 times, 8 times when c = 6 and b ≤ 2 and 6 times when c = 3 and (a, b) ∈
{(1, 2), (2, 1), (2, 2)}, for a contribution of 6 × 14 = 84;
• 4 appears 8 times when b = 4 and a, c ∈ {1, 2}, for a contribution of 4 × 8 = 32;
• 3 appears 2 times when c = 3 and a = b = 1, for a contribution of 3 × 2 = 6;
• 2 appears 7 times when a, b, c ∈ {1, 2} and (a, b, c) 6= (1, 1, 1), for a contribution of 2 × 7 = 14;
• 1 appears 1 time when a = b = c = 1, for a contribution of 1 × 1 = 1.

The result is again 192 + 84 + 32 + 6 + 14 + 1 = 329.

3. [30] Let ABC be an acute triangle with incenter I and circumcenter O. Assume that ∠OIA = 90◦ .
Given that AI = 97 and BC = 144, compute the area of 4ABC.
Proposed by: Evan Chen
Answer: 14040
We present five different solutions and outline a sixth and seventh one. In what follows, let a = BC,
b = CA, c = AB as usual, and denote by r and R the inradius and circumradius. Let s = 21 (a + b + c).
In the first five solutions we will only prove that

∠AIO = 90◦ =⇒ b + c = 2a.

Let us see how this solves the problem. This lemma implies that s √
= 216. If we let E be the foot of I
on AB, then AE = s − BC = 72, consequently the inradius is r = 972 − 722 = 65. Finally, the area
is sr = 216 · 65 = 14040 .

First Solution. Since OI ⊥ DA, AI = DI. Now, it is a well-known fact that DI = DB = DC (this is
occasionally called “Fact 5”). Then by Ptolemy’s Theorem,

DB · AC + DC · AB = DA · BC =⇒ AC + AB = 2BC.

Second Solution. As before note that I is the midpoint of AD. Let M and N be the midpoints of AB
and AC, and let the reflection of M across BI be P ; thus BM = BP . Also, M I = P I, but we know
M I = N I as I lies on the circumcircle of triangle AM N . Consequently, we get P I = N I; moreover
by angle chasing we have

∠IN C = ∠AM I = 180◦ − ∠BP I = ∠IP C.

Thus triangles IN C and P IC are congruent (CI is a bisector) so we deduce P C = N C. Thus,

1
BC = BP + P C = BM + CN = (AB + AC).
2

Third Solution. We appeal to Euler’s Theorem, which states that IO2 = R(R − 2r).
Thus by the Pythagorean Theorem on 4AIO (or by Power of a Point) we may write

abc
(s − a)2 + r2 = AI 2 = R2 − IO2 = 2Rr =
2s
with the same notations as before. Thus, we derive that

abc = 2s (s − a)2 + r2

= 2(s − a) (s(s − a) + (s − b)(s − c))

1
= (s − a) (b + c)2 − a2 + a2 − (b − c)2


2
= 2bc(s − a).

From this we deduce that 2a = b + c, and we can proceed as in the previous solution.

Fourth Solution. From Fact 5 again (DB = DI = DC), drop perpendicular from I to AB at E; call
M the midpoint of BC. Then, by AAS congruency on AIE and CDM , we immediately get that
CM = AE. As AE = 21 (AB + AC − BC), this gives the desired conclusion.
 Fifth Solution. This solution avoids angle-chasing and using the fact that BI and CI are angle-
bisectors. Recall the perpendicularity lemma, where

W X ⊥ Y Z ⇐⇒ W Y 2 − W Z 2 = XY 2 − XZ 2 .

Let B 0 be on the extension of ray CA such that AB 0 = AB. Of course, as in the proof of the angle
bisector theorem, BB 0 k AI, meaning that BB 0 ⊥ IO. Let I 0 be the reflection of I across A; of course, I 0
is then the incenter of triangle AB 0 C 0 . Now, we have B 0 I 2 −BI 2 = B 0 O2 −BO2 by the perpendicularity
and by power of a point B 0 O2 − BO2 = B 0 A · B 0 C. Moreover BI 2 + B 0 I 2 = BI 2 + BI 02 = 2BA2 + 2AI 2
by the median formula. Subtracting, we get BI 2 = AI 2 + 21 (AB)(AB − AC). We have a similar
expression for CI, and subtracting the two results in BI 2 − CI 2 = 21 (AB 2 − AC 2 ). Finally,

