2023 Session A

A1. For a positive integer n, let fn (x) = cos(x) cos(2x) cos(3x) · · · cos(nx). Find the smallest
n such that |fn′′ (0)| > 2023.

Answer: 18

Solution 1: The Taylor series is

x2 (2x2 ) (nx)2
    
fn (x) = 1 − + ··· 1− + ··· ··· 1 − + ···
2 2 2
x 2
12 + 2 2 + · · · + n2 + · · · .


=1−
2
Therefore (using the well-known summation formula for sums of squares)

n(n + 1)(2n + 1)
fn′′ (0) = −(12 + 22 + · · · + n2 ) = − .
6

The question is then to find the minimum n such that n(n+1)(2n+1)

6 > 2023. One can calculate
that this occurs at n = 18, where the sum is 3 · 19 · 37 = 2109 (and at n = 17 it is 1785).

Solution 2: By the product rule,

fn′ (x) = − sin(x) cos(2x) cos(3x) · · · cos(nx) − 2 cos(x) sin(2x) cos(3x) · · · cos(nx)
− · · · − n cos(x) cos(2x) · · · cos((n − 1)x) sin(nx)
= −fn (x) (tan(x) + 2 tan(2x) + · · · + n tan(nx))

for x sufficiently small that all the tangents are well-defined. Applying the product rule again
and substituting x = 0,

fn′′ (0) = −fn′ (0) (tan(0) + 2 tan(2 · 0) + · · · + n tan(n · 0))

− fn (0) sec2 (0) + 4 sec2 (2 · 0) + · · · + n2 sec2 (n · 0)

= −(1 + 4 + · · · + n2 ).
′′ (0)| = 1 + 4 + · · · + 172
One can compute directly (or using the formula in Solution 1) that |f17
′′ 2
= 1785 and |f18 (0)| = 1 + 4 + · · · + 18 = 1785 + 324 = 2109, so the answer is 18.

1
A2. Let n be an even positive integer. Let p be a monic, real polynomial of degree 2n; that
is to say, p(x) = x2n + a2n−1 x2n−1 + · · · + a1 x + a0 for some real coefficients a0 , . . . , a2n−1 .
Suppose that p(1/k) = k 2 for all integers k such that 1 ≤ |k| ≤ n. Find all other real numbers
x for which p(1/x) = x2 .

Answer: ±1/n!

Solution 1: The given condition can be equivalently written as p(x) = x12 for x = ± k1 , k =
1, . . . , n. Now define g(x) := x2 p(x)−1, and note that p(1/x) = x2 is equivalent to g(1/x) = 0.
Notice that g is a monic polynomial of degree 2n + 2, and by the preceding observation, it has
roots at all x = ± k1 . Unique factorization of polynomials (and/or the Fundamental Theorem
of Algebra) then implies that
     
1 1 1 1
g(x) = (x − 1)(x + 1) x − x+ ··· x − x+ · (x2 + ax + b)
2 2 n n
   
2

2 1 1
= x −1 x − · · · x − 2 · (x2 + ax + b),
2
4 n
where a and b are real numbers.
In order to determine these final coefficients, first note that by definition of g(x), the
(−1)n
coefficient of x is zero. But on the other hand, this coefficient is a, so a = 0. Now
n!2
consider the value at x = 0, which gives
(−1)n
g(0) = −1 = b.
n!2
We therefore conclude (using that n is even) that b = −n!2 . In all,
   
2

2 1 1
· · · x − 2 · x2 − n!2 .
2


g(x) = x − 1 x −
4 n
1
Finally, we see that g(1/x) has the additional roots x = ± n! .

Solution 2: We first show that p is even.

Claim. Suppose that q is a monic, degree 2n polynomial. If there exists a sequence of distinct
positive values x1 , . . . , xn such that q(xj ) = q(−xj ) for 1 ≤ j ≤ n, then q is even.
Proof. The polynomial q(x) − q(−x) has degree at most 2n − 1, but has (at least) 2n roots
±x1 , . . . , ±xn . Therefore, q(x) − q(−x) is identically zero, so q is even.

Thus p(x) = s(x2 ), where s is a monic, degree n polynomial such that s(1/k 2 ) = k 2 for
1 ≤ k ≤ n. Let h(x) := x · s(x) − 1. Then h is a monic, degree n + 1 polynomial with roots
at k12 , so
   
1 1
h(x) = (x − 1) x − · · · x − 2 (x + b)
4 n
for some real number b. Plugging in x = 0 gives
(−1)n
h(0) = −1 = b.
n!2
Finally, substituting x2 for x gives h(x2 ) = x2 p(x) − 1, and the remainder of the proof
proceeds as in Solution 1.

