On the Complexity of Elementary Modal Logics

Modal logics are widely used in computer science. The complexity of modal satisfiability problems has been investigated since the 1970s, usually proving results on a case-by-case basis. We prove a very general classification for a wide class of relevant logics: Many important subclasses of modal logics can be obtained by restricting the allowed models with first-order Horn formulas. We show that the satisfiability problem for each of these logics is either NP-complete or PSPACE-hard, and exhibit a simple classification criterion. Further, we prove matching PSPACE upper bounds for many of the PSPACE-hard logics.

💡 Research Summary

The paper “On the Complexity of Elementary Modal Logics” investigates the computational complexity of satisfiability problems for a broad family of modal logics that are defined by restricting the class of admissible Kripke frames with first‑order universal Horn formulas. The authors consider a fixed first‑order Horn constraint ψ̂ over the binary relation R (the accessibility relation of a frame) and define K(ψ̂) as the modal logic whose formulas are satisfiable exactly on frames that satisfy ψ̂. Classical modal logics such as K, K4, S4, S5, etc., correspond to simple Horn constraints expressing reflexivity, transitivity, symmetry, or combinations thereof.

The central research question is: for a given universal Horn formula ψ̂, what is the complexity of the decision problem K(ψ̂)‑SAT (given a modal formula φ, does there exist a ψ̂‑model and a world where φ holds)? The authors establish a clean dichotomy: every such logic falls into one of two categories—either its satisfiability problem is NP‑complete or it is PSPACE‑hard (and, for many of the latter, PSPACE‑complete). No intermediate complexity classes appear within this family.

The dichotomy is driven by a structural analysis of ψ̂. If ψ̂ enforces at most one of the three classic frame properties (reflexivity, transitivity, symmetry) in a “simple” way, then every satisfiable formula has a model whose size is bounded by a polynomial in the length of the formula. This “polynomial‑size model property” yields an NP upper bound via a standard guess‑and‑check algorithm, and NP‑hardness follows from the fact that any non‑trivial modal logic already subsumes propositional SAT. The authors give a precise classification criterion based on the Horn clauses: essentially, ψ̂ must not contain a combination of clauses that simultaneously forces two of the three properties in a non‑trivial way.

Conversely, when ψ̂ contains a combination that forces at least two of the properties (e.g., both transitivity and reflexivity, or transitivity and symmetry), the authors adapt Ladner’s classic PSPACE‑hardness construction. They reduce quantified Boolean formulas (QBF) to modal formulas such that any model of the resulting formula must embed a “quantifier tree” reflecting the alternation of existential and universal quantifiers. The Horn constraints guarantee that the frame must realize this tree structure, and thus any algorithm solving K(ψ̂)‑SAT would solve QBF, establishing PSPACE‑hardness. The paper shows that this argument works for all Horn constraints that are not covered by the NP‑case, thereby proving the dichotomy.

Beyond hardness, the authors also provide matching PSPACE upper bounds for many of the PSPACE‑hard logics. They introduce a “tree‑like model property” for these logics: although the frames may be complex, any satisfiable formula can be satisfied in a model that is locally tree‑shaped and respects the Horn constraints in a step‑wise fashion. Leveraging this property, they design a tableau‑style decision procedure that runs in polynomial space, generalizing earlier PSPACE algorithms for specific logics (e.g., K4, S4). Consequently, for a large subclass of the PSPACE‑hard logics, K(ψ̂)‑SAT is PSPACE‑complete.

The paper is organized as follows. Section 2 sets up notation, defines universal elementary modal logics, and recalls Ladner’s results. Section 3 develops tools for bounding model size, which are crucial for the NP proofs. Section 4 contains the core technical contributions: after introducing the Horn‑based classification (4.1), relating Horn formulas to homomorphisms (4.2), and proving NP results for special Horn families (4.3‑4.4), the authors present the main dichotomy theorem (Corollary 4.29) in 4.5. Sections 4.6‑4.7 establish the tree‑like model property and the PSPACE algorithm, respectively. Section 4.8 derives several corollaries, including an optimality result for Ladner’s hardness condition and concrete applications to many known modal logics. The paper concludes with a summary and open questions in Section 5.

In summary, the work delivers a unified, systematic classification of the complexity of elementary modal logics defined by universal Horn frame constraints. It shows that within this natural and expressive family, the only possible complexities are NP‑complete or PSPACE‑complete, thereby eliminating the possibility of intermediate complexities. The results both subsume many previously known case‑by‑case analyses and provide new PSPACE algorithms for a broad range of logics, offering valuable guidance for the design and analysis of modal systems in computer‑science applications.