Finite-State Abstractions for Probabilistic Computation Tree Logic

Daniel Wagner, Michael Huth, William Knottenbelt
Probabilistic Computation Tree Logic (PCTL) is the established temporal logic for probabilistic verification of discrete-time Markov chains. Probabilistic model checking is a technique that verifies or refutes whether a property specified in this logic holds in a Markov chain. But Markov chains are often infinite or too large for this technique to apply. A standard solution to this problem is to convert the Markov chain to an abstract model and to model check that abstract model. The problem
more » ... s thesis therefore studies is whether or when such finite abstractions of Markov chains for model checking PCTL exist. This thesis makes the following contributions. We identify a sizeable fragment of PCTL for which 3-valued Markov chains can serve as finite abstractions; this fragment is maximal for those abstractions and subsumes many practically relevant specifications including, e.g., reachability. We also develop game-theoretic foundations for the semantics of PCTL over Markov chains by capturing the standard PCTL semantics via a two-player games. These games, finally, inspire a notion of p-automata, which accept entire Markov chains. We show that p-automata subsume PCTL and Markov chains; that their languages of Markov chains have pleasant closure properties; and that the complexity of deciding acceptance matches that of probabilistic model checking for p-automata representing PCTL formulae. In addition, we offer a simulation between p-automata that under-approximates language containment. These results then allow us to show that p-automata comprise a solution to the problem studied in this thesis.
doi:10.25560/6348 fatcat:43p6ox4g5rdrnexzjpxuhjj7d4