Compositional Design of Stochastic Timed Automata

In this paper, we study the model of stochastic timed automata and we target the definition of adequate composition operators that will allow a compositional approach to the design of stochastic systems with hard real-time constraints. This paper achieves the first step towards that goal. Firstly, we define a parallel composition operator that (we prove) corresponds to the interleaving semantics for that model; we give conditions over probability distributions, which ensure that the operator is well-defined; and we exhibit problematic behaviours when this condition is not satisfied. We furthermore identify a large and natural subclass which is closed under parallel composition. Secondly, we define a bisimulation notion which naturally extends that for continuous-time Markov chains. Finally, we importantly show that the defined bisimulation is a congruence w.r.t. the parallel composition, which is an expected property for a proper modular approach to system design.