Lightweight Monte Carlo Verification of Markov Decision Processes with Rewards

Abstract

Markov decision processes are useful models of concurrency optimisation problems, but are often intractable for exhaustive verification methods. Recent work has introduced lightweight approximative techniques that sample directly from scheduler space, bringing the prospect of scalable alternatives to standard numerical model checking algorithms. The focus so far has been on optimising the probability of a property, but many problems require quantitative analysis of rewards. In this work we therefore present lightweight statistical model checking algorithms to optimise the rewards of Markov decision processes. We consider the standard definitions of rewards used in model checking, introducing an auxiliary hypothesis test to accommodate reachability rewards. We demonstrate the performance of our approach on a number of standard case studies.