Similarity quantification for linear stochastic systems as a set-theoretic control problem

For the formal verification and design of control systems, abstractions with quantified accuracy are crucial. Such similarity quantification is hindered by the challenging computation of approximate stochastic simulation relations. This is especially the case when considering accurate deviation bounds between a stochastic continuous-state model and its finite-state abstraction. In this work, we give a comprehensive computational approach and analysis for linear stochastic systems. More precisely, we develop a computational method that characterizes the set of possible simulation relations and optimally trades off the error contributions on the system's output with deviations in the transition probability. To this end, we establish an optimal coupling between the models and simultaneously solve the approximate simulation relation problem as a set-theoretic control problem using the concept of invariant sets. We show the variation of the guaranteed satisfaction probability as a function of the error trade-off in a case study where a formal specification is given as a temporal logic formula.