On non-singular renewal kernels with an application to a semigroup of transition kernels (Q1084768): Difference between revisions

Let \(\nu\) be a probability measure with finite mean \(\mu\) on the Borel subsets \({\mathcal R}\) of \({\mathbb{R}}_+\), let \(U=\sum^{\infty}_{n=0}\nu^{*n}\) denote the corresponding renewal measure. If a convolution power of \(\nu\) has a non-vanishing absolutely continuous component (\(\nu\) is spread out) then Stone's decomposition applies [\textit{Ch. Stone}, Ann. Math. Stat. 37, 271-275 (1966; Zbl 0147.162)]: \(U=U_ 0+U_ 1\), where \(U_ 0\) is a finite measure and \(U_ 1\) is absolutely continuous with density \(u_ 1\) satisfying \(u_ 1(t)\to 1/\mu\) as \(t\to \infty.\) In the present paper this is generalized to renewal kernels \(R=\sum^{\infty}_{n=0}K^{*n}\), where K(x,A) is a non-negative transition kernel from a general measurable space (E,\({\mathcal E})\) into \((E\times {\mathbb{R}}_+,{\mathcal E}\otimes {\mathcal R}):\) a spread-out condition allows minorization of some convolution power of K by a product measure on (E,\({\mathcal E})\otimes (E\times {\mathbb{R}}_+,{\mathcal E}\otimes {\mathcal R})\)- this separates x - which in turn yields an imbedded renewal measure to which Stone's result can be applied. A semigroup \(\{R_ t(x,dy): t\in {\mathbb{R}}_+\}\) of transition kernels on (E,\({\mathcal E})\) can be written as Lebesgue density of the renewal kernel associated with \[ K(x,dy\times dt)=e^{-t}R_ t(x,dy)dt, \] this construction is the basis for the eventual applications to continuous parameter transition semigroups.