Maths Olympiad Prep

Library / /4 of 7

Algebra Difficulty 8.8 Shortlist Prove it China

Let aa, bb, and dd be integers satisfying d0d \ge 0, a2|a| \ge 2, and b(a+1)d+1b \ge (|a| + 1)^{d+1}. Suppose f(x)f(x) is a real-coefficient polynomial of degree dd, and for each positive integer nn, let rnr_n denote the remainder of f(n)an\lfloor f(n)a^n \rfloor modulo bb.

Proof: If the sequence rnr_n is ultimately periodic, then f(x)f(x) is a rational-coefficient polynomial.

Note 1: For every real number xx, x\lfloor x \rfloor denotes the largest integer that is not greater than xx.

Note 2: A sequence ana_n is called ultimately periodic if there exist positive integers n0n_0 and TT such that for every integer nn0n \ge n_0, we have an+T=ana_{n+T} = a_n.

Solution

Proof. Lemma Let integers aa and real number zz satisfy a2|a| \ge 2. For any non-negative integer nn, if
anz<1a+1, ||a^n z|| < \frac{1}{|a| + 1},
then zz is an integer. Here, x||x|| is defined as the distance between the real number xx and the nearest integer, that is, x=min({x},1{x})||x|| = \min(\{x\}, 1 - \{x\}).

Proof of the lemma: Without loss of generality, assume a2a \ge 2 (otherwise, replace aa with a-a). Suppose the conclusion is false. Without loss of generality, assume 0<z120 < z \le \frac{1}{2} (otherwise, replace zz with mzm \mp z, where mm is some suitable integer).
Since z(0,1a1(a+1))z \in (0, \frac{1}{a^{-1}(a+1)}), we can assume that
1ak(a+1)z<1ak1(a+1)(kZ0), \frac{1}{a^k(a+1)} \le z < \frac{1}{a^{k-1}(a+1)} \quad (k \in \mathbb{Z}_{\ge 0}),
which implies 1a+1akz<11a+1\frac{1}{a+1} \le a^k z < 1 - \frac{1}{a+1}. Consequently, akz1a+1||a^k z|| \ge \frac{1}{a+1}, leading to a contradiction. Thus, the lemma is proven.

Back to the original question. Assume, for the sake of contradiction, that f(x)=adxd+ad1xd1++a1x+a0f(x) = a_d x^d + a_{d-1} x^{d-1} + \cdots + a_1 x + a_0 is not a polynomial with rational coefficients. Let tt be the largest index such that atQa_t \notin \mathbb{Q}. Define
{f(n)anb}=rn+εnb,εn[0,1). \left\{ \frac{f(n)a^n}{b} \right\} = \frac{r_n + \varepsilon_n}{b}, \quad \varepsilon_n \in [0, 1).
Let g(x)=Δtf(x)=i=0t(1)i(ti)f(x+ti)g(x) = \Delta^t f(x) = \sum_{i=0}^{t} (-1)^i \binom{t}{i} f(x + t - i), where Δ\Delta denotes the finite difference operator. It is easy to see that except for the constant term, all coefficients of g(x)g(x) are rational numbers. Let g(x)=cdtxdt++c1x+c0g(x) = c_{d-t}x^{d-t} + \cdots + c_1x + c_0, and let KK be the least common multiple of the denominators of c1,c2,,cdtc_1, c_2, \cdots, c_{d-t}. Now,
{g(n)an+tb}={i=0t(1)i(ti)f(n+ti)an+tiaib}={i=0t(1)i(ti)ai(rn+ti+εn+ti)b}={i=0trn+ti(1)i(ti)aib+i=0tεn+ti(1)i(ti)b}. \left\{ \frac{g(n)a^{n+t}}{b} \right\} = \left\{ \sum_{i=0}^{t} \frac{(-1)^i \binom{t}{i} f(n+t-i) a^{n+t-i} \cdot a^i}{b} \right\} \\ = \left\{ \sum_{i=0}^{t} \frac{(-1)^i \binom{t}{i} a^i (r_{n+t-i} + \varepsilon_{n+t-i})}{b} \right\} \\ = \left\{ \sum_{i=0}^{t} r_{n+t-i} \frac{(-1)^i \binom{t}{i} a^i}{b} + \sum_{i=0}^{t} \varepsilon_{n+t-i} \frac{(-1)^i \binom{t}{i}}{b} \right\}.

Let δn=i=0tεn+ti(1)i(ti)aib\delta_n = \sum_{i=0}^{t} \varepsilon_{n+t-i} \cdot \frac{(-1)^i \binom{t}{i} a^i}{b}. Note that
i=0t(1)i(ti)aib=i=0t(ti)aib=(a+1)tb(a+1)db1a+1, \sum_{i=0}^{t} \left| \frac{(-1)^i \binom{t}{i} a^i}{b} \right| = \sum_{i=0}^{t} \frac{\binom{t}{i} |a|^i}{b} = \frac{(|a|+1)^t}{b} \le \frac{(|a|+1)^d}{b} \le \frac{1}{|a|+1},
which implies that for any positive integers nn and nn',
δnδni=0tεn+tiεn+ti(ti)aib<i=0t(ti)aib1a+1. \left| \delta_n - \delta_{n'} \right| \le \sum_{i=0}^{t} \left| \varepsilon_{n+t-i} - \varepsilon_{n'+t-i} \right| \cdot \frac{\binom{t}{i} |a|^i}{b} < \sum_{i=0}^{t} \frac{\binom{t}{i} |a|^i}{b} \le \frac{1}{|a|+1}.
We choose positive integers MM and nn such that MM is a multiple of the smallest positive period of {rn}\{r_n\} and bKMbK|M, and when nn0n \ge n_0, we have bKan+M+tan+tbK|a^{n+M+t} - a^{n+t}. Then, for nn0n \ge n_0,
g(n+M)an+M+tbg(n)an+tb=δn+Mδn<1a+1, \left\| \frac{g(n+M)a^{n+M+t}}{b} - \frac{g(n)a^{n+t}}{b} \right\| = \left\| \delta_{n+M} - \delta_n \right\| < \frac{1}{|a|+1},
where the first equality holds because rn+M+ti=rn+tir_{n+M+t-i} = r_{n+t-i} for i=0,1,,ti = 0, 1, \dots, t. On the other hand,
g(n+M)an+M+tbg(n)an+tb=c0b(an+M+tan+t)+i=1dtcib((n+M)ian+M+tnian+t)=c0(aM1)an0+tbann0. \begin{aligned} & \left| \frac{g(n+M)a^{n+M+t}}{b} - \frac{g(n)a^{n+t}}{b} \right| \\ &= \left| \frac{c_0}{b} (a^{n+M+t} - a^{n+t}) + \sum_{i=1}^{d-t} \frac{c_i}{b} ((n+M)^i a^{n+M+t} - n^i a^{n+t}) \right| \\ &= \left| \frac{c_0(a^M - 1)a^{n_0+t}}{b} \cdot a^{n-n_0} \right|. \end{aligned}
According to the lemma, we have c0(aM1)an0+tbZ\frac{c_0(a^M - 1)a^{n_0+t}}{b} \in \mathbb{Z}, but this contradicts the irrationality of c0c_0! Hence, the proof is complete. \square

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement and solution reproduced as published; topic and difficulty added by this site.