1
BI 2 − CI 2 = [(BC + AB − AC)2 − (BC − AB + AC)2 ]
4
from which again, the result 2BC = AB + AC follows.

Sixth Solution, outline. Use complex numbers, setting I = ab+bc+ca, A = −a2 , etc. on the unit circle
(scale the picture to fit in a unit circle; we calculate scaling factor later). Set a = 1, and let u = b + c
and v = bc. Write every condition in terms of u and v, and the area in terms of u and v too. There
2
should be two equations relating u and v: 2u + v + 1 = 0 and u2 = 130 97 v from the right angle and the
144 to 97 ratio, respectively. The square area can be computed in terms of u and v, because the area
itself is antisymmetric so squaring it suffices. Use the first condition to homogenize (not coincidentally
the factor (1 − b2 )(1 − c2 ) = (1 + bc)2 − (b + c)2 = (1 + v)2 − u2 from the area homogenizes perfectly. . .
because AB · AC = AI · AIA , where IA is the A-excenter, and of course the way the problem is set up
AIA = 3AI.), and then we find the area of the scaled down version. To find the scaling factor simply
determine |b − c| by squaring it, writing in terms again of u and v, and comparing this to the value of
144.

Seventh Solution, outline. Trigonometric solutions are also possible. One can you write everything in
terms of the angles and solve the equations; for instance, the ∠AIO = 90◦ condition can be rewritten
2 sin B C
2 2 sin 2
as 21 cos B−C
2 = 2 sin B2 sin C2 and the 97 to 144 ratio condition can be rewritten as sin A
97
= 144 .
A B C
The first equation implies sin 2 = 2 sin 2 sin 2 , which we can plug into the second equation to get
cos A2 .

4. [30] Let n > 1 be an odd integer. On an n × n chessboard the center square and four corners are
deleted. We wish to group the remaining n2 − 5 squares into 21 (n2 − 5) pairs, such that the two squares
in each pair intersect at exactly one point (i.e. they are diagonally adjacent, sharing a single corner).
For which odd integers n > 1 is this possible?
Proposed by: Evan Chen
Answer: 3,5
Constructions for n = 3 and n = 5 are easy. For n > 5, color the odd rows black and the even rows
white. If the squares can be paired in the way desired, each pair we choose must have one black cell
and one white cell, so the numbers of black cells and white cells are the same. The number of black
cells is n+1
2 n − 4 or
n+1
2 n − 5 depending on whether the removed center cell is in an odd row. The
number of white cells is n−1 n−1
2 n or 2 n − 1. But
 
n+1 n−1
n−5 − n=n−5
2 2

so for n > 5 this pairing is impossible. Thus the answer is n = 3 and n = 5.

5. [35] Find all prime numbers p such that y 2 = x3 + 4x has exactly p solutions in integers modulo p.
In other words, determine all prime numbers p with the following property: there exist exactly p
ordered pairs of integers (x, y) such that x, y ∈ {0, 1, . . . , p − 1} and

p divides y 2 − x3 − 4x.

Proposed by: Evan Chen

Answer: p = 2 and p ≡ 3 (mod 4)
Clearly p = 2 works with solutions (0, 0) and (1, 1) and not (0, 1) or (1, 0).
If p ≡ 3 (mod 4) then −1 is not a quadratic residue, so for x3 + 4x 6= 0, exactly one of x3 + 4x
and −x3 − 4x is a square and gives two solutions (for positive and negative y), so there’s exactly two
solutions for each such pair {x, −x}. If x is such that x3 + 4x = 0, there’s exactly one solution.
If p ≡ 1 (mod 4), let i be a square root of −1 (mod p). The right hand side factors as x(x + 2i)(x − 2i).
For x = 0, 2i, −2i this is zero, there is one choice of y, namely zero. Otherwise, the right hand side is
nonzero. For any fixed x, there are either 0 or 2 choices for y. Replacing x by −x negates the right
hand side, again producing two choices for y since −1 is a quadratic residue. So the total number of
solutions (x, y) is 3 (mod 4), and thus there cannot be exactly p solutions.
Remark: This is a conductor 36 elliptic curve with complex multiplication, and the exact formula for
the number of solutions is given in http://www.mathcs.emory.edu/~ono/publications-cv/pdfs/
026.pdf.
6. [35] A nonempty set S is called well-filled if for every m ∈ S, there are fewer than 21 m elements of S
which are less than m. Determine the number of well-filled subsets of {1, 2, . . . , 42}.
Proposed by: Casey Fu
43


Answer: 21 − 1

Let an be the number of well-filled subsets whose maximum element is n (setting a0 = 1). Then it’s
easy to see that

a2k+1 = a2k + a2k−1 + · · · + a0

a2k+2 = (a2k+1 − Ck ) + a2k + · · · + a0 .

where Ck is the number of well-filled subsets of size k + 1 with maximal element 2k + 1.

