1: \begin{abstract}

2: 

3: We consider a probability-measure $p^*$ on $\mathcal M_d(\mathbb C)^\mathbb N$, infinite product of a non-atomic Borel probability measure $p$ of support $\mathbb C$. For $(A_n)_{n\in\mathbb N}\in\mathcal M_d(\mathbb C)^\mathbb N$, let $P_n:=A_1\cdots A_n$, we prove that the sequence $n\mapsto P_n/{\Vert P_n\Vert}$ diverges $p^*$-a.e.. For this reason we are more interested by the convergence of the sequence of vectors $n\mapsto P_nV/{\Vert P_nV\Vert}$, where $V$ is a $d$-dimensional column-vector such that $\forall n, P_nV\ne0$; for instance, if $\forall n,\ A_n\in\mathcal M$, finite set of positive $d\times d$ matrices, and if no couple of matrices of $\mathcal M$ has a common left-eigenvector, then $n\mapsto P_nV/{\Vert P_nV\Vert}$ converges for any $(A_n)_{n\in\mathbb N}\in\mathcal M^{\mathbb N}$ and any nonnegative column-vector~$V$, while $n\mapsto P_n/{\Vert P_n\Vert}$ diverges except if $(A_n)_{n\in\mathbb N}$ belongs to the countable set $\mathcal M_0\subset\mathcal M^{\mathbb N}$ of the eventually constant sequences. The purpose of this paper is to give in Theorem A some sufficient conditions on the sequence of nonnegative matrices $(A_n)_{n\in\mathbb N}$ , for $n\mapsto P_nV/{\Vert P_nV\Vert}$ to converge, and to apply this theorem to the study of a sofic (i.e. linearly representable) measure $\mu$; we choose this measure for Theorem~A to be the only way to prove that $\mu$ satisfies the multifractal formalism.

4: \end{abstract}

5: