Abstract
This paper is centered on the random graph generated by a Doeblin-type coupling of discrete time processes on a countable state space whereby when two paths meet, they merge. This random graph is studied through a novel subgraph, called a bridge graph, generated by paths started in a fixed state at any time. The bridge graph is made into a unimodular network by marking it and selecting a root in a specified fashion. The unimodularity of this network is leveraged to discern global properties of the larger Doeblin graph. Bi-recurrence, i.e., recurrence both forwards and backwards in time, is introduced and shown to be a key property in uniquely distinguishing paths in the Doeblin graph, and also a decisive property for Markov chains indexed by $\mathbb{Z} $. Properties related to simulating the bridge graph are also studied.
Citation
François Baccelli. Mir-Omid Haji-Mirsadeghi. James T. Murphy III. "Doeblin trees." Electron. J. Probab. 24 1 - 36, 2019. https://doi.org/10.1214/19-EJP375
Information