Skip to main content

Approximate Verification of the Symbolic Dynamics of Markov Chains

S. Akshay ( IRISA/ENS Cachan Bretagne )

A finite state Markov chain M is often viewed as a probabilistic transition system. An alternative view - which we follow here - is to regard M as a linear transform operating on the space of probability distributions over its set of nodes. The novel idea here is to discretize the probability value space [0,1] into a finite set of intervals. A concrete probability distribution over the nodes is then symbolically represented as a tuple D of such intervals. The i-th component of the discretized distribution D will be the interval in which the probability of node i falls. The set of discretized distributions is a finite set and each trajectory, generated by repeated applications of M to an initial distribution, will induce a unique infinite string over this finite set of letters. Hence, given a set of initial distributions, the symbolic dynamics of M will consist of an infinite language L over the finite alphabet of discretized distributions. We investigate whether L meets a specification given as a linear time temporal logic formula whose atomic propositions assert that the current probability of a node falls in an interval. Unfortunately, even for restricted Markov chains (for instance, irreducible and aperiodic chains), we do not know at present if and when L is an (omega)-regular language. To get around this we develop the notion of an epsilon-approximation, based on the transient and long term behaviors of M. Our main results are that, one can effectively check whether (i) for each infinite word in L, at least one of its epsilon-approximations satisfies the specification; (ii) for each infinite word in L all its epsilon-approximations satisfy the specification. These verification results are strong in that they apply to all finite state Markov chains. Further, the study of the symbolic dynamics of Markov chains initiated here is of independent interest and can lead to other applications.

Speaker bio

S. Akshay obtained his Ph.D in 2010 from LSV, Ecole Normale Superieure de Cachan, France. Since then, he completed a one-year research fellowship at the National University of Singapore and is currently pursuing a post-doctoral fellowship at IRISA, ENS Cachan Bretagne in Rennes. His research interests include modeling and verification of distributed timed systems and probabilistic verification with applications to systems biology.

Share this: