On a representation of measurable automaton transformations by stochastic automata (Q1241431)

From MaRDI portal
scientific article
Language Label Description Also known as
English
On a representation of measurable automaton transformations by stochastic automata
scientific article

    Statements

    On a representation of measurable automaton transformations by stochastic automata (English)
    0 references
    1979
    0 references
    Let \((X,\mathcal A)\) be a measurable, \((Y,\mathcal B)\) be a Borel space, and denote by \((X^*,\mathcal A^*)\) the direct sum of \(((X^n,\mathcal A^n))_{n\geq 0}\). If \(K\) is a transition probability from \((X^*,\mathcal A^*)\) to \((Y^*,\mathcal B^*)\), \(K\) is said to be a stochastic transformation iff for every \(v\in X^*\), \(K(v)(Y^{| v|}) = 1\), and if for every \(x\in X\), \(B\in\mathcal B^*\), \(K(vx)(B\times Y) = K(v)(B)\) holds \((| v|\) denoting the length of \(v)\). By means of a disintegration theorem due to \textit{D. Rhenius} [Markoffsche Entscheidungsprozesse mit unvollständiger Information und Anwendungen in der Lerntheorie. Thesis, Hamburg (1971), see also Ann. Stat. 2, 1327--1334 (1974; Zbl 0294.49007)] it is shown that \(K\) is a stochastic transformation iff it is the behavior of an initial stochastic automaton. If \(Y\) is \(\sigma\)-compact and Polish, an automaton transformation \(R\) (i.e. the behavior of an initial complete nondeterministic automaton, cf. \textit{A. Schmitt} [Computing 4, 56--74 (1969; Zbl 0213.02203)]) is said to be measurable iff \(R\) is closed-valued, and if \(X^*\ni v\rightarrow R(v)\subset Y^*\) is (weakly) measurable [\textit{C. J.Himmelberg}, Fundam. Math, 87, 53--72 (1975; Zbl 0296.28003)]. The main result of this paper is that \(R\) is measurable iff there exists a stochastic transformation \(K\) such that \(\forall v\in X^*\) : \(R(v)=\text{supp}(K(v))\), \(\text{supp}\) denoting support. This is shown by means of measurable selections [\textit{D. H. Wagner}, SIAM J. Control Optim. 15, 859--903 (1977; Zbl 0407.28006)] for the weakly measurable relation that assigns to every \(v\) the set of all probabilities which have their support in \(R(v)\). This result implies that a closed-valued relation \(F\) from \(X\) to \(Y\) is measurable iff there exists a transition probability from \((X,\mathcal A)\) to \((Y,\mathcal B)\) such that \(F(x) = \text{supp}(K(x))\) holds for every \(x\in X.\)
    0 references
    0 references
    measurable automaton transformations
    0 references
    stochastic automata
    0 references