Measures on probabilistic automata

In this thesis we consider nondeterministic probabilistic processes modeled by automata. Our purpose is the analysis of the problem of approximated bisimulations. These relations are used, generally, to simplify the models of some systems and to model agents and attackers in security protocols. For the latter field there are several proposals to use metrics, which are the quantitative analogue of probabilistic bisimilarity and allow a greater precision. A metric is about a degree of similarity between states. Starting from the formalisation of approximate (bi)simulation given in Turrini's work, we define two metrics on states and on distributions. These metrics are based on the concept of error allowed during the simulation of a state with respect to another one. We investigate the relation between these metrics with a largely used one, the Kantorovich metric, and discover that they are equivalent. Then we recast for probabilistic automata the transformer of measures proposed by De Alfaro et al., obtaining a new functional F that is a conservative extension of the transformers proposed in the literature. We show that the minimum fix point of F coincides with its over-aproximated by the measures derived from Turrini's work thus showing the existence of a strict relation between the Turrini’s approximate bisimulations with the literature on metrics.