Random Fibonacci sequences and the number $1.13198824\dots$ (Q4955865)

The asymptotic behaviour (in the sense of almost sure convergence) of the following randomized Fibonacci sequence \((t_n)_{n\geq 1}\) is studied: Let \(t_1= t_2= 1\), and for \(n\geq 3\), \(t_n= \varepsilon_{n-1}t_{n-1}+ \varepsilon_{n-2}t_{n-2}\), where the random variables \((\varepsilon_n)_{n\geq 1}\) are independent taking either \(+1\) or \(-1\) with probability \(1/2\). It is well known [see e.g. \textit{P. Bougerol} and \textit{J. Lacroix}, Random products of matrices with applications to Schrödinger operators. Birkhäuser, Basel (1985; Zbl 0572.60001)], that \(\frac 1n\log|t_n|\to \gamma\) as \(n\to \infty\) for a constant \(\gamma>0\) with probability 1. The main result of this paper is the numerically exact determination of the constant \(\gamma\) by showing that \(e^\gamma= 1.13198824\ldots\). Finding this exact value involves the theory of random matrix products, Stern-Brocot division of the real line, a fractal measure, a computer calculation, and a rounding error analysis to validate the computer calculation.