We proceed to compute Ck . One can think of such a subset as a sequence of numbers 1 ≤ s1 < · · · <
sk+1 ≤ 2k + 1 such that si ≥ 2i − 1 for every 1 ≤ i ≤ k + 1. Equivalently, letting si = i + 1 + ti it’s
the number of sequences 0 ≤ t1 ≤ · · · ≤ tk+1 ≤ k + 1 such that ti ≥ i for every i. This gives the list of
x-coordinates of steps up in a Catalan path from (0, 0) to (k + 1, k + 1), so
 
1 2(k + 1)
Ck =
k + 2 (k + 1)

is equal to the (k + 1)th Catalan number.

From this we can solve the above recursion to derive that
 
n
an = .
b(n − 1)/2c

Consequently, for even n,  

n+1
a0 + · · · + an = an+1 = .
bn/2c
Putting n = 42 gives the answer, after subtracting off the empty set (counted in a0 ).
7. [40] Let q(x) = q 1 (x) = 2x2 + 2x − 1, and let q n (x) = q(q n−1 (x)) for n > 1. How many negative real
roots does q 2016 (x) have?
Proposed by: Ernest Chiu
22017 +1
Answer: 3

Define g(x) = 2x2 − 1, so that q(x) = − 12 + g x + 1

. Thus 2
 
N 1 N 1
q (x) = 0 ⇐⇒ =g x+
2 2

where N = 2016.
But, viewed as function g : [−1, 1] → [−1, 1] we have that g(x) = cos(2 arccos(x)). Thus, the equation
q N (x) = 0 is equivalent to   
1 1
cos 22016 arccos x + = .
2 2
Thus, the solutions for x are
 
1 π/3 + 2πn
x = − + cos n = 0, 1, . . . , 22016 − 1.
2 22016

So, the roots are negative for the values of n such that

1 π/3 + 2πn 5
π< < π
3 22016 3
which is to say
1 2016 1
(2 − 1) < n < (5 · 22016 − 1).
6 6
The number of values of n that fall in this range is 61 (5 · 22016 − 2) − 16 (22016 + 2) + 1 = 16 (4 · 22016 + 2) =
1 2017
3 (2 + 1).

8. [40] Compute
π
2 sin θ + 3 cos θ − 3
Z
dθ.
0 13 cos θ − 5

Proposed by: Carl Lian

3π 4 3
Answer: 13 − 13 log 2

We have
π π/2
2 sin θ + 3 cos θ − 3 2 sin 2x + 3 cos 2x − 3
Z Z
dθ = 2 dx
0 13 cos θ − 5 0 13 cos 2x − 5
π/2
4 sin x cos x − 6 sin2 x
Z
=2 dx
0 8 cos2 x − 18 sin2 x
π/2
sin x(2 cos x − 3 sin x)
Z
=2 dx
0 (2 cos x + 3 sin x)(2 cos x − 3 sin x)
Z π/2
sin x
=2 .
0 2 cos x + 3 sin x
 To compute the above integral we want to write sin x as a linear combination of the denominator and
its derivative:
Z π/2 Z π/2 1
sin x − 13 [−3(2 cos x + 3 sin x) + 2(3 cos x − 2 sin x)]
2 =2
0 2 cos x + 3 sin x 0 2 cos x + 3 sin x
"Z #
π/2 Z π
2 −2 sin x + 3 cos x
=− (−3) + 2
13 0 0 2 cos x + 3 sin x
 
