Safety alternating automata on data words

A data word is a sequence of pairs of a letter from a finite alphabet and an element from an infinite set, where the latter can only be compared for equality. Safety one-way alternating automata with one register on infinite data words are considered, their nonemptiness is shown EXPSPACE-complete, and their inclusion decidable but not primitive recursive. The same complexity bounds are obtained for satisfiability and refinement, respectively, for the safety fragment of linear temporal logic with freeze quantification. Dropping the safety restriction, adding past temporal operators, or adding one more register, each causes undecidability.

💡 Research Summary

The paper investigates the theoretical properties of one‑way alternating automata equipped with a single register (safety 1ARA₁) that run over infinite data words—sequences of (letter, datum) pairs where data can only be compared for equality. A data word is thus a finite‑alphabet stream enriched with an equivalence relation on positions. The authors focus on the safety acceptance condition: a language is safe if every rejected word has a finite prefix that already guarantees rejection. This mirrors the classic notion of safety in ω‑languages.

The main contributions are fourfold. First, the non‑emptiness problem for safety 1ARA₁ is shown to be EXPSPACE‑complete. The proof proceeds by translating a given safety automaton into a faulty counter automaton (CA) whose counters may spontaneously increase (incrementing errors). This translation is log‑space computable, preserving non‑emptiness while moving the problem into a counter‑based setting. The authors then prove an inductive counting argument: if the CA is non‑empty, it admits an accepting run of length at most doubly exponential in the size of the original automaton. Consequently, a deterministic EXPSPACE algorithm decides non‑emptiness.

Second, the inclusion problem between two safety 1ARA₁ is decidable, but its complexity exceeds any primitive‑recursive bound. By reducing inclusion to a combination of non‑emptiness checks and exploiting the known Π₀¹‑hardness of non‑emptiness for unrestricted 1ARA₁, the authors establish that inclusion is decidable yet not primitive‑recursive.

Third, the paper connects safety 1ARA₁ with the safety fragment of future‑time linear temporal logic equipped with a single freeze register (safety LTL↓₁). A safety LTL formula is one where every “until” operator occurs under an odd number of negations (equivalently, the dual “release” operator is used). The authors give a log‑space translation from safety LTL↓₁ to safety 1ARA₁, showing that satisfiability for the logic is also EXPSPACE‑complete and that refinement (implication) is decidable but not primitive‑recursive. This positions safety LTL↓₁ as a highly expressive yet still decidable logic for data ω‑words, incomparable in expressive power with FO²(∼,<,+1).

Finally, the authors demonstrate the fragility of decidability: removing the safety restriction, adding past temporal operators, or allowing a second register each leads to undecidability (Π₀¹‑hardness). These results underline that safety is the crucial syntactic discipline that keeps the decision problems within the EXPSPACE realm.

Overall, the work provides a detailed complexity landscape for automata and logics over data words, showing that safety‑restricted alternating automata with one register are both expressive enough for many verification tasks and amenable to algorithmic analysis, while any modest relaxation quickly pushes the problems beyond elementary or even primitive‑recursive bounds.