Linear-Time and May-Testing in a Probabilistic Reactive Setting

International audience We consider reactive probabilistic labelled transition systems (rplts), a model where internal choices are refined by probabilistic choices. In this setting, we study the relationship between linear-time and may-testing semantics, where an angelic view of nondeterminism is tak...

Full description

Bibliographic Details
Main Authors: Acciai, Lucia, Boreale, Michele, Nicola, Rocco
Other Authors: Università degli Studi di Firenze = University of Florence Firenze (UNIFI), Roberto Bruni, Juergen Dingel, TC 6, WG 6.1
Format: Conference Object
Language:English
Published: HAL CCSD 2011
Subjects:
Online Access:https://hal.inria.fr/hal-01583326
https://hal.inria.fr/hal-01583326/document
https://hal.inria.fr/hal-01583326/file/978-3-642-21461-5_2_Chapter.pdf
https://doi.org/10.1007/978-3-642-21461-5_2
Description
Summary:International audience We consider reactive probabilistic labelled transition systems (rplts), a model where internal choices are refined by probabilistic choices. In this setting, we study the relationship between linear-time and may-testing semantics, where an angelic view of nondeterminism is taken. Building on the model of d-trees of Cleaveland et al., we first introduce a clean model of probabilistic may-testing, based on simple concepts from measure theory. In particular, we define a probability space where statements of the form “p may pass test o” naturally correspond to measurable events. We then obtain an observer-independent characterization of the may-testing preorder, based on comparing the probability of sets of traces, rather than of individual traces. This entails that may-testing is strictly finer than linear-time semantics. Next, we characterize the may-testing preorder in terms of the probability of satisfying safety properties, expressed as languages of infinite trees rather than traces. We then identify a significative subclass of rplts where linear and may-testing semantics do coincide: these are the separated rplts, where actions are partitioned into probabilistic and nondeterministic ones, and at each state only one type is available.