2 3π π/2
=− − + 2 log(3 sin x + 2 cos x) |0
13 2
 
2 3π 3
=− − + 2 log
13 2 2
3π 4 3
= − log .
13 13 2

9. [40] Fix positive integers r > s, and let F be an infinite family of sets, each of size r, no two of which
share fewer than s elements. Prove that there exists a set of size r − 1 that shares at least s elements
with each set in F .
Proposed by: Victor Wang
This is a generalization of 2002 ISL C5.
Solution 1. Say a set S s-meets F if it shares at least s elements with each set in F . Suppose no
such set of size (at most) r − 1 exists. (Each S ∈ F s-meets F by the problem hypothesis.)
Let T be a maximal set such that T ⊆ S for infinitely many S ∈ F , which form F 0 ⊆ F (such T exists,
since the empty set works). Clearly |T | < r, so by assumption, T does not s-meet F , and there exists
U ∈ F with |U ∩ T | ≤ s − 1. But U s-meets F 0 , so by pigeonhole, there must exist u ∈ U \ T belonging
to infinitely many S ∈ F 0 , contradicting the maximality of T .
Comment. Let X be an infinite set, and a1 , . . . , a2r−2−s elements not in X. Then F = {B ∪ {x} :
B ⊆ {a1 , . . . , a2r−2−s }, |B| = r − 1, x ∈ X} shows we cannot replace r − 1 with any smaller number.
Solution 2. We can also use a more indirect approach (where the use of contradiction is actually
essential).
Fix S ∈ F and a ∈ S. By assumption, S \ {a} does not s-meet F , so there exists S 0 ∈ F such that S 0
contains at most s − 1 elements of S \ {a}, whence S ∩ S 0 is an s-set containing a. We will derive a
contradiction from the following lemma:
Lemma. Let F, G be families of r-sets such that any f ∈ F and g ∈ G share at least s elements. Then
there exists a finite set H such that for any f ∈ F and g ∈ G, |f ∩ g ∩ H| ≥ s.
Proof. Suppose not, and take a counterexample with r + s minimal; then F, G must be infinite and
r > s > 0.
Take arbitrary f0 ∈ F and g0 ∈ G; then the finite set X = f0 ∪ g0 meets F, G. For every subset Y ⊆ X,
let FY = {S ∈ F : S ∩ X = Y }; analogously define GY . Then the FY , GY partition F, G, respectively.
For any FY and y ∈ Y , define FY (y) = {S \ {y} : S ∈ FY }.
Now fix subsets Y, Z ⊆ X. If one of FY , GZ is empty, define HY,Z = ∅.
Otherwise, if Y, Z are disjoint, take arbitrary y ∈ Y , z ∈ Z. By the minimality assumption, there
exists finite HY,Z such that for any f ∈ FY (y) and g ∈ GZ (z), |f ∩ g ∩ HY,Z | ≥ s.
If Y, Z share an element a, and s = 1, take HY,Z = {a}. Otherwise, if s ≥ 2, we find again by
minimality a finite HY,Z (a) such that for f ∈ FY (a) and g ∈ GZ (a), |f ∩ g ∩ HY,Z | ≥ s − 1; then take
HY,Z = HY,Z (a) ∪ {a}.
S
Finally, we see that H = Y,Z⊆X HY,Z shares at least s elements with each f ∩ g (by construction),
contradicting our assumption.
10. [50] Let ABC be a triangle with incenter I whose incircle is tangent to BC, CA, AB at D, E, F .
Point P lies on EF such that DP ⊥ EF . Ray BP meets AC at Y and ray CP meets AB at Z. Point
Q is selected on the circumcircle of 4AY Z so that AQ ⊥ BC.
Prove that P , I, Q are collinear.
Proposed by: Evan Chen

X Y

Q E
Z

P
F

B D C

The proof proceeds through a series of seven lemmas.

Lemma 1. Lines DP and EF are the internal and external angle bisectors of ∠BP C.

Proof. Since DEF the cevian triangle of ABC with respect to its Gregonne point, we have that


−1 = EF ∩ BC, D; B, C .

Then since ∠DP F = 90◦ we see P is on the Apollonian circle of BC through D. So the conclusion
follows.

Lemma 2. Triangles BP F and CEP are similar.

Proof. Invoking the angle bisector theorem with the previous lemma gives
BP BP CP CP
= = = .
BF BD CD CE
But ∠BF P = ∠CEP , so 4BF P ∼ 4CEP .

Lemma 3. Quadrilateral BZY C is cyclic; in particular, line Y Z is the antiparallel of line BC through
∠BAC.
Proof. Remark that ∠Y BZ = ∠P BF = ∠ECP = ∠Y CZ.

Lemma 4. The circumcircles of triangles AY Z, AEF , ABC are concurrent at a point X such that
4XBF ∼ 4XCE.

Proof. Note that line EF is the angle bisector of ∠BP Z = ∠CP Y . Thus
ZF ZP YP YE
= = = .
FB PB PC EC
Then, if we let X be the Miquel point of quadrilateral ZY CB, it follows that the spiral similarity
mapping segment BZ to segment CY maps E to F ; therefore the circumcircle of 4AEF must pass
through X too.

Lemma 5. Ray XP bisects ∠F XE.

Proof. The assertion amounts to

XF BF FP
= = .
XE EC PE
The first equality follows from the spiral similarity 4BF X ∼ 4CEX, while the second is from
4BF P ∼ 4CEP . So the proof is complete by the converse of angle bisector theorem.

Lemma 6. Points X, P , I are collinear.

Proof. On one hand, ∠F XI = ∠F AI = 12 ∠A. On the other hand, ∠F XP = 12 ∠F XE = 12 ∠A. Hence,

X, Y , I collinear.

Lemma 7. Points X, Q, I are collinear.

Proof. On one hand, ∠AXQ = 90◦ , because we established earlier that line Y Z was antiparallel to
line BC through ∠A, hence AQ ⊥ BC means exactly that ∠AZQ = AY Q = 90◦ . On the other hand,
∠AXI = 90◦ according to the fact that X lies on the circle with diameter AI. This completes the
proof of the lemma.

Finally, combining the final two lemmas solves the problem.