The relation between a generalized Fibonacci sequence and the length of Cunningham chains

TL;DR

This paper explores the connection between generalized Fibonacci sequences and Cunningham chain lengths, proposing a new relation that suggests an upper bound of the chain length proportional to logarithm of the prime, based on a divisor function.

Contribution

It introduces a novel relation linking generalized Fibonacci sequences and Cunningham chain lengths, providing a potential method to estimate upper bounds of chain lengths.

Findings

Relation between divisor function on Fibonacci-like sequences and Cunningham chain structure

Proposed upper bound of chain length proportional to logarithm of prime

Reduction of prime-based problem to natural number analysis

Abstract

Let $p$ be a prime number. A chain $\{p,2p+1,4p+3,\cdots,(p+1)2^{l(p)-1}-1\}$ is called the Cunningham chain generated by $p$ if all elements are prime number and $(p+1)2^{l(p)}-1$ is composite. Then $l(p)$ is called the length of the Cunningham chain. It is conjectured by Bateman and Horn in 1962 that the number of prime $p\leq N$ such that $l(p)\geq k$ is asymptotically equal to $B_k N/(\log N)^k$ with a real $B_k>0$ for all natural number $k$. This suggests that $l(p)=\Omega(\log p/\log\log p)$. However, so far no good estimation is known. It has not even been proven whether $\limsup_{p\to\infty} l(p)$ is infinite or not. All we know is that $l(p)=5$ if $p=2$ and $l(p)<p$ for odd $p$ by Fermat's little theorem. Let $\alpha\geq3$ be an integer. In this article, a generalized Fibonacci sequence $\mathcal{F}_\alpha=\{F_n\}_{n=0}^\infty$ is defined as $F_0=0,F_1=1, F_{n+2}=\alpha F_{n+1}+F_n (n\geq0)$, and ${}_{\mathcal{F}_\alpha}\sigma(n)=\sum_{d\mid n, 0<d\in\mathcal{F}_\alpha}d$ is called a divisor function on $\mathcal{F}_\alpha$. Then we obtain an interesting relation between the iteration of ${}_{\mathcal{F}_\alpha}\sigma$ and the length of Cunningham chains. For two primes $p$ and $q$, the fact $p=2q+1$ or $2q-1$ is equivalent to ${}_{\mathcal{F}_\alpha}\sigma({}_{\mathcal{F}_\alpha}\sigma(F_p))={}_{\mathcal{F}_\alpha}\sigma(F_q)$ for some $\alpha$. By this relation, we get $l(p)\ll\log p$ under a certain condition. It seems that this sufficient condition is plausible by numerical test. Furthermore, the condition, written in terms of prime numbers, can be replaced by the condition written in terms of natural numbers. This implies that the problem of upper estimation of $l(p)$ is reduced to that on natural numbers.