HMMT February 2015

Saturday 21 February 2015

Algebra
1. Let Q be a polynomial
Q(x) = a0 + a1 x + · · · + an xn ,
where a0 , . . . , an are nonnegative integers. Given that Q(1) = 4 and Q(5) = 152, find Q(6).
Answer: 254 Since each ai is a nonnegative integer, 152 = Q(5) ≡ a0 (mod 5) and Q(1) = 4 =⇒
ai ≤ 4 for each i. Thus, a0 = 2. Also, since 54 > 152 = Q(5), a4 , a5 , . . . , an = 0.
Now we simply need to solve the system of equations

5a1 + 52 a22 + 53 a33 = 150

a1 + a2 + a3 = 2

to get

a2 + 6a3 = 7.

Since a2 and a3 are nonnegative integers, a2 = 1, a3 = 1, and a1 = 0. Therefore, Q(6) = 63 + 62 + 2 =

254.
1
2. The fraction 2015 has a unique “(restricted) partial fraction decomposition” of the form

1 a b c
= + + ,
2015 5 13 31
where a, b, c are integers with 0 ≤ a < 5 and 0 ≤ b < 13. Find a + b.
Answer: 14 This is equivalent to 1 = 13 · 31a + 5 · 31b + 5 · 13c.1 Taking modulo 5 gives 1 ≡ 3 · 1a
(mod 5), so a ≡ 2 (mod 5). Taking modulo 13 gives 1 ≡ 5 · 5b = 25b ≡ −b (mod 13), so b ≡ 12
(mod 13). The size constraints on a, b give a = 2, b = 12, so a + b = 14.
Remark. This problem illustrates the analogy between polynomials and integers, with prime powers
(here 51 , 131 , 311 ) taking the role of powers of irreducible polynomials (such as (x − 1)1 or (x2 + 1)3 ,
when working with polynomials over the real numbers).
Remark. The “partial fraction decomposition” needs to be restricted since it’s only unique “modulo
1”. Abstractly, the abelian group (or Z-module) Q/Z has a “prime power direct sum decomposition”
(more or less equivalent to Bezout’s identity, or the Chinese remainder theorem), but Q itself (as an
abelian group under addition) does not.
You may wonder whether there’s a similar “prime power decomposition” of Q that accounts not just
for addition, but also for multiplication (i.e. the full ring structure of the rationals). In some sense,
the “adeles/ideles” serve this purpose, but it’s not as clean as the partial fraction decomposition (for
additive structure alone)—in fact, the subtlety of adeles/ideles reflects much of the difficulty in number
theory!
3. Let p be a real number and c 6= 0 an integer such that

1 − (1 + x)10
 
c − 0.1 < xp < c + 0.1
1 + (1 + x)10

for all (positive) real numbers x with 0 < x < 10−100 . (The exact value 10−100 is not important. You
could replace it with any “sufficiently small number”.)
Find the ordered pair (p, c).
1 Note that this does actually have integer solutions by Bezout’s identity, as gcd(13 · 31, 5 · 31, 5 · 13) = 1.

Algebra
 Answer: (−1, −5) This is essentially a problem about limits, but phrased concretely in terms of
“small numbers” (like 0.1 and 10−100 ).
10
−10x+O(x2 )
We are essentially studying the rational function f (x) := 1−(1+x)
1+(1+x)10 = 2+O(x) , where the “big-O”
notation simply make precise the notion of “error terms”.2
Intuitively, f (x) ≈ −10x
2 = −5x for “small nonzero x”. (We could easily make this more precise if
we wanted to, by specifying the error terms more carefully, but it’s not so important.) So g(x) :=
xp f (x) ≈ −5xp+1 for “small nonzero x”.

• If p + 1 > 0, g will approach 0 (“get very small”) as x approaches 0 (often denoted x → 0), so
there’s no way it can stay above the lower bound c − 0.1 for all small nonzero x.
• If p + 1 < 0, g will approach −∞ (“get very large in the negative direction”) as x → 0, so there’s
no way it can stay below the upper bound c + 0.1 for all small nonzero x.
• If p + 1 = 0, g ≈ −5 becomes approximately constant as x → 0. Since c is an integer, we must
have c = −5 (as −5 is the only integer within 0.1 of −5).