2
Remark. Although this is not needed in the proof of the claim, one can use Lagrange inter-
polation to say slightly more. For 1 ≤ aj ≤ n, let aj := q(±xj ). Define L(x) as the unique
polynomial of degree strictly less than n with the values L(x2j ) = aj for all j (Lagrange
interpolation gives a formula for L). Then

q(x) = x2 − x21 · · · x2 − x2n + L(x2 ).

3
A3. Determine the smallest positive real number r such that there exist differentiable func-
tions f : R → R and g : R → R satisfying

(a) f (0) > 0,

(b) g(0) = 0,

(c) |f ′ (x)| ≤ |g(x)| for all x,

(d) |g ′ (x)| ≤ |f (x)| for all x, and

(e) f (r) = 0.

Answer: π/2

Solution 1: Let h(x) := f (x)2 + g(x)2 . Then, using the AM-GM inequality,

|h′ (x)| = 2(f (x)f ′ (x) + g(x)g ′ (x)) ≤ 2 |f (x)||f ′ (x)| + |g(x)||g ′ (x)|

≤ 4|f (x)||g(x)| ≤ 2 f (x)2 + g(x)2 = 2h(x).

Thus, h(x)e2x has a nonnegative derivative. For x ≥ 0, it follows that h(x)e2x ≥ h(0)e0 =
f (0)2 , so h(x) ≥ f (0)2 e−2x > 0.
Now define θ(x) := tan−1 fg(x) (x) on the largest interval around x = 0 on which f (x) > 0
(this includes a neighborhood of 0 by continuity of f ). Then

1 g ′ (x)f (x) − g(x)f ′ (x) g ′ (x)f (x) − g(x)f ′ (x)

θ′ (x) = g(x)2
· = ,
1+ f (x)2 f (x)2 + g(x)2
f (x)2

which implies that |θ′ (x)| ≤ 1, and (since θ(0) = 0) therefore |θ(x)| ≤ x for x ≥ 0 such that
θ is defined.
Finally, observe that f (x)2 = h(x) cos2 θ(x), and therefore f (x)2 ≥ f (0)2 e−2x cos2 x > 0
for x ∈ (0, π/2). Thus, r ≥ π/2. Letting f (x) = cos x and g(x) = sin x shows that r = π/2 is
possible.

Solution 2: Notice that f (x) = cos x and g(x) = sin x satisfy all the conditions of the
problem with r = π/2.
To see that no smaller value of r is possible, we claim that f (x) ≥ f (0) cos x for 0 ≤ x <
π/2. If not, then there is some z ∈ (0, π/2) such that f (z)/ cos z < f (0). Since f (x)/ cos x
is continuous for 0 ≤ x < π/2 and f (0)/ cos 0 = f (0) > 0, we can choose z such that
f (x)/ cos x > 0 [whence f (x) > 0] for 0 ≤ x ≤ z. Since f (x)/ cos x is differentiable for
0 < x < z, the Mean Value Theorem implies that its derivative must be negative at some
y ∈ (0, z). Thus, f ′ (y) cos y + f (y) sin y < 0. Since |f ′ (y)| ≤ |g(y)| and cos y > 0, it follows
that f (y) sin y − |g(y)| cos y < 0.
Let h(x) = f (x) sin x − |g(x)| cos x. Since h(0) = 0, h(y) < 0, and h is continuous, there
is a greatest value w ∈ [0, y) such that h(w) ≥ 0. Then for w < x < y, we have h(x) < 0 and
f (x) sin x ≥ 0 [since f (x) > 0 for 0 ≤ x ≤ z and y < z], so |g(x)| cos x > 0, and thus |g(x)| > 0,
for all such x. In particular, h is differentiable on the interval (w, y), so by the Mean Value
Theorem h′ (x) < 0 for some x ∈ (w, y). Further, since g is continuous and nonzero on (w, y),

4
it does not change sign on this interval; without loss of generality, assume that g is positive on
(w, y) [otherwise, replace g with −g]. Then h′ (x) = (f ′ (x) + g(x)) sin x + (f (x) − g ′ (x)) cos x.
Since h′ (x) < 0, sin x > 0, and cos x > 0, we must have f ′ (x) + g(x) < 0 or f (x) − g ′ (x) < 0.
But since f (x) > 0 and g(x) > 0, this would require either |f ′ (x)| > g(x) = |g(x)| or
|g ′ (x)| > f (x) = |f (x)|, a contradiction.

5
A4. Let v1 , . . . , v12 be unit vectors in R3 from the origin to the vertices of a regular icosa-
hedron. Show that for every vector v ∈ R3 and every ε > 0, there exist integers a1 , . . . , a12
such that ∥a1 v1 + · · · + a12 v12 − v∥ < ε.

