Canonicity in GFG and Transition-Based Automata

Abstract

Minimization of deterministic automata on finite words results in a {\em canonical\/} automaton. For deterministic automata on infinite words, no canonical minimal automaton exists, and a language may have different minimal deterministic Büchi (DBW) or co-Büchi (DCW) automata. In recent years, researchers have studied {\em good-for-games\/} (GFG) automata -- nondeterministic automata that can resolve their nondeterministic choices in a way that only depends on the past. Several applications of automata in formal methods, most notably synthesis, that are traditionally based on deterministic automata, can instead be based on GFG automata. The {\em minimization\/} problem for DBW and DCW is NP-complete, and it stays NP-complete for GFG Büchi and co-Büchi automata. On the other hand, minimization of GFG co-Büchi automata with {\em transition-based\/} acceptance (GFG-tNCWs) can be solved in polynomial time. In these automata, acceptance is defined by a set $α$ of transitions, and a run is accepting if it traverses transitions in $α$ only finitely often. This raises the question of canonicity of minimal deterministic and GFG automata with transition-based acceptance. In this paper we study this problem. We start with GFG-tNCWs and show that the safe components (that is, these obtained by restricting the transitions to these not in $α$) of all minimal GFG-tNCWs are isomorphic, and that by saturating the automaton with transitions in $α$ we get isomorphism among all minimal GFG-tNCWs. Thus, a canonical form for minimal GFG-tNCWs can be obtained in polynomial time. We continue to DCWs with transition-based acceptance (tDCWs), and their dual tDBWs. We show that here, while no canonical form for minimal automata exists, restricting attention to the safe components is useful, and implies that the only minimal tDCWs that have no canonical form are these for which the transition to the GFG model results in strictly smaller automaton, which do have a canonical minimal form.