Remark. Why does (p, c) = (−1, −5) actually satisfy the inequality? This is where the 10−100 kicks
in: for such small values of x, the “error” |g(x) − (−5)| of the approximation g ≈ −5 does actually
lie within the permitted threshold of ±0.1. (You can easily work out the details yourself, if you’re
interested. It’s something you might want to work out once or twice in your life, but rational functions
are “well-behaved” enough that we can usually rely on our intuition in these kinds of scenarios.)

4. Compute the number of sequences of integers (a1 , . . . , a200 ) such that the following conditions hold.

• 0 ≤ a1 < a2 < · · · < a200 ≤ 202.

• There exists a positive integer N with the following property: for every index i ∈ {1, . . . , 200}
there exists an index j ∈ {1, . . . , 200} such that ai + aj − N is divisible by 203.

Answer: 20503 Let m := 203 be an integer not divisible by 3. We’ll show the answer for general
such m is m⌈ m−1
2 ⌉.
Let x, y, z be the three excluded residues. Then N works if and only if {x, y, z} ≡ {N − x, N − y, N − z}
(mod m). Since x, y, z (mod m) has opposite orientation as N − x, N − y, N − z (mod m), this is
equivalent to x, y, z forming an arithmetic progression (in some order) modulo m centered at one of
x, y, z (or algebraically, one of N ≡ 2x ≡ y + z, N ≡ 2y ≡ z + x, N ≡ 2z ≡ x + y holds, respectively).
Since 3 ∤ m, it’s impossible for more than one of these congruences to hold (or else x, y, z would have to
be equally spaced modulo m, i.e. x − y ≡ y − z ≡ z − x). So the number of distinct 3-sets corresponding
to arithmetic progressions is m⌈ m−1
2 ⌉ (choose a center and a difference, noting that ±d give the same
arithmetic progression). Since our specific m = 203 is odd this gives m m−12 = 203 · 101 = 20503.
Remark. This problem is a discrete analog of certain so-called Frieze patterns. (See also Chapter 6,
Exercise 5.8 of Artin’s Algebra textbook.)

5. Let a, b, c be positive real numbers such that a+b+c = 10 and ab+bc+ca = 25. Let m = min{ab, bc, ca}.
Find the largest possible value of m.
25
Answer: 9 Without loss of generality, we assume that c ≥ b ≥ a. We see that 3c ≥ a + b + c = 10.
10
Therefore, c ≥ 3 .
2 For instance, the O(x2 ) refers to a function bounded by C|x|2 for some positive constant C, whenever x is close enough to

0 (and as the 10−100 suggests, that’s all we care about).

Algebra
 Since

0 ≤ (a − b)2
= (a + b)2 − 4ab
= (10 − c)2 − 4 (25 − c(a + b))
= (10 − c)2 − 4 (25 − c(10 − c))
= c(20 − 3c),

we obtain c ≤ 20
3 . Consider m = min{ab, bc, ca} = ab, as bc ≥ ca ≥ ab. We compute ab = 25−c(a+b) =
25 − c(10 − c) = (c − 5)2 . Since 10 20 25 25
3 ≤ c ≤ 3 , we get that ab ≤ 9 . Therefore, m ≤ 9 in all cases and
5 5 20
the equality can be obtained when (a, b, c) = ( 3 , 3 , 3 ).

6. Let a, b, c, d, e be nonnegative integers such that 625a + 250b + 100c + 40d + 16e = 153 . What is the
maximum possible value of a + b + c + d + e?
Answer: 153 The intuition is that as much should be in e as possible. But divisibility obstructions
like 16 ∤ 153 are in our way. However, the way the coefficients 54 > 53 · 2 > · · · are set up, we can at
least easily avoid having a, b, c, d too large (speifically, ≥ 2). This is formalized below.
First, we observe that (a1 , a2 , a3 , a4 , a5 ) = (5, 1, 0, 0, 0) is a solution. Then given a solution, replacing
(ai , ai+1 ) with (ai − 2, ai+1 + 5), where 1 ≤ i ≤ 4, also yields a solution. Given a solution, it turns out
all solutions can be achieved by some combination of these swaps (or inverses of these swaps).
Thus, to optimize the sum, we want (a, b, c, d) ∈ {0, 1}4 , since in this situation, there would be no way
to make swaps to increase the sum. So the sequence of swaps looks like (5, 1, 0, 0, 0) → (1, 11, 0, 0, 0) →
(1, 1, 25, 0, 0) → (1, 1, 1, 60, 0) → (1, 1, 1, 0, 150), yielding a sum of 1 + 1 + 1 + 0 + 150 = 153.
Why is this optimal? Suppose (a, b, c, d, e) maximizes a + b + c + d + e. Then a, b, c, d ≤ 1, or else
we could use a replacement (ai , ai+1 ) → (ai − 2, ai+1 + 5) to strictly increase the sum. But modulo 2
forces a odd, so a = 1. Subtracting off and continuing in this manner3 shows that we must have b = 1,
then c = 1, then d = 0, and finally e = 150.
Remark. The answer is coincidentally obtained by dropping the exponent of 153 into the one’s place.