Solution 1: We first claim that the vertices of a regular pentagon centered at the origin
in R2 generate a dense additive subgroup. Identify R2 ∼ = C, and assume without loss of
2πin
generality that the vertices are the fifth roots of unity ζ5n = e 5 . Define

2π π
r := ζ5 + ζ54 = 2 cos < 2 cos = 1
5 3
√
(in fact, r = 5−1 2
2 , which follows from observing that r + r = 1). Then the positive powers
of r accumulate at 0, and are all contained in the subring Z[ζ5 ]. Therefore this subring is
dense in R, since it contains all {mrn | m ∈ Z, n ≥ 0}. Furthermore, it similarly contains a
dense subset of ζ5 R, and thus a dense subset of R + ζ5 R = C.
Now suppose that v1 , . . . , v5 are the neighbors of some fixed vertex v in the icosahedron.
Then v1 − v2 , v2 − v3 , . . . , v5 − v1 are the vertices of a regular pentagon in the plane perpen-
dicular to the line through 0 and v. Therefore the set of vertices generates a dense set in this
plane, and similarly, in the plane perpendicular to (say) the line through 0 and v1 . These
two planes span R3 .

Solution 2: Write the vertex-vectors as ±v1 , . . . , ±v6 where v2 , . . . , v6 are each adjacent to
v1 , and are adjacent to each other in the pairs (v2 , v3 ), (v3 , v4 ), (v4 , v5 ), (v5 , v6 ), (v6 , v2 ). Then
v2 is also adjacent to −v4 and −v5 , etc. Since all sides of the icosahedron have the same
length and vj · vj = 1 for all j, the value of vj · vk is the same for all pairs of adjacent vertices
(vj , vk ). Thus,

v2 · v3 = v3 · v4 = v4 · · · v5 = v5 · v6 = v6 · v2
= −v2 · v4 = −v2 · v5 = −v3 · v5 = −v3 · v6 = −v4 · v6 .

Then the cross terms 2vj · vk cancel each other in the following calculation:

(v2 + · · · + v6 ) · (v2 + · · · + v6 ) = v2 · v2 + · · · + v6 · v6 = 5.
√
By symmetry and adjacency, v2 + · · · v6 is a positive multiple of v1 , so v2 + · · · v6 = 5v1 .
By Kronecker’s
√ Theorem on Diophantine approximation, the integer linear combinations of
1 and 5 are dense in the reals, so the integer linear combinations of v1 and v2 + · · · + v6
are dense in the line spanned by v1 . By the analogous argument, appropriate integer linear
combinations of v1 , . . . , v6 are dense in the line spanned by v2 and in the line spanned by v3 .
Since v1 , v2 , v3 are not coplanar, they span three-space. Thus, every vector v in three-space
can be written c1 v1 + c2 v2 + c3 v3 , and since each term in this sum can be approximated
arbitrarily closely by an integer linear combination of v1 , . . . , v6 , so can v.

6
A5. For a nonnegative integer k, let f (k) be the number of ones in the base 3 representation
of k. Find all complex numbers z such that
31010
X−1
(−2)f (k) (z + k)2023 = 0.
k=0

√
1 − 31010 1 − 31010 32020 − 1
Answer: , ± i
2 2 4
Solution 1: If n is an integer, a quasi-base-3 representation of n is n = aN 3N + aN −1 3N −1 +
· · · + a1 · 3 + a0 , where all aj ∈ {−1, 0, 1}. This can also be written in the shorthand
(aN aN −1 · · · a0 )3 , with parentheses used as needed for clarity. For example, 7 = 1(−1)13 . If
the leading digit is required to be ±1, then it is a standard fact that this representation is
unique (both existence and uniqueness are easily proven by induction). However, here we
will also be interested in representations with some number of leading zeros. Let AN :=
3N − 1
= 11 · · · 13 . It can similarly be shown that if |n| ≤ AN , then n has a unique quasi-
2 | {z }
N digits
base-3 representation consisting of exactly N digits.
Define f0,N (ℓ) to be the number of zeros when ℓ is written in its N -digit quasi-base-3
representation. For example, f0,3 (8) = 1 and f0,5 (8) = 3, as the left-extended quasi-base-3
expansion is 8 = . . . 0010(−1)3 .
Claim. Let u = z + AN . The sum in the problem statement is equivalent to (with N = 1010)
AN
X
SN (u) := (−2)f0,N (ℓ) (u + ℓ)2N +3 . (1)
ℓ=−AN

This follows from the summation shift k = ℓ + AN , as well as the straightforward observa-
tion that f (ℓ + AN ) = f0,N (ℓ) (note that ℓ 7→ ℓ + AN is a bijection from N -digit quasi-base-3
representations to N -digit base-3 representations for the range ℓ = −AN , . . . , AN , while the
definition of f is unchanged by the presence of leading zeros).
If d is an integer, define the centered, d-shifted second-difference operator by

∆2d (h(x)) := h(x + d) − 2h(x) + h(x − d).

