On the relations between Markov chain lumpability and reversibility (Q2406431)

The paper discusses a number of properties of ``large'' Markov chains linked to aggregation of states, which become more and more topical in relation to big data in computer science. In turn, they all relate to stationarity and reversibility. The authors introduce strict lumpability, strong and exact lumpability which are all various versions of the same idea about equal transition probabilities for certain states, yet with some differences, sometimes subtle enough. As an example, strong lumpability turns out to be equivalent to the Markov property of an aggregated process (Proposition 5). One of the main results is Theorem 2 which links exact lumpability for the process with the strong lumpability for its reversed. Other closely related properties are new notions of weak similarity, lambda-rho-reversibility and autoreversibility, all introduced by the authors in this paper or recently.