7. Suppose (a1 , a2 , a3 , a4 ) is a 4-term sequence of real numbers satisfying the following two conditions:

• a3 = a2 + a1 and a4 = a3 + a2 ;
• there exist real numbers a, b, c such that

an2 + bn + c = cos(an )

for all n ∈ {1, 2, 3, 4}.

Compute the maximum possible value of

cos(a1 ) − cos(a4 )

over all such sequences (a1 , a2 , a3 , a4 ).

√
Answer: −9 + 3 13 Let f (n) = cos an and m = 1. The second (“quadratic interpolation”)
condition on f (m), f (m + 1), f (m + 2), f (m + 3) is equivalent to having a vanishing third finite
difference
f (m + 3) − 3f (m + 2) + 3f (m + 1) − f (m) = 0.
3 This is analogous to the “number theoretic” proof of the uniqueness of the base 2 expansion of a nonnegative integer.

Algebra
 This is equivalent to

f (m + 3) − f (m) = 3 [f (m + 2) − f (m + 1)]
⇐⇒ cos(am+3 ) − cos(am ) = 3 (cos(am+2 ) − cos(am+1 ))
   
am+2 + am+1 am+2 − am+1
= −6 sin sin
2 2
a  a 
m+3 m
= −6 sin sin .
2 2
Set x = sin am+3 and y = sin a2m . Then the above rearranges to



2

(1 − 2x2 ) − (1 − 2y 2 ) = −6xy ⇐⇒ x2 − y 2 = 3xy.

√
Solving gives y = x −3±2 13 . The expression we are trying to maximize is 2(x2 − y 2 ) = 6xy, so we want
√
x, y to have the same sign; thus y = x −3+2 13 .
Then |y| ≤ |x|, so since |x|, |y| ≤ 1, to maximize 6xy we can simply set x = 1, for a maximal value of
√
−3+ 13
√
6· 2 = −9 + 3 13.
8. Find the number of ordered pairs of integers (a, b) ∈ {1, 2, . . . , 35}2 (not necessarily distinct) such that
ax + b is a “quadratic residue modulo x2 + 1 and 35”, i.e. there exists a polynomial f (x) with integer
coefficients such that either of the following equivalent conditions holds:
• there exist polynomials P, Q with integer coefficients such that f (x)2 − (ax + b) = (x2 + 1)P (x) +
35Q(x);
• or more conceptually, the remainder when (the polynomial) f (x)2 − (ax + b) is divided by (the
polynomial) x2 + 1 is a polynomial with (integer) coefficients all divisible by 35.

Answer: 225 By the Chinese remainder theorem, we want the product of the answers modulo 5
and modulo 7 (i.e. when 35 is replaced by 5 and 7, respectively).
First we do the modulo 7 case. Since x2 + 1 is irreducible modulo 7 (or more conceptually, in F7 [x]),
exactly half of the nonzero residues modulo x2 + 1 and 7 (or just modulo x2 + 1 if we’re working in
2
F7 [x]) are quadratic residues, i.e. our answer is 1 + 7 2−1 = 25 (where we add back one for the zero
polynomial).
Now we do the modulo 5 case. Since x2 + 1 factors as (x + 2)(x − 2) modulo 5 (or more conceptually,
in F5 [x]), by the polynomial Chinese remainder theorem modulo x2 + 1 (working in F5 [x]), we want
the product of the number of polynomial quadratic residues modulo x ± 2. By centering/evaluating
polynomials at ∓2 accordingly, the polynomial squares modulo these linear polynomials are just those
reducing to integer squares modulo 5.4 So we have an answer of (1 + 5−1 2
2 ) = 9 in this case.
Our final answer is thus 25 · 9 = 225.
Remark. This problem illustrates the analogy between integers and polynomials (specifically here,
polynomials over the finite field of integers modulo 5 or 7), with x2 + 1 (mod 7) or x ± 2 (mod 5)
taking the role of a prime number. Indeed, just as in the integer case, we expect exactly half of the
(coprime) residues to be (coprime, esp. nonzero) quadratic residues.
9. Let N = 302015 . Find the number of ordered 4-tuples of integers (A, B, C, D) ∈ {1, 2, . . . , N }4 (not
necessarily distinct) such that for every integer n, An3 + Bn2 + 2Cn + D is divisible by N .
Answer: 24 Note that n0 = n0 , n1 = n1 , n2 = 2 n2 + n1 , n3 = 6 n3 + 6 n2 + n1 (generally see

