Automata-theoretic and bounded model checking for linear temporal logic

Abstract

In this work we study methods for model checking the temporal logic LTL. The focus is on the automata-theoretic approach to model checking and bounded model checking.

We begin by examining automata-theoretic methods to model check LTL safety properties. The model checking problem can be reduced to checking whether the language of a finite state automaton on finite words is empty. We describe an efficient algorithm for generating small finite state automata for so called non-pathological safety properties. The presented implementation is the first tool able to decide whether a formula is non-pathological. The experimental results show that treating safety properties can benefit model checking at very little cost. In addition, we find supporting evidence for the view that minimising the automaton representing the property does not always lead to a small product state space. A deterministic property automaton can result in a smaller product state space even though it might have a larger number states.

Next we investigate modular analysis. Modular analysis is a state space reduction method for modular Petri nets. The method can be used to construct a reduced state space called the synchronisation graph. We devise an on-the-fly automata-theoretic method for model checking the behaviour of a modular Petri net from the synchronisation graph. The solution is based on reducing the model checking problem to an instance of verification with testers. We analyse the tester verification problem and present an efficient on-the-fly algorithm, the first complete solution to tester verification problem, based on generalised nested depth-first search.

We have also studied propositional encodings for bounded model checking LTL. A new simple linear sized encoding is developed and experimentally evaluated. The implementation in the NuSMV2 model checker is competitive with previously presented encodings. We show how to generalise the LTL encoding to a more succint logic: LTL with past operators. The generalised encoding compares favourably with previous encodings for LTL with past operators. Links between bounded model checking and the automata-theoretic approach are also explored.