A Note on Monitors and Buchi Automata

When a property needs to be checked against an unknown or very complex system, classical exploration techniques like model-checking are not applicable anymore. Sometimes a monitor can be used, that checks a given property on the underlying system at runtime. A monitor for a property L is a deterministic finite automaton M-L that after each finite execution tells whether (1) every possible extension of the execution is in L, or (2) every possible extension is in the complement of L, or neither (1) nor (2) holds. Moreover, L being monitorable means that it is always possible that in some future the monitor reaches (1) or (2). Classical examples for monitorable properties are safety and cosafety properties. On the other hand, deterministic liveness properties like "infinitely many a's" are not monitorable. We discuss various monitor constructions with a focus on deterministic omega-regular languages. We locate a proper subclass of deterministic omega-regular languages but also strictly larger than the subclass of languages which are deterministic and codeterministic; and for this subclass there exist canonical monitors which also accept the language itself. We also address the problem to decide monitorability in comparison with deciding liveness. The state of the art is as follows. Given a Buchi automaton, it is PSPACE-complete to decide liveness or monitorability. Given an LTL formula, deciding liveness becomes EXPSPACE-complete, but the complexity to decide monitorability remains open.