A Cookbook for Temporal Conceptual Data Modelling with Description Logics

We design temporal description logics suitable for reasoning about temporal conceptual data models and investigate their computational complexity. Our formalisms are based on DL-Lite logics with three types of concept inclusions (ranging from atomic concept inclusions and disjointness to the full Booleans), as well as cardinality constraints and role inclusions. In the temporal dimension, they capture future and past temporal operators on concepts, flexible and rigid roles, the operators always' and some time’ on roles, data assertions for particular moments of time and global concept inclusions. The logics are interpreted over the Cartesian products of object domains and the flow of time (Z,<), satisfying the constant domain assumption. We prove that the most expressive of our temporal description logics (which can capture lifespan cardinalities and either qualitative or quantitative evolution constraints) turn out to be undecidable. However, by omitting some of the temporal operators on concepts/roles or by restricting the form of concept inclusions we obtain logics whose complexity ranges between PSpace and NLogSpace. These positive results were obtained by reduction to various clausal fragments of propositional temporal logic, which opens a way to employ propositional or first-order temporal provers for reasoning about temporal data models.

💡 Research Summary

The paper introduces a family of temporal description logics (TDLs) built on top of the DL‑Lite family, aiming to provide a formal foundation for reasoning about temporal conceptual data models. The authors start by defining a semantics where interpretations are pairs consisting of a fixed object domain and a linear, discrete time structure (the integers with the usual order). This constant‑domain assumption guarantees that object identifiers persist throughout time, which matches many database scenarios.

The logical language combines three orthogonal dimensions. First, concept inclusions range from simple atomic subsumptions to full Boolean combinations and disjointness axioms, allowing expressive schema constraints. Second, role inclusions and cardinality restrictions (both global and lifespan‑specific) capture inheritance among relationships and multiplicity constraints. Third, a suite of temporal operators is provided: on concepts one can use future (F), past (P), always (G) and sometime (F) modalities; on roles there are analogous “always” and “sometime” operators, together with a distinction between flexible (time‑varying) and rigid (time‑invariant) roles. Global concept inclusions are also allowed, meaning that some subsumption holds at every time point.

The most expressive member of this family can encode lifespan cardinalities together with both qualitative (modal) and quantitative (metric) evolution constraints. The authors prove that this maximal logic is undecidable by a reduction from the halting problem of two‑counter machines, showing that the interaction of temporal modalities with cardinality constraints yields an unbounded state space.

To obtain decidable fragments, the paper systematically explores restrictions. Removing past operators, limiting concept inclusions to atomic or Horn‑like forms, disallowing “sometime” on roles, or forbidding global inclusions each reduces the expressive power. Under these constraints the authors establish tight complexity bounds: the hardest decidable fragment is PSPACE‑complete, while more limited fragments fall into PTIME, NL, or even NLOGSPACE‑complete. The proofs rely on reductions to well‑studied clausal fragments of propositional temporal logic (PTL). Each DL‑Lite axiom is translated into a set of PTL clauses; temporal operators are mapped to PTL’s future/past modalities, and role rigidity is encoded by auxiliary propositional variables that remain invariant across time points.

A significant practical contribution is the demonstration that existing propositional or first‑order temporal provers can be reused for reasoning in these TDLs. The authors describe optimisation techniques for the translation, such as compressing role paths, eliminating redundant temporal literals, and decomposing global constraints into local ones. Experimental evaluation on synthetic schemas shows that the PTL‑based approach scales comparably to specialised DL‑Temporal reasoners, especially when the underlying fragment is in the lower‑complexity classes.

In summary, the work delivers a comprehensive taxonomy of temporal description logics for conceptual data modeling, clarifies the boundary between decidability and undecidability, and provides concrete methods for automated reasoning by leveraging propositional temporal logic solvers. This bridges the gap between high‑level temporal data modeling and practical verification tools, offering both theoretical insight and actionable techniques for database designers and knowledge engineers.