Handling Conflicts in Depth-First Search for LTL Tableau to Debug Compliance Based Languages

Providing adequate tools to tackle the problem of inconsistent compliance rules is a critical research topic. This problem is of paramount importance to achieve automatic support for early declarative design and to support evolution of rules in contract-based or service-based systems. In this paper we investigate the problem of extracting temporal unsatisfiable cores in order to detect the inconsistent part of a specification. We extend conflict-driven SAT-solver to provide a new conflict-driven depth-first-search solver for temporal logic. We use this solver to compute LTL unsatisfiable cores without re-exploring the history of the solver.

💡 Research Summary

The paper addresses the pressing need for automated detection of inconsistent compliance rules in contract‑oriented and service‑based systems. While many existing tools can verify temporal properties, few can pinpoint the exact subset of LTL (Linear Temporal Logic) specifications that cause unsatisfiability. The authors propose a novel conflict‑driven depth‑first‑search (DFS) solver that integrates SAT‑solver conflict learning techniques directly into the on‑the‑fly tableau construction for LTL.

First, the paper reviews LTL syntax and semantics, and recalls the classic tableau method where each state is a set of closure formulas and transitions are driven by the X (next) operator. Traditional on‑the‑fly approaches explore the tableau using DFS or SCC detection but treat conflicts merely by backtracking, without learning from them. Consequently, the same contradictory sub‑formulae are revisited many times, leading to exponential blow‑up.

The core contribution is the adaptation of SAT‑solver conflict analysis to temporal logic tableau exploration. Each closure formula ψ is associated with a fresh Boolean variable x_ψ indicating its presence in the current state. The tableau expansion rules (conjunction, disjunction, F/G, U/W) are encoded as implication clauses over these Boolean variables, forming a propositional SAT problem that mirrors the temporal unfolding. When a state contains both x_h and ¬x_h, a conflict is detected. The solver then extracts a conflict clause (¬x_h ∨ ¬x_¬h) and learns it, preventing the same combination of choices from being explored again. This learned clause also enables non‑chronological backjumping: the search jumps directly to the most recent decision point involved in the conflict, rather than stepping back one level at a time.

To further prune the search space, the authors introduce a “fair prime implicant” strategy. After unit propagation, only the minimal set of literals that must hold for a fair continuation of the path is kept, discarding redundant branches that would otherwise lead to the same SCC. The combination of conflict learning and prime‑implicant pruning yields a dramatically smaller exploration tree.

Theoretical results establish soundness (every model found satisfies the original LTL formula), completeness (if a model exists the algorithm will find a fair SCC), and a PSPACE bound for deciding whether a given unsatisfiable core is minimal. The paper also shows that extracting a minimal unsatisfiable core is PSPACE‑complete, matching known results for LTL satisfiability.

An illustrative example involves a set of contract rules (r1, r2, r3) where a newly added rule r3.c conflicts with existing rules r3.a and r3.b. Using the proposed solver, the algorithm quickly discovers the conflict, learns the corresponding clause, and avoids re‑examining the same contradictory combination. The resulting unsatisfiable core consists of only three rules, which is close to a minimal core and far smaller than the full set of clauses that would be examined by traditional Unbounded Model Checking (UMC) or naive tableau methods.

Empirical evaluation (though limited to the toy example) demonstrates a substantial reduction in the number of explored states and conflict checks compared to UMC, confirming the practical benefit of integrating conflict learning into temporal logic verification.

In conclusion, the paper successfully bridges SAT‑solver conflict‑driven learning with LTL tableau DFS, delivering an efficient method for extracting small unsatisfiable cores in compliance‑based languages. Future work is suggested on scaling the approach to larger specifications, extending it to richer temporal logics such as MTL or TPTL, and integrating it into real‑world contract management tools.