Couplings of Markov chains by randomized stopping times. II: Short couplings for 0-recurrent chains and harmonic functions (Q1081211): Difference between revisions

From MaRDI portal
Added link to MaRDI item.
RedirectionBot (talk | contribs)
Removed claim: author (P16): Item:Q320226
Property / author
 
Property / author: Andreas Greven / rank
Normal rank
 

Revision as of 03:04, 13 February 2024

scientific article
Language Label Description Also known as
English
Couplings of Markov chains by randomized stopping times. II: Short couplings for 0-recurrent chains and harmonic functions
scientific article

    Statements

    Couplings of Markov chains by randomized stopping times. II: Short couplings for 0-recurrent chains and harmonic functions (English)
    0 references
    0 references
    1987
    0 references
    [Part I is to appear in ibid., see the preview Zbl 0596.60070.] We consider a 0-recurrent ergodic Markov chain on (E,\({\mathcal B})\), generated by a kernel P. Again we consider couplings of two chains \((_{\nu}X_ n)\), \((_{\mu}X_ n)\) starting with the initial distributions \(\nu\) respectively \(\mu\) and evolving with P. The coupling consists of two randomized stopping times: T, S, with \({\mathcal L}(_{\nu}X_ T)={\mathcal L}(_{\mu}X_ S).\) Under additional regularity assumptions we characterize the existence of ''short'' couplings for two chains \((_{\nu}X_ n)\), \((_{\mu}X_ n)\) by the property: \(<\nu -\mu,h>=0\) for all harmonic functions h fulfilling certain growth conditions. By ''short'' we mean that the probability to hit CÃ\({}_ m\) before T respectively S decays faster than the analogue quantity for the recurrence times of \(\nu\) and \(\mu\). Here the Ã\({}_ m\) are constructed in terms of recurrence times for a certain class of measures (Ã\({}_ m\uparrow E).\) We also show that the couplings of the chains \((_{\nu}X^ m_ k)_{k\in {\mathbb{N}}}\), \((_{\mu}X^ m_ k)_{k\in N}\) obtained by stopping the original chains, when first hitting CÃ\({}_ m\), converge to a coupling of the original chains, which is also distinguished from other exact couplings by space-time properties. We use these results to characterize the recurrent potential kernel.
    0 references
    0 references
    ergodic Markov chain
    0 references
    couplings
    0 references
    growth conditions
    0 references
    recurrent potential kernel
    0 references