No label defined (Q3973696)

[For the entire collection see Zbl 0705.00029.] Let \({\mathcal M}=(\mu_ \theta)_{\theta\in\Theta}\) be a family of probability measures on \(([0,1],{\mathcal B}([0,1]))\). The author calls \({\mathcal M}\) to be ``separable'' iff there exists a partition of \([0,1]\), of measurable sets \(E_ \theta\), \(\theta\in\Theta\), such that \(\mu_ \theta(E_{\theta'})=\delta_{\theta\theta'}=1\) if \(\theta=\theta'\), \(=0\) if \(\theta\neq\theta'\). If \({\mathcal M}\) satisfies only the second condition \(\mu_ \theta(E_{\theta'})=\delta_{\theta\theta'}\), then it is said to be only weakly separable. Consider now \((\xi_ n)_ n\) to be an \(r\)-order homogeneous stationary Markov chain which is valued on some finite set \(L=\{0,1,\dots,k-1\}\). Let \(\theta=(\theta_{i_ 1,i_ 2,\dots,i_ r;j})_{0\leq i_ 1,\dots,i_ r,j\leq k-1}\) be its transition kernel and \(\Theta\) be the set of all the transition kernels. Let \(\xi=\sum_{n=1}^ \infty \xi_ n/k^ n\). Then \(\xi\) belongs to [0,1]. Let \(\mu_ \theta\) be its distribution. The author proves that the family \((\mu_ \theta)_{\theta\in\Theta}\) is separable. The interesting fact is not that for \(\theta\neq\theta'\) the measures \(\mu_ \theta\) and \(\mu_{\theta'}\) are orthogonal (we already know that statistics on \(r\)- order Markov chains is possible) but that one can choose the supports of the probabilities \(\mu_ \theta\) to be mutually disjoint.