Maths Olympiad Prep

Library / /7 of 16

Number theory Difficulty 6.9 National olympiad Prove it Japan

Determine all polynomials f(x)f(x) with integer coefficients such that, for any integer n2n \ge 2, the following conditions hold:
 f(n)>0.\bullet\ f(n) > 0.
 f(n)\bullet\ f(n) divides nf(n)1n^{f(n)} - 1.

Solution

Solution: We will show that f(x)=(x1)mf(x) = (x-1)^m provides an answer. For n2n \ge 2 we have f(n)=(n1)m>0f(n) = (n-1)^m > 0, thus the first condition is satisfied.

Lemma 1. Let d,t,xd, t, x be positive integers. If x1x-1 is a multiple of dtd^t, then xd1x^d-1 is a multiple of dt+1d^{t+1}.

Proof. Since x1(modd)x \equiv 1 \pmod d, we have xd1++x+11++10(modd)x^{d-1} + \dots + x + 1 \equiv 1 + \dots + 1 \equiv 0 \pmod d. Thus xd1=(x1)(xd1++x+1)x^d - 1 = (x-1)(x^{d-1} + \dots + x + 1) is a multiple of dt+1d^{t+1} as desired. \blacksquare

Let n2n \ge 2. Since n(n1)01=n1n^{(n-1)0} - 1 = n-1 is a multiple of n1n-1, an induction using Lemma 1 shows that n(n1)m1n^{(n-1)m} - 1 is a multiple of (n1)m+1(n-1)^{m+1} for m0m \ge 0. Thus, we conclude that f(x)=(x1)mf(x) = (x-1)^m is an answer for m0m \ge 0.

Next we will show that no other answers exist. For integers a,ba, b with b0b \ne 0, we write bab \mid a to mean that aa is divisible by bb.

Lemma 2. For a positive integer nn and integers aa and bb, ab(modn)a \equiv b \pmod n implies f(a)f(b)(modn)f(a) \equiv f(b) \pmod n.

Proof. Write f(x)=a0+a1x++adxdf(x) = a_0 + a_1x + \dots + a_d x^d, where dd is a nonnegative integer and a0,a1,,ada_0, a_1, \dots, a_d are integers. Since nabn \mid a-b,
f(a)f(b)=i=1dai(aibi)=(ab)i=1dai(ai1+ai2b++abi2+bi1) f(a) - f(b) = \sum_{i=1}^{d} a_i (a^i - b^i) = (a-b) \sum_{i=1}^{d} a_i (a^{i-1} + a^{i-2}b + \dots + ab^{i-2} + b^{i-1})
is divisible by nn. \blacksquare

We will show that, for an odd prime pp and an integer n2n \ge 2, pf(n)p \mid f(n) implies pn1p \mid n-1. For a nonnegative integer kk, Lemma 2 shows that f(n+kp)f(n)0(modp)f(n + kp) \equiv f(n) \equiv 0 \pmod p. Thus we have
0(n+kp)f(n+kp)1nf(n+kp)1(modp) 0 \equiv (n + kp)^{f(n+kp)} - 1 \equiv n^{f(n+kp)} - 1 \pmod p
Since pp and p1p-1 are coprime, we can take a nonnegative integer kk such that p1n+kpp-1 \mid n + kp.
Then f(n+kp)f(p1)(modp1)f(n + kp) \equiv f(p-1) \pmod{p-1} holds and thus Fermat's little theorem shows that
0nf(n+kp)1nf(p1)1,(modp) 0 \equiv n^{f(n+kp)} - 1 \equiv n^{f(p-1)} - 1, \pmod{p}
therefore we have
nf(p1)1.(modp) n^{f(p-1)} \equiv 1. \pmod{p}
Since f(p1)(p1)f(p1)1f(p-1) \mid (p-1)^{f(p-1)} - 1, f(p1)f(p-1) and p1p-1 are coprime. Thus, there exists a positive integer rr such that f(p1)r1(modp1)f(p-1)r \equiv 1 \pmod{p-1}. By Fermat's little theorem we have
nnf(p1)r1r1,(modp) n \equiv n^{f(p-1)r} \equiv 1^r \equiv 1, \pmod{p}
that is, pn1p \mid n-1.

We have shown that, for an odd prime pp and an integer n2n \ge 2, pf(n)p \mid f(n) implies pn1p \mid n-1. In particular, for an odd prime pp, any prime factor of f(p+1)f(p+1) is also a prime factor of p+11=pp+1-1 = p, which means that f(p+1)f(p+1) is a power of pp.

Write f(x+1)=xm(b0+b1x++blxl)f(x+1) = x^m(b_0 + b_1x + \dots + b_lx^l) where ll and mm are nonnegative integers and b00,b1,,blb_0 \ne 0, b_1, \dots, b_l are integers. Let pp be an odd prime greater than b0|b_0|. Then f(p+1)f(p+1) is a multiple of pmp^m, while f(p+1)b0pm0(modpm+1)f(p+1) \equiv b_0p^m \ne 0 \pmod{p^{m+1}}. The argument above shows that f(p+1)=pmf(p+1) = p^m.

Let g(x)=f(x)(x1)mg(x) = f(x) - (x-1)^m. Then there exist infinitely many integers nn such that g(n)=0g(n) = 0. It follows that g(x)=0g(x) = 0 as polynomials. We conclude that f(x)=(x1)mf(x) = (x-1)^m.

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.