Model Checking Using Generalized Testing Automata

Abstract

Abstract. Geldenhuys and Hansen showed that a kind of ω-automata known as Testing Automata (TA) can, in the case of stuttering-insensitive properties, outperform the Büchi automata traditionally used in the automata-theoretic approach to model checking~ [10]. In previous work~[23], we compared TA against Transition-based Generalized Büchi Automata (TGBA), and concluded that TA were more interesting when counterexamples were expected, otherwise TGBA were more efficient. In this work we introduce a new kind of automata, dubbed Transition-based Generalized Testing Automata (TGTA), that combine ideas from TA and TGBA. Implementation and experimentation of TGTA show that they outperform other approaches in most of the cases.

Extending Testing Automata to All LTL

“…We have shown that TA are better than Büchi automata (BA and TGBA) when the formula to be verified is violated (i.e., a counterexample is found), but this is not the case when the property is verified since the entire product have to be visited twice to check for each acceptance mode of a TA. Then, in order to improve the TA approach, we proposed in [2] a new ω-automata for stutter-invariant properties, called Transition-based Generalized Testing Automata (TGTA) [2], that mixes features from both TA and TGBA.…”

“…In [2], we compared the LTL model checking approach using TGTA with the "traditional" approaches using BA, TGBA and TA. The results of these experimental comparisons show that TGTA compete well: the TGTA approach was statistically more efficient than the other evaluated approaches, especially when no counterexample is found (i.e., the property is verified) because it does not require a second pass.…”

“…The results of these experimental comparisons show that TGTA compete well: the TGTA approach was statistically more efficient than the other evaluated approaches, especially when no counterexample is found (i.e., the property is verified) because it does not require a second pass. Unfortunately, the TGTA constructed in [2] only represent stutter-invariant properties (LTL\ X [13]). …”

“…Let us now describe how to build a TGTA starting from a TGBA. The TGTA construction is inspired by the one presented in [2], with some changes introduced in the second step of this construction. Indeed, a TGTA is built in two steps as illustrated in Figure 3.…”

“…In previous work [2], we propose an improvement of TA called Transition-based Generalized Testing Automata (TGTA) that combine the advantages of both TA and TGBA, and without the disadvantages of TA (without introducing a second mode of acceptance and without the second pass of the emptiness check).…”