http://en.wikipedia.org/wiki/Stirling_numbers_of_the_second_kind). Thus the polynomial

rewrites as        
n n n n
6A + (6A + 2B) + (A + B + 2C) +D ,
3 2 1 0
1
4 Thisis more explicit than necessary. By the same reasoning as in the previous paragraph, we can abstractly count 1 + 5 2−1
quadratic residues modulo x ± 2 (irreducible polynomials in F5 [x]) each (and then multiply/square to get the answer for x2 + 1).

Algebra
 which by the classification of integer-valued polynomials is divisible by N always if and only if 6A, 6A+
2B, A + B + 2C, D are always divisible by N .
We can eliminate B and (trivially) D from the system: it’s equivalent to the system 6A ≡ 0 (mod N ),
4A − 4C ≡ 0 (mod N ), B ≡ −A − 2C (mod N ), D ≡ 0 (mod N ). So we want 12 times the number of
(A, C) with A ≡ 0 (mod N/6), C ≡ A (mod N/4). So there are N/(N/6) = 6 choices for A, and then
given such a choice of A there are N/(N/4) = 4 choices for C. So we have 6 · 4 · 12 = 24 solutions total.
10. Find all ordered 4-tuples of integers (a, b, c, d) (not necessarily distinct) satisfying the following system
of equations:
a 2 − b2 − c 2 − d 2 = c − b − 2
2ab = a − d − 32
2ac = 28 − a − d
2ad = b + c + 31.

Answer: (5, −3, 2, 3) We first give two systematic solutions using standard manipulations and
divisibility conditions (with some casework), and then a third solution using quaternionic number
theory (not very practical, so mostly for your cultural benefit).
Solution 1. Subtract the second equation from the third to get a(c − b + 1) = 30. Add the second
and third to get 2a(b + c) = −4 − 2d. Substitute into the fourth to get
31a − 2
2a(2ad − 31) = −4 − 2d ⇐⇒ a(31 − 2ad) = 2 + d ⇐⇒ d = ,
2a2 + 1
which in particular gives a 6≡ 1 (mod 3). Then plugging in a factor of 30 for a gives us the system of
equations b + c = 2ad − 31 and c − b + 1 = 30/a in b, c. Here, observe that b + c is odd, so c − b + 1
is even. Thus a must be odd (and from earlier a 6≡ 1 (mod 3)), so a ∈ {−1, ±3, 5, ±15}. Manually
checking these, we see that the only possibilities we need to check are (a, d) = (5, 3), (−1, −11), (−3, −5),
corresponding to (b, c) = (−3, 2), (11, −20), (5, −6). Then check the three candidates against first
condition a2 − b2 − c2 − d2 = c − b − 2 to find our only solution (a, b, c, d) = (5, −3, 2, 3).
Solution 2. Here’s an alternative casework solution. From 2ad = b + c + 31, we have that b + c is
odd. So, b and c has different parity. Thus, b2 + c2 ≡ 1 (mod 4). Plugging this into the first equation,
we get that a and d also have the same parity.
So, a2 − b2 − c2 − d2 ≡ −1 (mod 4). Thus, c − b − 2 ≡ −1 (mod 4). So, c ≡ b + 1 (mod 4).
From taking modulo a in the second and third equation, we have a | d + 32 and a | 28 − d. So, a | 60.
Now, if a is even, let a = 2k and d = 2m. Plugging this in the second and third equation, we get
2kc = 14 − k − m and 2kb = k − m − 16. So, k(c − b) = 15 − k.
15−k 15
We can see that k 6= 0. Therefore, c − b = k = k − 1.
15
But c − b ≡ 1 (mod 4). So, k − 1 ≡ 1 (mod 4), or 15
k ≡2 (mod 4) which leads to a contradiction.
So, a is odd. And we have a | 60. So, a | 15. This gives us 8 easy possibilities to check...
Solution 3. The left hand sides clue us in to the fact that this problem is secretly about quaternions.
Indeed, we see that letting z = a + bi + cj + dk gives
(z − i + j)z = −2 − 32i + 28j + 31k.

