Temporal Logics on Words with Multiple Data Values

The paper proposes and studies temporal logics for attributed words, that is, data words with a (finite) set of (attribute,value)-pairs at each position. It considers a basic logic which is a semantical fragment of the logic $LTL^\downarrow_1$ of Demri and Lazic with operators for navigation into the future and the past. By reduction to the emptiness problem for data automata it is shown that this basic logic is decidable. Whereas the basic logic only allows navigation to positions where a fixed data value occurs, extensions are studied that also allow navigation to positions with different data values. Besides some undecidable results it is shown that the extension by a certain UNTIL-operator with an inequality target condition remains decidable.

💡 Research Summary

The paper investigates temporal logics for attributed words, a generalisation of data words in which each position carries a finite set of (attribute, value) pairs rather than a single data value. This model captures realistic scenarios such as XML streams, log files, or sensor data where multiple pieces of metadata are attached to each event.

Basic logic.
The authors first define a core temporal logic that is a semantic fragment of the well‑known logic LTL↓₁ (linear‑time temporal logic with one register). The syntax includes the usual LTL operators (Next, Until, Since, etc.) together with a binding operator “↓x. φ”. The binding stores the value of a chosen attribute at the current position in a register x; subsequent sub‑formulae may compare other positions to this stored value. Crucially, the basic logic permits navigation only to positions that share the same stored value – a “fixed‑value navigation” restriction.

Decidability via data automata.
To decide the satisfiability (or model‑checking) problem for this logic, the authors translate any formula into a data automaton. A data automaton consists of a finite‑state control that reads the sequence of attribute names and a register‑based component that handles equality tests on the associated values. Because each position may contain several attributes, the construction uses a bounded number of registers, one per attribute, and encodes the binding and navigation operators as transitions with equality constraints. The emptiness problem for data automata is known to be EXPSPACE‑complete; therefore the basic logic is decidable with the same complexity bound. This result shows that, despite the richer word model, the core temporal fragment remains algorithmically tractable.

Extensions with different‑value navigation.
The paper then explores extensions that allow navigation to positions whose values differ from the stored one. Operators such as “◇≠x φ” (eventually reach a position with a different value) or “U≠” (until a different value) are introduced. These extensions dramatically increase expressive power, enabling specifications like “after a request with ID i, eventually see a response with a different ID”. However, the authors prove that unrestricted use of such operators leads to undecidability. By encoding a two‑counter Minsky machine, they show that the satisfiability problem becomes Σ₁¹‑complete, i.e., not recursively enumerable. Hence, naïvely adding inequality‑based navigation destroys the decidability frontier.

A decidable inequality‑until fragment.
Surprisingly, a carefully restricted inequality‑based operator retains decidability. The authors define an inequality‑until operator φ U≠ ψ that requires the path from the current position to the first position whose value differs from the stored one to satisfy φ, and ψ must hold at that first differing position. The key restriction is that the “different‑value” condition can be satisfied only once along any evaluation of the operator. Under this limitation, the translation to data automata can be adapted: the automaton guesses the unique “break point” where the value changes and checks φ up to that point and ψ at the break point. The resulting automaton remains within the class for which emptiness is decidable, preserving the EXPSPACE bound.

Contributions and impact.

1. Model definition: Introduction of attributed words as a natural extension of data words, together with a precise formalisation of their semantics.
2. Core logic decidability: A reduction of the basic fragment to data automata, yielding a decidability proof and an explicit complexity estimate.
3. Expressiveness vs. decidability trade‑off: Systematic study of extensions with inequality navigation, identification of undecidable extensions, and isolation of a maximal decidable fragment (the inequality‑until operator).

The results have practical relevance for verification of systems that manipulate rich metadata. For instance, one could automatically check that in a log stream, every login event (attribute userID) is eventually followed by a logout event with a different session identifier, or that XML documents satisfy temporal constraints across elements with distinct attribute values. Moreover, the methodology demonstrates how data automata—originally devised for single‑value data words—can be extended to handle multiple attributes per position, opening avenues for further research on richer data models and their associated logics.

In summary, the paper delineates the boundary between decidable and undecidable temporal specifications over multi‑valued data streams, providing both theoretical insights and a foundation for future tool development.