Automated Proof (or Disproof) of Linear Recurrences Satisfied by Pisot Sequences

Pisot sequences (sequences $a_n$ with initial terms $a_0=x, a_1=y$, and defined for $n>1$ by $a_n= \lfloor a_{n-1}^2/a_{n-2} + \frac{1}{2} \rfloor$) often satisfy linear recurrences with constant coefficients that are valid for all $n \geq 0$, but there are also cautionary examples where there is a linear recurrence that is valid for an initial range of values of $n$ but fails to be satisfied beyond that point, providing further illustrations of Richard Guy's celebrated "Strong Law of Small Numbers". In this paper we present a decision algorithm, fully implemented in an accompanying Maple program ({\tt Pisot.txt}), that first searches for a putative linear recurrence and then decides whether or not it holds for all values of $n$. We also explain why the failures happen (in some cases the `fake' linear recurrence may be valid for thousands of terms). We conclude by defining, and studying, higher-order analogs of Pisot sequences, and point out that similar phenomena occur there, albeit far less frequently. This article is dedicated to Richard K. Guy (b. Sept. 30, 1916) on his 100th birthday.