Nonnegative multiplicative processes reaching stationarity in finite time (Q1822138)

This paper follows the work of the author for Markov chains [Stochastic Processes Appl. 11, 91-99 (1981; Zbl 0457.60041)] and \textit{U. G. Rothblum} and \textit{C. Tan} [Linear Algebra Appl. 55, 155-167 (1983; Zbl 0534.60063)] for general (n\(\times n)\) non-negative matrices P. A non- negative vector \(\pi \in {\mathbb{R}}^ n\) satisfying \(\pi P=\rho \pi\) for some number \(\rho >0\) is called a stationary vector for P. An \(x\in {\mathbb{R}}^ n\) is accessible to \(\pi\) in k steps (is a ''\(\pi\) /k-vector'') if \(xP^{k-1}\neq \rho^{k-1}\pi\), \(xP^ k=\rho^ k\pi\) for some \(k\geq 1.\) Rothblum and Tan showed that it is possible to determine (in terms of P) a smallest set \(S\subseteq \{1,2,...,n\}\) (depending on \(\pi)\) such that a non-negative vector x is a \(\pi\) /k-vector for some \(k\geq 1\) if and only if S(x)\(\subseteq S\) where \(S(x)=\{i:\) \(i=1,...,n\), \(x_ i>0\}\). They also showed that a \(\pi\) /k-vector (not necessarily non-negative) exists if and only if \(1\leq k\leq m(P)\), where m(P) is the index of P at eigenvalue 0. In this paper the author proves the corresponding result for non-negative input vectors x: that a non-negative \(\pi\) /k-vector for P exists if and only if \(1\leq k\leq m(P_ S)\) where \(P_ S\) is the restriction of P to the set S. Other earlier results are also generalized. The continuing use of the words ''multiplicative processes'' in the title is undesirable because of confusion with branching stochastic processes.