Small sets and Markov transition densities. (Q1766078): Difference between revisions

Let \(K\) be a Markov kernel on a measurable state space \((S,\mathcal S)\). We say that \(K\) satisfies a minorization condition if an integer \(n\geq 1\), a non-vanishing function \(g\geq 0\) and a probability measure \(\mu \) may be found such that \[ K^ {n}(x,A)\geq g(x)\mu (A) \quad \text{for all }x\in S,\;A\in \mathcal S.\tag{1} \] A set \(C\subseteq S\) is called a small set of order \(n\) provided that (1) holds with \(g=\varrho \mathbf 1_ {C}\) for some \(\rho \in \mathopen ]0,1\mathclose [\). If \(S\) is a smooth manifold and the kernel \(K\) has a continuous density, then small sets of order 1 exist. However, it is proved that in general there are no small sets of order 1, even if \(K\) admits a measurable transition density. On the other hand, there always exist small sets of order 2 whenever the \(\sigma \)-algebra \(\mathcal S\) is countably generated and \(K\) has a measurable density. Moreover, the two-step transition density \(p^ {(2)}\) may be represented as \(p^ {(2)}(x,y) = \sum ^ \infty _ {i=0} \beta (x,i)r_ {i}(y)\), where \(r_ {i}\) are probability densities and \(\beta \) is a transition density from \(S\) to the set of positive integers \(\{1,2,\ldots \}\). Finally, some interesting refinements and applications of these results are discussed.