This operator satisfies some simple properties that will be helpful later for reducing longer
expressions.
Lemma 1. 1. ∆2d is a parity-preserving operator on functions (i.e. ∆2d (h) is an even/odd
function as h is).
2. ∆2d acts on monomials as
X n
∆2d (xn ) =2 d2j xn−2j .
n 2j
1≤j≤ 2

3. If p(x) is a polynomial of degree n, then ∆2d (p(x)) is a polynomial of degree n − 2 (or

is identically zero if n ≤ 1).

7
Proof. 1. Suppose that h is an even or odd function. By definition, this means that
h(−x) = sgn(h) · h(x), where sgn(h) = ±1 denotes the parity of h. Then

∆2d (h(−x)) = h(−x + d) − 2h(−x) + h(−x − d)

= h(−(x + d)) − 2h(−x) + h(−(x − d))
= sgn(h) · ∆2d (h(x)).

2. By the Binomial Theorem,

n   n   X n
X n X n
∆2d (xn ) = j n−j
d x + j n−j
(−d) x =2 d2j xn−2j .
j j n 2j
j=1 j=1 1≤j≤ 2

3. This follows by considering only the highest degree term in the sum from part 2.

We next show that sums of the form (1) can be written in terms of the second-difference
operator.

Lemma 2. If N is a positive integer, then

AN
X
∆23N −1 · · · ∆23 ∆21 h(x) = (−2)f0,N (ℓ) h(x + ℓ).
ℓ=−AN

Proof. The base case of N = 0 simply states that h(x) = h(x). For the inductive step,
AN
X
∆23N (−2)f0,N (ℓ) h(x + ℓ)
ℓ=−AN
AN
X
(−2)f0,N (ℓ) h x + 3N + ℓ − 2h(x + ℓ) + h x − 3N + ℓ




=
ℓ=−AN
3NX
+AN AN
X
= (−2)f0,N (ℓ) h(x + ℓ) + (−2)f0,N (ℓ)+1 h(x + ℓ)
ℓ=3N −AN ℓ=−AN
−3N
X +AN
+ (−2)f0,N (ℓ) h(x + ℓ)
ℓ=−3N −AN
AN +1
X
= (−2)f0,N +1 (ℓ) h(x + ℓ).
ℓ=−AN +1

The final line follows because in the first and third sum ℓ has leading coefficient ±1, so
f0,N +1 (ℓ) = f0,N (ℓ), whereas in the middle sum the leading coefficient is 0, so f0,N +1 (ℓ) =
f0,N (ℓ) + 1. Furthermore, the three summation ranges combine to one because AN +1 =
3N + AN by definition.

8
 In particular, Lemma 2 implies that

SN (u) = ∆23N −1 · · · ∆23 ∆21 u2N +3 , (2)

with the implicit notational assumption that the ∆2 operators now act on the variable u.
Lemma 1 parts 1 and 3 then imply that SN (u) reduces to an odd, cubic polynomial (i.e. of
the form au3 + bu). The exact coefficients can now be determined using Lemma 1 part 2.
In particular, only the two highest-order terms are relevant. For n ≥ 4, the Lemma states
that    
n 2 n−2 n 4 n−4
∆2d (xn ) = 2 + O xn−6 .


d x +2 d x
2 4
For a nonnegative integer m and real number x, the falling factorial is defined by (x)m :=
x(x − 1)(x − 2) · · · (x − m + 1).
Lemma 3. If N , k and d are positive integers, where k ≥ 4, then

∆2dN −1 · · · ∆2d ∆21 x2N +k

 
N (N −1) k k(k − 1)  2 2 2(N −1)

k−2

k−4
= (2N + k)2N · d x + 1 + d + ··· + d x +O x .
3·4
The statement is also true for k ≤ 3, except that any monomials in x with negative exponents
do not appear.
Proof. We induct on N . The base case is N = 1, and by Lemma 1,
(2 + k)4 4 k−2
∆21 x2+k = (2 + k)2 · 12 xk + 1 x .
3·4
For the inductive step, suppose that the statement is true for N . Then we have (replacing k
by 2 + k when applying the statement for N )

∆2dN ∆2dN −1 · · · ∆21 x2(N +1)+k

 h
= ∆2dN (2(N + 1) + k)2N · dN (N −1) x2+k
(2 + k)(1 + k)  2   
2 2(N −1) k k−2
+ 1 + d + ··· + d x +O x
3·4
 
N (N −1) 2N k (2 + k)4 4N k−2
= (2(N + 1) + k)2N · d (2 + k)(1 + k)d x + d x
3·4
(2 + k)(1 + k)  2   
2 2(N −1) 2N k−2 k−4
+ 1 + d + ··· + d · k(k − 1)d x +O x
3·4

k(k − 1)  2 
= (2(N + 1) + k)2N +2 · d N (N +1)
xk + 1 + d2 + · · · + d2(N −1) + d2N xk−2
3·4
 i
k−4
+O x .

