(Homogeneous) Markovian bridges (Q720746): Difference between revisions

The paper is devoted to the investigation of Markov bridges, i.e., Markov chains which begin at a given point and end at a given point. In Section 2, the basic notions from potential theory for Markov chains are collected. In particular, using the Doob transform, the transience criterion for a (sub-stochastic) matrix \(P\) is proved. Further, in Section 3, the Doob transform is used to construct a killed Markov chain, a Markov bridge, and to establish the so-called ``past-future extraction''-formula for it. In Section 4, time reversal is considered for mortal chains (i.e., when the respective transition matrix \(P\) is sub-stochastic, and \(\lim_{n\to \infty} \operatorname{E}_x^P 1_{\zeta\geq n}\), where \(\zeta\) is the life-time of the chain). It is shown that a mortal Markov chain is time-reversible, in particular, this result holds true for a Markov bridge. In Section 5 a Markov bridge is constructed as a superposition of killing and conditioning. It is shown that conditioning with respect to the initial point can transform a bridge into another bridge, and conditioning with respect to the final point can transform a Markov chain or a Markov bridge to a Markov bridge. In Sections 6--8, transformations of Markov bridges are considered, such as the confinement in a certain region and a time change. In Section 9, the analogue of the Veervat transform for bridges is established. In Sections 10--12, several extensions are considered, e.g., time changes with respect to a randomized stopping time, extensions of the process up to a Martin boundary, up-normalization of the bridges. Sections 13 deals with the coincidence of Markov bridges. Section 14 closes the presentation with the proof that there exists a class of Doob transformations which characterizes a given Markov bridge and the condition under which this class is unique.