Tableau-based decision procedure for the multi-agent epistemic logic with operators of common and distributed knowledge

We develop an incremental-tableau-based decision procedure for the multi-agent epistemic logic MAEL(CD) (aka S5_n (CD)), whose language contains operators of individual knowledge for a finite set Ag of agents, as well as operators of distributed and common knowledge among all agents in Ag. Our tableau procedure works in (deterministic) exponential time, thus establishing an upper bound for MAEL(CD)-satisfiability that matches the (implicit) lower-bound known from earlier results, which implies ExpTime-completeness of MAEL(CD)-satisfiability. Therefore, our procedure provides a complexity-optimal algorithm for checking MAEL(CD)-satisfiability, which, however, in most cases is much more efficient. We prove soundness and completeness of the procedure, and illustrate it with an example.

💡 Research Summary

The paper presents a deterministic, incremental tableau decision procedure for the multi‑agent epistemic logic MAEL(CD), also known as S5ₙ(CD). MAEL(CD) extends standard S5ₙ epistemic logic with two additional modalities: distributed knowledge (D) and common knowledge (C) over a finite set of agents Ag. While the ExpTime‑hardness of the satisfiability problem for this logic has been known for some time, no matching upper‑bound algorithm had been published. The authors fill this gap by constructing a tableau system that runs in exponential time, thereby proving ExpTime‑completeness of MAEL(CD)‑satisfiability and delivering a complexity‑optimal decision procedure.

The paper begins with a precise definition of the syntax and semantics of MAEL(CD). Formulas are built from propositional atoms using the usual Boolean connectives together with K_i φ (agent i knows φ), D φ (the group’s distributed knowledge of φ) and C φ (the group’s common knowledge of φ). Semantically, models are Kripke structures ⟨W, {R_i}{i∈Ag}, V⟩ where each R_i is an equivalence relation (reflecting S5 properties). The D‑operator is interpreted as truth in all worlds reachable via the intersection of all R_i, while C φ is defined as the least fixed point of the equation C φ ↔ φ ∧ ⋂{i∈Ag}K_i C φ.

The core contribution is the tableau construction. A tableau node (world) carries a label – a set of formulas that are required to hold at that world. The procedure starts with a single node labelled by the input formula ϕ. Three families of expansion rules are applied repeatedly:

1. α/β rules decompose propositional connectives (¬¬, φ∧ψ, φ∨ψ, etc.) into simpler components.
2. K‑rules handle individual knowledge: if K_i φ appears in a label, a successor node w′ is created (or reused) such that φ is added to w′’s label and the edge (w, w′) is marked with R_i.
3. D‑rules enforce distributed knowledge: D φ in a label forces φ to be added to every existing node and to any node that will be created later, reflecting the fact that D φ is true exactly when φ is known by the coalition as a whole.
4. C‑rules implement the common‑knowledge fixed point. When C φ occurs, the tableau adds both φ and, for every agent i, K_i C φ to the same label. If C φ is already present, no further expansion is performed, preventing infinite unfolding.

A crucial design choice is incrementality: new worlds are generated only when a rule explicitly requires them, and before creating a world the algorithm checks whether an identical label already exists, thereby avoiding redundant duplication. Contradictions (e.g., p and ¬p, or C φ together with ¬C φ) close a branch. A tableau is open if at least one branch remains contradiction‑free; otherwise it is closed. The input formula is satisfiable iff the tableau admits an open branch.

Soundness is proved by constructing a canonical model from any open branch: worlds correspond to labels, relations are defined by the K‑rules, and the truth lemma shows that every formula in a label holds in the associated world. Completeness is established by showing that any satisfiable formula yields a tableau that cannot be fully closed; the fixed‑point treatment of C guarantees that the necessary infinite unfolding is captured finitely.

Complexity analysis shows that each rule can be applied in polynomial time relative to the size of the current tableau. The number of distinct labels is bounded by 2^{|Sub(ϕ)|}, where Sub(ϕ) denotes the set of sub‑formulas of ϕ. Consequently, the total number of nodes and edges is at most exponential in |ϕ|, and the whole procedure terminates after at most exponential work. This matches the known ExpTime lower bound, establishing that MAEL(CD)‑satisfiability is ExpTime‑complete.

An illustrative example walks through the tableau for the formula C(p → K₁ p) ∧ ¬K₂ p. The construction demonstrates how the C‑rule introduces the fixed‑point component, how the K₁‑rule creates a successor world where p and K₁ p hold, and how the branch remains open, confirming satisfiability.

The paper concludes with a discussion of related work, notably earlier tableau approaches for plain S5ₙ and for logics with only common knowledge. The incremental method presented here improves both theoretical bounds and practical performance, as it avoids the combinatorial explosion typical of naïve global tableau constructions. Future directions include extending the procedure to logics with additional group modalities (e.g., shared knowledge, coalition operators), handling dynamic agent sets, and implementing the algorithm in a prototype tool for empirical evaluation on realistic multi‑agent scenarios.

In summary, the authors deliver a sound, complete, and deterministic exponential‑time tableau decision procedure for MAEL(CD), thereby closing the complexity gap for this expressive epistemic logic and providing a foundation for efficient automated reasoning in multi‑agent systems.