An Information-Theoretic Framework to Aggregate a Markov Chain

This paper is concerned with an information-theoretic framework to aggregate a large-scale Markov chain to obtain a reduced order Markov model. The Kullback-Leibler (K-L) divergence rate is employed as a metric to measure the distance between two stationary Markov chains. Model reduction is obtained by considering an optimization problem with respect to this metric. The solution is just the optimal aggregated Markov model. We show that the solution of the bi-partition problem is given by an eigenvalue problem. To construct a reduced order model with m super-states, a recursive algorithm is proposed and illustrated with examples.