Acta Applicandae Mathematica

, Volume 34, Issue 1, pp 175–188

Topological conditions enabling use of harris methods in discrete and continuous time

  • R. L. Tweedie
Part III: Regeneration

DOI: 10.1007/BF00994264

Cite this article as:
Tweedie, R.L. Acta Applicandae Mathematicae (1994) 34: 175. doi:10.1007/BF00994264
  • 72 Views

Abstract

This paper describes the role of continuous components in linking the topological and measuretheoretic (or regenerative) analysis of Markov chains and processes. Under Condition\(\mathcal{T}\) below we show the following parallel results for both discrete and continuous time models:
  1. (i)

    when the model is open set irreducible it is ϕ-irreducible;

     
  2. (ii)

    under (i), the measure-theoretic classification of the model as Harris recurrent or positive Harris recurrent is equivalent to a topological classification in terms of not leaving compact sets or of tightness of transition kernels;

     
  3. (iii)

    under (i), the ‘global’ classification of the model as transient, recurrent or positive recurrent is given by a “local’ classification of any individual reachable point;

     
  4. (iv)

    under (i), every compact set is a small set, so that through the Nummelin splitting there is pseudo-regeneration within compact sets, and compact sets are ‘test sets’ for stability;

     
  5. (v)

    even without irreducibility, there is always a Doeblin decomposition into a countable disjoint collection of Harris sets and a transient set. We conclude with a guide to verifying Condition\(\mathcal{T}\) and indicate that it holds under very mild constraints for a wide range of specific models: in particular a ϕ-irreducible Feller chain satisfies Condition\(\mathcal{T}\) provided only that the support of ϕ has nonempty interior.

     

Mathematics Subject Classifications (1991)

60J10

Key words

Irreducible Markov processesergodicityrecurrenceHarris recurrenceT-chainsDoeblin decompositioninvariant measures

Copyright information

© Kluwer Academic Publishers 1994

Authors and Affiliations

  • R. L. Tweedie
    • 1
  1. 1.Department of StatisticsColorado State UniversityFort CollinsUSA