Algorithmic metatheorems for decidable LTL model checking over infinite systems

By algorithmic metatheorems for a model checking problem P over infinite-state systems we mean generic results that can be used to infer decidability (possibly complexity) of P not only over a specific class of infinite systems, but over a large family of classes of infinite systems. Such results normally start with a powerful formalism of infinite-state systems, over which P is undecidable, and assert decidability when is restricted by means of an extra “semantic condition” C. We prove various algorithmic metatheorems for the problems of model checking LTL and its two common fragments LTL(Fs,Gs) and LTLdet over the expressive class of word/tree automatic transition systems, which are generated by synchronized finite-state transducers operating on finite words and trees. We present numerous applications, where we derive (in a unified manner) many known and previously unknown decidability and complexity results of model checking LTL and its fragments over specific classes of infinite-state systems including pushdown systems; prefix-recognizable systems; reversal-bounded counter systems with discrete clocks and a free counter; concurrent pushdown systems with a bounded number of context-switches; various subclasses of Petri nets; weakly extended PA-processes; and weakly extended ground-tree rewrite systems. In all cases,we are able to derive optimal (or near optimal) complexity. Finally, we pinpoint the exact locations in the arithmetic and analytic hierarchies of the problem of checking a relevant semantic condition and the LTL model checking problems over all word/tree automatic systems.

💡 Research Summary

The paper introduces a family of algorithmic metatheorems that turn the generally undecidable problem of Linear Temporal Logic (LTL) model checking on infinite‑state systems into a decidable one under a well‑defined semantic restriction. The authors work with the highly expressive class of word‑ and tree‑automatic transition systems, which are generated by synchronized finite‑state transducers acting on finite words or trees. Although LTL model checking is undecidable on this class in full generality, the paper shows that if the system satisfies a “semantic condition C” – namely that the predecessor operation on any regular (or regular tree) set yields another regular set – then all the standard automata‑theoretic constructions used for LTL (product with a Büchi automaton, complementation, fix‑point computation) stay within the regular world. Consequently, the model‑checking problem becomes decidable and its complexity can be precisely characterised.

Three metatheorems are proved: one for full LTL, one for the fragment LTL(Fs,Gs) that only uses finally and globally operators, and one for LTLdet, the deterministic‑Büchi‑automaton fragment of LTL. For the full logic the generic upper bound is 2‑EXPTIME, matching known lower bounds for many automatic structures. For LTL(Fs,Gs) the bound drops to EXPTIME, and for LTLdet it falls further to PSPACE, because deterministic automata avoid the costly complementation step.

The power of these metatheorems is demonstrated by systematically re‑deriving a large collection of known decidability and complexity results, and by establishing several new ones, for a variety of well‑studied infinite‑state models:

• Pushdown systems – captured as word‑automatic; LTL is 2‑EXPTIME‑complete, LTL(Fs,Gs) EXPTIME, LTLdet PSPACE.
• Prefix‑recognizable systems – also satisfy C; the same complexity profile as pushdown systems holds.
• Reversal‑bounded counter systems with discrete clocks and a free counter – the bounded reversal guarantees regularity of predecessor sets; LTL checking is EXPTIME, the fragments are PSPACE.
• Concurrent pushdown systems with a bounded number of context‑switches – the context‑switch bound enforces C; LTL becomes 2‑EXPTIME, fragments drop to EXPTIME.
• Various subclasses of Petri nets – weak extensions of PA‑processes and ground‑tree rewrite systems meet C; the metatheorems give EXPTIME or 2‑EXPTIME bounds depending on the exact subclass.
• Other weakly extended models – the same methodology yields optimal or near‑optimal complexities.

In each case the authors show that the semantic condition C can be checked with the same level of difficulty as the underlying model‑checking problem. Moreover, they locate the decision problem “does a given automatic system satisfy C?” in the analytical hierarchy as Σ₁¹‑complete, and they prove that LTL model checking over all automatic systems is Π₁¹‑complete. These results pinpoint the exact boundary between decidable and undecidable instances within the broad landscape of automatic structures.

Overall, the paper provides a unifying framework: by reducing LTL model checking on many infinite‑state systems to regular language operations under condition C, it not only explains why a host of previously disparate results share the same complexity, but also offers a systematic recipe for tackling new classes of infinite systems. The metatheorems are both technically deep – involving intricate constructions of synchronized transducers and careful analysis of regularity preservation – and practically valuable, as they give researchers a clear checklist (automatic presentation + condition C) that guarantees decidable LTL verification with known optimal complexity.