When Symmetry Yields NP-Hardness: Affine ML-SAT on S5 Frames

Hemaspaandraetal.~[JCSS 2010] conjectured that satisfiability for multi-modal logic restricted to the connectives XOR and 1, over frame classes T, S4, and S5, is solvable in polynomial time. We refute this for S5 frames, by proving NP-hardness.

💡 Research Summary

The paper addresses a conjecture made by Hemaspaandra et al. (JCSS 2010) that the satisfiability problem for multi‑modal logic restricted to the exclusive‑or (⊕) connective and the constant true (1) should be solvable in polynomial time on the frame classes T, S4, and S5. The authors refute this conjecture for the S5 class by proving that the problem is NP‑hard.

The authors begin by recalling the basics of modal logic, Kripke semantics, and the classification of frames (reflexive, transitive, symmetric). They define the decision problem S5‑⊕‑MSAT₂{□,◇} as: given a formula built from propositional variables, the binary XOR connective, and the modal operators □ and ◇ (with two modalities), does there exist an S5‑frame and a world where the formula holds?

The core of the proof is a polynomial‑time many‑one reduction from 3‑SAT. Starting from a 3‑CNF formula f, they first transform it into a new formula f′ (Lemma 6) that introduces fresh “buffer” variables b_i and guarantees that every conjunction g∧h occurring in f′ consists of sub‑formulas that are strongly independent: each side contains at least one clause whose variables do not appear in the other side. This independence property is crucial for the later model construction.

Next, each literal a is encoded as φ(a)=□₁a⊕□₂□₁a, and each negated literal ¬a as φ(¬a)=□₁¬a⊕□₂□₁¬a. A clause C = l₁∨l₂∨l₃ is encoded as φ(C)=◇₁(φ(l₁)⊕φ(l₂)⊕φ(l₃)). For a conjunction g∧h, the authors define auxiliary formulas ψ_g = φ(g)⊕□₂φ(g) and ψ_h analogously, and then set φ(g∧h)=◇₁(ψ_g⊕ψ_h)⊕◇₁ψ_g⊕◇₁ψ_h. The construction ensures that the size of the resulting modal formula φ(f) remains polynomial in the size of the original 3‑SAT instance.

The proof then introduces the notion of a “nice model”. A nice model for a sub‑formula is an S5‑frame that contains a distinguished equivalence class (the core) where every world satisfies the sub‑formula, and the truth values of the underlying propositional variables in the core correspond exactly to a satisfying assignment of the original 3‑SAT instance. Lemma 9 shows how to build a nice model for a single clause. Lemma 10 proves that extending a nice model with an arbitrary additional component (while connecting the cores via the reflexive relation) preserves the niceness property. Lemma 11 exploits the strong independence of g and h: given a model where both φ(g) and φ(h) hold, one can modify it to obtain a model where φ(g) holds but φ(h) does not, by falsifying a clause that uses variables exclusive to h. Lemma 12 combines the previous lemmas to construct a nice model for the conjunction φ(g∧h) from nice models of φ(g) and φ(h). The construction heavily relies on the symmetry of S5: worlds are linked bidirectionally by the □₁ relation, guaranteeing that every world can reach every other, which is essential for the ◇₁ operators to behave as intended.

By recursively applying Lemma 12 according to the syntactic tree of f′, the authors obtain a global nice model for φ(f). Consequently, f is satisfiable iff φ(f) is satisfiable on an S5‑frame. This establishes that S5‑⊕‑MSAT₂{□,◇} is NP‑hard, thereby disproving the conjecture for S5.

The paper concludes with a brief discussion of the remaining open cases: S4‑⊕‑MSAT₁{□,◇} and T‑⊕‑MSAT₂{□}. The authors note that the NP‑hardness proof crucially uses the symmetry property of S5, which is absent in S4 and T, so the same technique does not directly apply. They suggest that further work is needed to resolve the complexity of these two fragments.