Applying Lemma 3 (with d = 3 and k = 3) to (2) gives

3 · 2 32N − 1
 
N (N −1) 3
SN (u) = (2N + 3)2N · 3 u + u .
3 · 4 32 − 1

9
 ) ( √
32N −1 32N − 1
The cubic u3
+ has the roots u ∈ 0, ±
16 u i . Finally, these are translated to
4
roots of the original expression in z using (1).

Solution 2: For nonnegative integers n, let

−1
3X n
1
gn (z) = (−2)f (k) (z + k)2n+5 ,
(2n + 5)!
k=0

′′
and notice that the equation to be solved is (2023!)g1010 (z) = 0. Let cn = (3n − 1)/2. We
will prove by induction on n that
3 (32n − 1)(z + cn )
 
′′ n2 −n (z + cn )
gn (z) = 3 + .
6 96

Since g0 (z) = z 5 /5!, we have g0′′ (z) = z 3 /3! = (z + c0 )3 /6, verifying the base case. Assume
now the formula for gn′′ above holds for a particular n ≥ 0. Observe that f (j + 3n ) = f (j) + 1
and f (j + 2 · 3n ) = f (j) for 0 ≤ j < 3n . Then, with the substitutions ℓ = k − 3n and
m = k − 2 · 3n in the corresponding sums below,

3 −1 n+1
′′ (2n + 7)(2n + 6) X
gn+1 (z) = (−2)f (k) (z + k)2n+5
(2n + 7)!
k=0
n −1
3X −1
3X n
1 f (k) 2n+5 1
= (−2) (z + k) + (−2)f (ℓ)+1 (z + ℓ + 3n )2n+5
(2n + 5)! (2n + 5)!
k=0 ℓ=0
n −1
3X
1
+ (−2)f (m) (z + m + 2 · 3n )2n+5
(2n + 5)!
m=0
= gn (z) − 2gn (z + 3n ) + gn (z + 2 · 3n )
Z 3n Z 3n Z 3n Z 3n +t
′ ′
= n
gn (z + 3 + t)dt − gn (z + t)dt = gn′′ (z + s)ds dt
0 0 0 t
Z 2·3n Z min(3n ,s)
= gn′′ (z + s)dt ds
0 max(0,s−3n )
Z 2·3n Z 3n
= n
(3 − |s − 3 n
|)gn′′ (z + s)ds = (3n − |u|)gn′′ (z + u + 3n )du
0 −3n
3n
(z + u + 3n + cn )3 (32n − 1)(z + u + 3n + cn )
Z  
n2 −n n
=3 (3 − |u|) + dt.
−3n 6 96

The integrand is 3n − |u| times a cubic polynomial of u. Since 3n − |u| is an even function of
u, its integral from −3n to 3n times an odd power of u is zero, so we can eliminate the cubic
and linear terms from the cubic polynomial of u. Having done so, the integrand becomes an
even function of u, which we can integrate instead from 0 to 3n and double the result. Using

10
also the fact that 3n + cn = cn+1 , we have
Z 3n
(z + cn+1 )3 (32n − 1)(z + cn+1 )
 
′′ n2 −n n 2 (z + cn+1 )
gn+1 (z) = 2 · 3 (3 − u) 3u + + du
0 6 6 96
 4n
(z + cn+1 ) 32n (z + cn+1 )3 (32n − 1)(z + cn+1 )
 
n2 −n 3
=2·3 + +
4 6 2 6 96
3
  2n 2n
 
2n n2 −n (z + cn+1 ) 3 (3 − 1)
=3 ·3 + + (z + cn+1 )
6 12 96
3 32n+2 − 1
 
(n+1)2 −(n+1) (z + cn+1 )
=3 + (z + cn+1 ) ,
6 96

completing the induction.

′′
Recall that the answers to the problem are the roots of g1010 , which are those z for which
z + c1010 = 0 or (z + c1010 )2 = −(32020 − 1)/16, yielding the answers given above.

Solution 3: For a sequence of integers b, set p(b) = #{i : bi = 0}. For nonnegative integers
n, m, define !m
X n−1
X
hn,m (y) = (−2)p(b) y + bi 3i
b∈{−1,0,1}n i=0