Taking norms gives N (z − i + j)N (z) = 22 + 322 + 282 + 312 = 2773 = 47 · 59. By the triangle inequality,
N (z), N (z − i + j) aren’t too far apart, so they must be 47, 59 (in some order).
Thus z, z − i + j are Hurwitz primes.5 We rely on the following foundational lemma in quaternion
number theory:
5 For the purposes of quaternion number theory, it’s simpler to work in the the Hurwitz quaternions H = hi, j, k, 1+i+j+k 2
iZ ,
which has a left- (or right-) division algorithm, left- (resp. right-) Euclidean algorithm, is a left- (resp. right-) principal ideal
domain, etc. There’s no corresponding division algorithms when we’re working with the Lipschitz quaternions, i.e. those with
integer coordinates.

Algebra
 Lemma. Let p ∈ Z be an integer prime, and A a Hurwitz quaternion. If p | N (A), then the HA + Hp
(a left ideal, hence principal) has all element norms divisible by p, hence is nontrivial. (So it’s either
Hp or of the form HP for some Hurwitz prime P .)
In our case, it will suffice to apply the lemma for A = −2 − 32i + 28j + 31k at primes p = 47 and q = 59
to get factorizations (unique up to suitable left/right unit multiplication) A = QP and A = P ′ Q′
(respectively), with P, P ′ Hurwitz primes of norm p, and Q, Q′ Hurwitz primes of norm q. Indeed,
these factorizations come from HA + Hp = HP and HA + Hq = HQ′ .
We compute by the Euclidean algorithm:
HA + H(47) = H(−2 − 32i + 28j + 31k) + H(47)
= H(−2 + 15i − 19j − 16k) + H(47)
−2 + 15i − 19j − 16k
= [H(47 · 18) + H(47)(−2 − 15i + 19j + 16k)]
47 · 18
−2 + 15i − 19j − 16k
= [H18 + H(−2 + 3i + j − 2k)]
18
−2 + 15i − 19j − 16k
= H(−2 + 3i + j − 2k)
18
−54 − 90i + 54j − 36k
=H
18
= H(−3 − 5i + 3j − 2k).
Thus6 there’s a unit7 ǫ such that P = ǫ(−3 − 5i + 3j − 2k).
Similarly, to get P ′ , we compute
AH + 47H = (−2 − 32i + 28j + 31k)H + 47H
= (−2 + 15i − 19j − 16k)H + 47H
−2 + 15i − 19j − 16k
= [(47 · 18)H + 47(−2 − 15i + 19j + 16k)H]
47 · 18
−2 + 15i − 19j − 16k
= [18H + (−2 + 3i + j − 2k)H]
18
−2 + 15i − 19j − 16k
= (−2 + 3i + j − 2k)H
18
−54 + 18i + 18j + 108k
= H
18
= (−3 + i + j + 6k)H,
so there’s a unit ǫ′ with P ′ = (−3 + i + j + 6k)ǫ′ .
Finally, we have either z = ǫ(−3 − 5i + 3j − 2k) for some ǫ, or z − i + j = (−3 + i + j + 6k)ǫ′ for some
ǫ′ . Checking the 24 + 24 cases (many of which don’t have integer coefficients, and can be ruled out
immediately) gives z = iP = 5 − 3i + 2j + 3k as the only possibility.
Remark. We have presented the most conceptual proof possible. It’s also possible to directly compute
based on the norms only, and do some casework. For example, since 47 ≡ 3 (mod 4), it’s easy to
check the only ways to write it as a sum of four squares are (±5)2 + (±3)2 + (±3)2 + (±2)2 and
(±3)2 + (±1)2 + (±1)2 + (±6)2 .
Remark. For a systematic treatment of quaternions (including the number theory used above), one
good resource is On Quaternions and Octonions: Their Geometry, Arithmetic, and Symmetry by John
H. Conway and Derek A. Smith. A more focused treatment is the expository paper Factorization of
Hurwitz Quaternions by Boyd Coan and Cherng-tiao Perng.
For an example of interesting research in this rather exotic area, see the Metacommutation of Hurwitz
primes paper by Henry Cohn and Abhinav Kumar.
6 Hidden computations: we’ve used 47 · 18 = 846 = 22 + 152 + 192 + 162 , and 18 = N (−2 + 3i + j − 2k).
7 i.e. one of ±1, ±i, ±j, ±k, ±1±i±j±k
2

Algebra