Given an integer sequence $a_{n+1}=a_{n}^{2}-a_{n}-1$, prove that $\forall n\in \mathbb{Z}$, $a_{n+1}$ and $2n+1$ are coprime

coprimesequences-and-series

This is a very interesting infinite integer sequence problem I came across. The crux of the matter lies in comparing the values and the indices – we are supposed to check whether $a_{n+1}$ and $2n+1$ are coprime as opposed to any two members of the sequence itself. I tried plugging in a few numbers. In the cases when $a_{1}$ is $-1$,$0$,$1$ or $2$ the sequence starts oscillating between $-1$ and $1$ and the proof becomes trivial. In general, however, the numbers obviously become large really fast. What kind of approach should be taken?

Thank you for the help

Best Answer

The very first observation is that if for some prime number $p$ and positive integer $n$ we have $p|a_n$ then $a_m$ is not divisible by $p$ for all $m>n$ (*) (the modulo will be $0,-1,1,-1,1,\dots$). We will return to this observation later in this post.

Let $P(x)=x^2-x-1$
For any odd prime number $p$, consider the sequence $b_n = (a_n \mod p)$ for all $n \ge 0$.
Note that for any number $a$, $P(a) \equiv P(1-a) (\mod p)$ hence the modulo of $P(a)$ mod $p$ has at most $\frac{p+1}{2}$ different values .
So by pigeonhole principle, there are two distinct numbers $1 \le m<n \le \frac{p+1}{2}+1$ such that $b_m=b_n$, i.e.
From the position $m$, the sequence $(b_k)$ repeats with the period $n-m$ (1).
We will show next there is not any $k$ that lies between $m,n$ such that $b_k=0$(2).
Indeed, if its true, that leads to $b_{k+m-n}=0$, which means $a_k$ and $a_{k+m-n}$ are divisible by $p$. This is a contradiction to our observation (*).
From (1) and (2), we conclude that $b_k \ne 0$ for all $k \ge m$. Besides $m \le \frac{p+1}{2}$, thus in general, $b_k \ne 0 $ with all $k \ge \frac{p+1}{2}$.
In other words, $a_k$ is not divisible by $p$ for all $k\ge \frac{p+1}{2}$. Hence the conclusion.

Related Question