As described in Solution 1, the polynomial in the problem is hn,m (y) for n = 1010, m = 2n+3
3n −1
and y = z + 1 + 3 + · · · + 3n−1
P= z + 2 .
We have that p(b) = n − i |bi |, and we can expand the polynomial by the Binomial and
Multinomial Formulas as follows:
m   n−1
!k
X P X m X
hn,m (y) = (−2)n− |bi | y m−k bi 3i
n
k
b∈{−1,0,1} k=0 i=0
m     n−1
X m m−k X
−
P
i |bi |
X k Y
= y (−2)n (−2) (bi 3i )ai
k a0 , a1 , . . .
k=0 b∈{−1,0,1}n a0 +···+an−1 =k i=0
aj ≥0
m     n−1
X m m−k X k X X Y
= y (−2)n ··· (−2)−|bi | (bi 3i )ai
k a0 , a1 , . . .
k=0 a0 +···+an−1 =k b0 ∈{−1,0,1} bn−1 ∈{−1,0,1} i=0
aj ≥0
 
m     n−1
X m m−k X k Y X
= y (−2)n  (−2)−|bi | (bi 3i )ai 
k a0 , a1 , . . .
k=0 a0 +···+an−1 =k i=0 bi ∈{−1,0,1}
aj ≥0

Notice that the terms in the final parentheses evaluate as (recall that the correct conven-
tion for powers of 0 in multinomial expansions is 00 = 1, and 0ℓ = 0 for ℓ ≥ 1)

(
X 0 , if ai = 0 or odd
(−2)−|bi | (bi 3i )ai = (−2)−1 (−3i )ai +0ai +(−2)−1 (3i )ai =
bi ∈{−1,0,1}
−3ai i , if ai ≥ 2 is even

11
 P the only nonzero summands can occur only when all ai ≥ 2 and are even and so
Thus
k= ai ≥ 2n and is even. Since k ≤ m = 2n + 3, we have either k = 2n and then ai = 2
for all i, or k = 2m + 2 and then aj = 4 for some j and ai = 2 for all i ̸= j. The summation
simplifies as

 
   n−1    n−1
 3 2n + 3 2n Y 2n + 3 2n + 2 X Y
hn,2n+3 (y) = (−2)n  2i
(−34j ) (−32i )

y 2n 2, 2, . . .
(−3 ) + y
2n + 2 4, 2, 2, · · · 
i=0 j=0 0≤i≤n−1
i̸=j
 
n−1   n−1
Y 2n + 3 (2n)! 3 (2n + 2)! X
= (−2)n (−1)n 32i  y + y(2n + 3) 32j 
3 2n 12 · 2n
i=0 j=0

Factoring out (2n+3)!

6·2n and simplifying the geometric sum over j we are left with looking
for the solutions of
32n − 1
y3 + y = 0,
16
√
32n −1 3n −1
whose roots are {0, ±i 4 }. Finally, recalling that z =y− 2 , the roots of the original
√
3n −1 n 2n
polynomial are then {− 2 , − 3 2−1 ± i 3 4 −1 } with n = 2020.

12
A6. Alice and Bob play a game in which they take turns choosing integers from 1 to
n. Before any integers are chosen, Bob selects a goal of “odd” or “even”. On the first
turn, Alice chooses one of the n integers. On the second turn, Bob chooses one of the
remaining integers. They continue alternately choosing one of the integers that has not yet
been chosen, until the nth turn, which is forced and ends the game. Bob wins if the parity of
{k : the number k was chosen on the kth turn} matches his goal. For which values of n does
Bob have a winning strategy?

Answer: Bob can always win by choosing the goal that matches the parity of n.

Solution 1: We say that k is a “fixed point” if k is chosen on the kth turn.

If n is even, then Bob can win by following a simple “mirror” strategy. Divide the available
numbers into adjacent pairs (1, 2), (3, 4), . . . , (n − 1, n). Whenever Alice chooses a number
from some pair, Bob chooses the other number from the pair on his turn. If on turn 2j + 1
Alice chooses 2j + 1, then Bob also creates a fixed point on turn 2j + 2, thereby adding two
to the total number of fixed points. Otherwise, Alice and Bob add zero fixed points on turns
2j + 1 and 2j + 2. There are therefore an even number of fixed points after each of Bob’s
turns, and since the game ends after the nth turn, which is Bob’s, he wins the game.
Now suppose that n = 2m + 1 is odd. For the remainder of the proof, denote Alice’s
choices by A1 , A3 , . . . , A2m+1 , and Bob’s by B2 , . . . , B2m . In particular, on the first turn of
the game, Alice chooses A1 , followed by Bob choosing B2 , and so on, and all of the As and
Bs must be distinct integers from 1 to 2m + 1. Let Fk be the number of fixed points after k
turns, reduced modulo 2. We have F0 = 0, and Bob wins if Fn = 1.
If n = 1, then Bob clearly wins. Otherwise, for m ≥ 1 we claim that Bob wins by playing
according to the following rules. Throughout 2j will denote Bob’s current turn in the game,
beginning by applying Rule R1 on turn 2, followed by whichever of Rule R2 or R3 applies
on turn 2j for j ≥ 2. The rule statements include several assumed properties that will be
justified inductively later, most importantly that F2j−2 = 1 for j ≥ 2.

