Journées de Probabilités 2007

Élise Janvresse

Exponential growth of random Fibonacci sequences and generalizations

(Joint work with Thierry de la Rue and Benoît Rittaud)

We study two kinds of generalized random Fibonacci sequences defined by F1=a, F2=b and one of the following induction formulas:
linear case: for n>0, Fn+2 = Fn+1 Fn
non-linear case: n>0, Fn+2 = | Fn+1 Fn|
where is of the form for some k 3, and each sign is independent and either + with probability p or - with probability 1-p (0 < p 1). For k=3, =1 and we recover random Fibonacci sequences.
Our main result is that the exponential growth of Fn for 0 < p 1 in the linear case, and for 1/k p 1 in the non-linear case, is almost surely given by

where is an explicit function of p depending on the case we consider, taking values in [0, 1]k-1, and is an explicit probability distribution on R+ defined inductively on generalized Stern-Brocot intervals.
Our method relies on some reduction process and makes use of Rosen continued fraction expansion. In the linear case, it gives an explicit form for Furstenberg's invariant measure. It is easier in the non-linear case to which the standard theory does not apply.