Maths Olympiad Prep

Track / Stage 8 / 101 of 180 #1801 of 1964

Problem 1801

IMO Shortlist mid-range; USAMO P2/P5
Number theory Difficulty 8.2 Prove it

Let kk be a positive integer. Show that if there exists a sequence a0,a1,a_0,a_1,\ldots of integers satisfying the condition an=an1+nkn for all n1,a_n=\frac{a_{n-1}+n^k}{n}\text{ for all } n\geq 1, then k2k-2 is divisible by 33.

Proposed by Okan Tekman, Turkey

This one wants a proof. Work it on paper, then read the official solution and mark yourself. Be honest about it: the record is only any use to you if it is.

Official solution

1. Lemma 1: There exists a polynomial P(x)Z[x] P(x) \in \mathbb{Z}[x] with degree k1 k-1 such that
xP(x)P(x1)=xk+c xP(x) - P(x-1) = x^k + c
for some constant cZ c \in \mathbb{Z} .

Proof: Write P(x)=ak1xk1+ak2xk2++a1x+a0 P(x) = a_{k-1}x^{k-1} + a_{k-2}x^{k-2} + \cdots + a_1x + a_0 . Now, solve for the ai a_i 's in descending order, starting with ak1 a_{k-1} and working downwards. By comparing the xk x^k -coefficients on both sides of the equation, we clearly have ak1=1 a_{k-1} = 1 . If ai+1,,ak1 a_{i+1}, \cdots, a_{k-1} have been determined for some i0 i \ge 0 , then ai a_i is uniquely determined by examining the xi+1 x^{i+1} -coefficients on both sides of the equation. Furthermore, we can see that ai a_i is integral. \blacksquare

2. As guaranteed by Lemma 1, let P(x) P(x) be a monic polynomial of degree k1 k-1 with integer coefficients and cZ c \in \mathbb{Z} be an integer such that
xP(x)P(x1)=xk+c. xP(x) - P(x-1) = x^k + c.

3. Lemma 2: c=0 c = 0 and an=P(n) a_n = P(n) for all nZ0 n \in \mathbb{Z}_{\ge 0} .

Proof: Let bn=P(n)an b_n = P(n) - a_n for all nZ0 n \in \mathbb{Z}_{\ge 0} . Note that from the condition of the problem:
nanan1=nk,nN. na_n - a_{n-1} = n^k, \quad \forall n \in \mathbb{N}.
From the equation xP(x)P(x1)=xk+c xP(x) - P(x-1) = x^k + c , we have:
nP(n)P(n1)=nk+c,nN. nP(n) - P(n-1) = n^k + c, \quad \forall n \in \mathbb{N}.
Therefore, subtracting the two equations gives:
nbnbn1=c,nN. nb_n - b_{n-1} = c, \quad \forall n \in \mathbb{N}.
Let M=max(b0,c) M = \max(|b_0|, |c|) . We claim now that this implies bi[2M,2M] b_i \in [-2M, 2M] for all iZ0 i \in \mathbb{Z}_{\ge 0} . For i=0,1 i = 0, 1 , this is immediate from the definition of M M . For i>1 i > 1 , we can prove it easily by induction on i i . Hence, we've shown that the sequence {bi} \{b_i\} is bounded by 2M 2M in absolute value.

4. Now, notice that for n>3M n > 3M , we have:
bn=bn1+cnbn1n+cn<2M3M+M3M=1, |b_n| = \left|\frac{b_{n-1} + c}{n}\right| \le \left|\frac{b_{n-1}}{n}\right| + \left|\frac{c}{n}\right| < \frac{2M}{3M} + \frac{M}{3M} = 1,
and hence bn=0 b_n = 0 . Therefore, c=(3M+2)b3M+2b3M+1=0 c = (3M+2)b_{3M+2} - b_{3M+1} = 0 , as claimed. From here, it easily follows from downwards induction (with base case n=3M+1 n = 3M+1 ) that bn=0 b_n = 0 for all nZ0 n \in \mathbb{Z}_{\ge 0} . This implies that P(n)an=0an=P(n) P(n) - a_n = 0 \Rightarrow a_n = P(n) for all nZ0 n \in \mathbb{Z}_{\ge 0} . \blacksquare

5. From the lemma, as c=0 c = 0 , we know that:
xP(x)P(x1)=xk. xP(x) - P(x-1) = x^k.

6. Let's take this equation modulo x4+x2+1 x^4 + x^2 + 1 . It's clear that the remainder of P P modulo x4+x2+1 x^4 + x^2 + 1 is of the form R(x)=ax3+bx2+cx+d R(x) = ax^3 + bx^2 + cx + d for some a,b,c,dZ a, b, c, d \in \mathbb{Z} . Now, observe that P(x)R(x)=(x2+x+1)(x2x+1)Q(x) P(x) - R(x) = (x^2 + x + 1)(x^2 - x + 1)Q(x) for some QZ[x] Q \in \mathbb{Z}[x] . In particular, observe that x2x+1P(x)R(x) x^2 - x + 1 \mid P(x) - R(x) and x2x+1P(x1)R(x1) x^2 - x + 1 \mid P(x-1) - R(x-1) , since (x1)2+(x1)+1=x2x+1 (x-1)^2 + (x-1) + 1 = x^2 - x + 1 . Therefore, viewing the equation in modulo x2x+1 x^2 - x + 1 , we obtain:
xR(x)R(x1)xk. xR(x) - R(x-1) \equiv x^k.

7. Plugging the formula for R R into the equation yields:
ax4+bx3+cx2+dxa(x1)3b(x1)2c(x1)dxk. ax^4 + bx^3 + cx^2 + dx - a(x-1)^3 - b(x-1)^2 - c(x-1) - d \equiv x^k.
Noting that for iZ2 i \in \mathbb{Z}_{\ge 2} , we have xixi1xi2 x^i \equiv x^{i-1} - x^{i-2} and (x1)i(x1)i1(x1)i2 (x-1)^i \equiv -(x-1)^{i-1} - (x-1)^{i-2} , we have that:
(c+da)x(b+c)+(bc)x+(cad)xk, (c+d-a)x - (b+c) + (b-c)x + (c-a-d) \equiv x^k,
i.e.
(b+da)x(a+b+d)xk. (b+d-a)x - (a+b+d) \equiv x^k.

8. Observe that:
10x+1, 1 \equiv 0 \cdot x + 1,
x1x+0, x \equiv 1 \cdot x + 0,
x31, x^3 \equiv -1,
and
x4(1)x+0. x^4 \equiv (-1) \cdot x + 0.
So since b+da,a+b+d b+d-a, a+b+d are either both 0 or both nonzero (since a,b,dZ a, b, d \in \mathbb{Z} ), we cannot have k{0,1,3,4} k \in \{0, 1, 3, 4\} . Furthermore, as xkxk+6 x^k \equiv x^{k+6} for kZ0 k \in \mathbb{Z}_{\ge 0} , we actually have that k k cannot be 0,1,3, 0, 1, 3, or 4 4 (mod 6 6 ). In other words, k2 k \equiv 2 (mod 3 3 ), as desired. \blacksquare

The final answer is k2 \boxed{ k-2 } is divisible by 3.

Source: NuminaMath-1.5, licensed Apache-2.0. Statement and solution reproduced as published; topic, difficulty and ordering added by this site.