(R1) (a) If A1 is in {1, 2}, then Bob chooses the other integer in this pair as B2 . This
results in either 0 or 2 fixed points, so F2 = 0. Now rename the remaining integers
3, . . . , n to 1, . . . , n − 2, and inductively restart the game.
(b) If A1 = a ≥ 3, then Bob chooses B2 = 2, so that F2 = 1. If n = 3, then a = 3 and
the game ends with the forced value A3 = 1, so F3 = 1. Otherwise, proceed to the
following rules.

(R2) If j ≥ 2 and A2j−1 = 2j − 1, then (we will show later that 2j ≤ a in this case):

(a) If 2j < a, Bob chooses B2j = 2j. Then F2j = F2j−2 + 2 mod 2 = 1.
(b) If 2j = a, then Bob chooses B2j = 1. Then F2j = F2j−2 + 1 mod 2 = 0. Rename
the remaining integers a+1, . . . , n to 1, . . . , n−a and inductively restart the game.

(R3) If j ≥ 2 and A2j−1 ̸= 2j − 1, then Bob chooses B2j = 2j + 1 if it has not been previously
chosen; otherwise, Bob chooses B2j to be an arbitrary unchosen integer not equal to
2j. Then F2j = F2j−2 = 1.

Since Rules R1b, R2a, and R3 all end with F2j = 1, and the other rules result in a restart
that resets j to 1, the assumption that F2j−2 = 1 for j ≥ 2 is true by induction. Bob’s

13
last turn occurs when j = m. If this turn results in a restart (from Rule R1a or R2b),
then F2m = 0, and Alice is forced to choose a fixed point on her final turn, so Fn = 1 and
Bob wins. Otherwise, Rule R1b or R2a or R3 applies on Bob’s last turn, and in each case
F2m = 1. Rule R1b explains why Fn = 1 in that case. If Rule R2a applies on turn 2m, then
a = n = 2m + 1, so Alice cannot choose An = n, and Fn = 1. If Rule R3 applies on turn
2m, then either Bob chooses B2m = 2m + 1, or 2m + 1 has already been chosen; again, Alice
cannot choose An = n, and Fn = 1.
To complete the proof, we need to verify that the strategy above respects the rules of
the game, that the case 2j > a never occurs when Rule R2 applies, and that the inductive
restart in Rule R2b is valid. In the arguments below, we assume that Bob has been able to
apply the rules on all turns before the turn in question.

Claim. If Rule R3 applies, then Bob is able to follow its instructions. Further, Rule R3 will
apply on all of Bob’s remaining turns.

Proof. If 2j + 1 has already been chosen, then since there are n − (2j − 1) = 2m − 2j + 2 ≥ 2
remaining integers, there is at least one choice for B2j other than 2j. After Bob’s turn, Alice
cannot choose A2j+1 = 2j + 1. Thus, Rule R3 applies on turn 2j + 2, and by induction it
applies on all of Bob’s subsequent turns.

It follows that if Rule R2 applies, then Rule R3 could not have previously been applied.
Thus, after the most recent restart (if any), Rule R1b must have been applied on turn 2, and
Rule R2a must have been applied on turn 2i for 2 ≤ i < j.

Claim. If Rule R2 applies, then 2j ≤ a. Further, if Rule R2a applies, then Bob is able to
choose B2j = 2j. If Rule R2b applies, then a is even, and all integers from 1 to a are chosen
on turns 1 to a.

Proof. After Rule R1b was applied, 2 and a had been chosen. Since Rule R2 has applied
ever since, Alice has chosen all of the odd numbers from 3 to 2j − 1, and since Rule R2a
was applied on the previous turns, Bob has chosen all of the even numbers from 4 to 2j − 2.
Since a is different from the other chosen numbers, and a ̸= 1, we must have a ≥ 2j. If Rule
R2a applies, 2j ̸= a, so Bob can choose B2j = 2j. If Rule R2b applies, then Alice chose
A1 = a = 2j on her first turn, and Bob chooses B2j = 1. Then all of the numbers from 1 to
a (and only those numbers) have been chosen.

Solution 2: Let S = {k : the number k was chosen on the k th turn}; at the beginning of
the game, S is empty, and each turn either adds an element to S or keeps S the same. Call
a number “available” if it hasn’t been chosen yet.
If n is even and Bob chooses the “even” goal, then Bob can win by following the rules
below on the kth turn (where k is even).

(1) If all numbers greater than k are available, then Bob chooses the one remaining available
number in {1, . . . , k}. This rule always applies when k = n.

(2) If k ≤ n − 2 and Rule 1 doesn’t apply, then Bob chooses k + 1 if available; if not, he
chooses an available number greater than k + 1 if possible; otherwise, Bob can choose
any remaining value other than k (since k < n, there is more than one available number
for Bob to choose).

14
Claim. If k is even, then after k turns, S has an even number of elements, and either 0 or
at least 2 of the numbers greater than k have been chosen.

