On the number of residues of linear recurrences

From MaRDI portal
Publication:2065753



Abstract: For every nonconstant monic polynomial ginmathbbZ[X], let mathfrakM(g) be the set of positive integers m for which there exist an integer linear recurrence (sn)ngeq0 having characteristic polynomial g and a positive integer M such that (sn)ngeq0 has exactly m distinct residues modulo M. Dubickas and Novikas proved that mathfrakM(X2−X−1)=mathbbN. We study mathfrakM(g) in the case in which g is divisible by a monic quadratic polynomial finmathbbZ[X] with roots such that and is not a root of unity. We show that this problem is related to the existence of special primitive divisors of certain Lehmer sequences, and we deduce some consequences on mathfrakM(g). In particular, for , we prove that minmathfrakM(g) for every integer mgeq7 with meq10 and 4midm.


An integer sequence \(\mathbf{s}=(s_{n})_{n\geq0}\) is\ called a linear recurrence relation if there are integers \(c_{1},...,c_{r}\in\mathbb{Z}\) such that for every integer \(n\geq r\), we have \[ s_{n}=c_{1}s_{n-1}+c_{2}s_{n-2}+...+c_{r}s_{n-r}. \] The \(r\) values \(s_{0},...,s_{r-1}\) are called the \textit{initial conditions} of the integer sequence \(\mathbf{s}\), and \[ g(X)=X^{r}-c_{1}X^{r-1}-c_{2}X^{r-2}-...-c_{r} \] is called the \textit{characteristic polynomial} of \(\mathbf{s}\). Let \(g\in\mathbb{Z}[X]\) be any nonconstant monic polynomial. Also, let \(\mathfrak{M}(g)\) be the set of positive integers \(m\) such that there exists an integer linear recurrence relation \((s_{n})_{n\geq0}\) with characteristic polynomial \(g\) and a positive integer \(M\) for which \((s_{n})_{n\geq0}\) has exactly \(m\) distinct residues modulo \(M\). \textit{A. Dubickas} and \textit{A. Novikas} [J. Number Theory 215, 120--137 (2020; Zbl 1473.11039)] have shown that \(\mathfrak{M}(X^{2}-X-1)=\mathbb{N}\). The author of the paper under review examines \(\mathfrak{M}(g)\) in the case in which \(g\) is divisible by a monic quadratic polynomial \(f\in\mathbb{Z}[X]\) with roots \(\alpha,\beta\) such that \(\alpha\beta=\pm1\) and \(\alpha/\beta\) is not a root of unity. Also, the author demonstrates that this problem is pertained to the existence of special primitive divisors of certain Lehmer sequences.











This page was built for publication: On the number of residues of linear recurrences

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2065753)