The proof is by induction on even values of k from 0 to n, the base case k = 0 being the
beginning of the game with 0 elements in S and 0 numbers chosen. Assume now that the
claim hold for some even k < n.
If k = 0 or if Bob applies Rule 1 on the kth turn, then all numbers from 1 to k are chosen
before Alice makes the (k + 1)st turn. If Alice chooses k + 1 (adding an element to S) or
k + 2, then Bob applies Rule 1 on the (k + 2)nd turn to choose k + 2 (adding an element to
S) or k + 1, keeping an even number of elements in S and leaving 0 chosen numbers greater
than k + 2. If, on the other hand, Alice chooses a number greater than k + 2, then exactly 1
number greater than k + 2 has been chosen before Bob makes the (k + 2)nd turn. Bob must
then apply Rule 2, and since k + 2 ≤ n − 2 in that case, there is still an available number
greater than k + 2 for him to choose. Thus, Alice’s and Bob’s turns add no elements to S,
and result in 2 chosen numbers greater than k + 2.
If k > 0 and Bob applies Rule 2 on the kth turn, then either Bob chooses k + 1, or k + 1
was previously chosen; either way, Alice can’t add an element to S on the (k + 1)st turn.
Since Rule 1 didn’t apply on the kth turn, by the inductive hypothesis there were already
at least 2 chosen numbers greater than k, and thus at least 1 chosen number greater than
k + 1. If before Bob makes the (k + 2)nd turn, the only chosen number greater than k + 1 is
k + 2, then Bob applies Rule 1 and chooses a number strictly less than k + 2; this adds no
new element to S, and leaves 0 chosen numbers greater than k. Otherwise, there is at least
1 chosen number greater than k + 2 before Bob makes the (k + 2)nd turn, so Bob applies
Rule 2, which never adds an element to S. Either he chooses a number greater than k + 2,
making at least 2 such numbers chosen, or else all of the numbers (and in particular, at least
2 numbers) greater than k + 2 were already chosen before Bob’s turn.
This completes the induction that proves the claim. Applying the claim with k = n shows
that Bob wins when n is even.
If n is odd and Bob chooses the “odd” goal, then Bob can win by following the rules
below on the kth turn; since k is even, k < n.

(1) If S has an odd number of elements, then Bob chooses k + 1 if available; otherwise,
Bob can choose any number other than k (since k < n, Bob must have an option other
than k).

(2) If S has an even number of elements, then Bob chooses k if available; otherwise, Bob
chooses a number less than k (since only k − 1 numbers are chosen before Bob’s turn,
there must be an available number less than or equal to k).

If Bob applies Rule 1, then he doesn’t add an element to S, and Alice can’t add an element
to S on the (k + 1)st turn, so the number of elements in S remains odd after the (k + 1)st
turn. By induction, Bob applies Rule 1 on all future turns, and the number of elements in S
remains the same for the rest of the game, so Bob wins.
Assume hereafter that Bob is never able to apply Rule 1 for the entire game.

Claim. If k is even, then after k turns, either S has an even number of elements and no
numbers greater than k have been chosen, or S has an odd number of elements and exactly
one number greater than k has been chosen.

15
 In the base case k = 0, there are 0 elements in S and no elements at all have been chosen.
Proceeding by induction, assume that the claim holds for some even k < n − 1.
If after k turns S has an even number of elements, then by the inductive hypothesis all
numbers less than or equal to k have been chosen so far, and only those numbers. To prevent
Bob from applying Rule 1, Alice must not choose k + 1 on the (k + 1)st turn. If Alice chooses
k + 2, Bob chooses k + 1 on the (k + 2)nd turn; then S still has an even number of elements
and no numbers greater than k + 2 have been chosen. If Alice chooses a number greater than
k + 2, Bob chooses k + 2, adding an element to S; then S has an odd number of elements,
and exactly one number greater than k + 2 has been chosen.
If after k turns S has an odd number of elements, then Alice must choose k + 1 on the
(k +1)st turn, adding an element to S to prevent Bob from applying Rule 1. By the inductive
hypothesis, exactly one other number greater than k was chosen before Alice’s turn, and we
now know that number couldn’t have been k + 1. If that number is k + 2, then all but one of
the numbers from 1 to k + 2 have been chosen prior to the (k + 2)nd turn, and Bob chooses
the remaining avalaible number in that range; then S has an even number of elements, and
no numbers greater than k + 2 have been chosen. If, on the other hand, a number greater
than k + 2 was previously chosen, Bob chooses k + 2, adding another element to S; then S
has an odd number of elements, and exactly on number greater than k + 2 has been chosen.
This completes the induction that proves the claim. Applying the claim with k = n − 1,
either S has an even number of elements and Alice is forced to choose n on the nth turn,
adding an element to S, or S has an odd number of elements and n is already chosen, so
Alice is unable to add an element to S on the nth turn; either way, Bob wins.

16