Belief Base Semantics

• Belief base semantics is a computational framework that replaces abstract possible worlds with finite, explicit belief bases for clear epistemic modeling.
• It supports both explicit and implicit belief modalities, enabling dynamic belief change and fine-grained reasoning in multi-agent settings.
• Practical applications include symbolic model checking, AGM-style revision, and graded distributed belief analysis in complex systems.

Belief base semantics is a foundational paradigm in epistemic logic and belief revision that dispenses with the traditional, primitive notions of possible worlds and accessibility relations. Instead, it reconstructs these concepts from more elementary data structures—typically finite (possibly prioritized or graded) sets of formulas specifying agents’ explicit beliefs. This approach provides a transparent, computationally grounded framework for modeling explicit and implicit belief, as well as belief-change operations, across a spectrum of multi-agent settings. Belief base semantics has been developed through a series of works, notably by Lorini and others, and now underpins advances in both static epistemic modeling and dynamic epistemic logic (Lorini, 2019, Lorini, 2018, Lorini et al., 27 Nov 2025, Souza et al., 2019).

1. Syntax and Core Semantic Concepts

A belief base is a finite set (possibly prioritized or graded) of formulas from a base logic such as propositional or first-order logic. For the multi-agent case, a multi-agent belief base (MBB) is a tuple $B = (B_1, \dots, B_n, s)$, where $B_i$ is the explicit belief base of agent $i$, $B_i \subseteq L^-$, and $s \subseteq PROP$ is the actual propositional state (Lorini, 2019, Lorini, 2018). The explicit-belief language is defined inductively:

$$\alpha ::= p \mid \neg\alpha \mid \alpha \wedge \alpha \mid \Delta_i \alpha,$$

where $p\in PROP$, $\Delta_i \alpha$ means agent $i$ explicitly believes $\alpha$. For implicit/epistemic modeling, the language is extended with modalities ($K_i$ for knowledge, $I_i$ for implicit belief), resulting in formulas like:

$$\phi ::= \alpha \mid \neg\phi \mid \phi\wedge\phi \mid K_i\phi,$$

interpreted over pairs $(B, C)$ with $B$ an MBB and $C$ a context (common ground) (Lorini, 2019).

Truth in $(B, C)$ is defined recursively, with $(B, C) \models K_i\phi$ iff, for all $B'\in C$ such that $B\to_i B'$ (i.e., $B'$ satisfies all explicit beliefs from $B_i$), $(B', C)\models \phi$.

2. Belief Bases versus Kripke-Style Semantics

Belief base semantics reconstructs possible worlds and accessibility/doxastic relations from explicit, finitary structures:

• Worlds are not abstract sets of atomic valuations, but syntactic tuples of belief bases and a propositional state.
• Accessibility (alternatives) is defined via belief satisfaction: $B \to_i B'$ iff all explicit beliefs in $B_i$ are satisfied in $B'$.
• Contexts $C$ represent the common ground and restrict admissible states (Lorini, 2019, Lorini, 2018).

In classical Kripke structures, both states and accessibility relations are primitive, with no constraint on the internal structure of possible worlds. In contrast, belief base models provide fine-grained control, allowing explicit reasoning about limited reasoning, non-closure under logical consequence, and inconsistent or partial bases. The semantic link $B_i \alpha \rightarrow I_i \alpha$ is a theorem in this setting, as explicit belief entails implicit belief by construction (Lorini, 2018).

3. Model-Theoretic and Computational Properties

Universal Model

The universal epistemic model is constructed without induction over belief levels. Let $C_\top$ be the set of all belief bases; then $(B, C_\top)$ serves as a universal model containing all finite belief hierarchies. This allows for a compact representation of the entire epistemic universe, with all accessibility relations and higher-order beliefs generated “on the fly” from explicit bases (Lorini, 2019).

Meta-Theoretic Results

Key results include:

• Soundness and completeness: Belief base logics (with explicit/implicit belief modalities) are sound and complete with respect to both belief-base semantics and their Kripke-style counterparts (Lorini, 2018).
• Finite model property: Satisfiability can always be witnessed in a finite belief-base model via standard filtration techniques (Lorini, 2018).
• PSPACE-completeness: The satisfiability and model checking problems for languages with explicit and implicit/only-belief modalities are PSPACE-complete (Lima et al., 2023).

4. Belief-Base Revision and Dynamic Epistemic Logic

Belief base semantics natively supports fine-grained belief change operations:

• AGM-style base revision: Bases are revised by new information $\mu$ by selecting maximal (or most credible) consistent subbases that accommodate $\mu$. Operators such as CSRG, CSRW, and CSIR leverage evidence-theoretic notions (credibility via Dempster-Shafer theory) to select consistent subbases (Ktari et al., 2020).
• Prioritized bases and graphs: Iterated belief revision is handled using priority graphs $(\Phi, \prec)$, where $\prec$ encodes base element priorities. These translate to preference relations over worlds, supporting classic postulates (AGM, Darwiche-Pearl) and directly inducing update dynamics in dynamic epistemic logic frameworks (Souza et al., 2019, Souza et al., 2019).
• Dynamic extensions: Operators for private belief expansion or public announcement are definable at the level of belief bases, with reduction axioms and explicit semantic effects on individual bases (Lima et al., 2023).

Model-based revision operators can be defined semantically via total preorders or min-friendly faithful assignments, as shown in (Falakh et al., 2021). For general logics and unions of bases, these operators are fully characterized by min-completeness, min-retractivity, and min-expressibility over the set of admissible bases.

5. Advanced Extensions: Graded and Distributed Belief

Belief base semantics is extendable to graded, collective, and distributed epistemic attitudes:

• Graded belief bases: Bases become multisets or functions $B_i: \text{Formulas} \to \mathbb{N}_\infty$ encoding the strength of explicit beliefs. Group belief is computed by merging bases (pointwise sum), and modal operators express distributed belief at various strength thresholds (Lorini et al., 27 Nov 2025).
• Distributed belief operators: Doxastic accessibility is parameterized by a threshold $k$ representing the tolerance for violating merged beliefs, yielding operators $D_J^k \varphi$ for “group $J$ distributively believes $\varphi$ with strength at least $k$" (Lorini et al., 27 Nov 2025).
• PSPACE-completeness and decision procedures: The associated logics maintain the finite model/property and admit tableau-based decision procedures or reductions to QBF for practical model checking.

6. Equivalence, Limitations, and Meta-Theoretical Analysis

Belief base semantics achieves notable equivalence and refinement properties:

• For individual and distributed epistemic languages (with explicit/implicit modalities), validities coincide with those of classic Kripke models, both in the finite and “universal” context. The universal context can make the logic strictly stronger for only-belief variants (Lorini, 2019).
• Limits are encountered in the representation of natural revision operators and certain iterated AGM postulates: not all can be captured by graph-based transformations due to the inability to uniquely “name” sets of minimal worlds using base elements (Souza et al., 2019, Souza et al., 2019).
• In logics with full Boolean expressivity (“disjunctive” logics), the entire class of AGM operators admits a total-preorder semantics on belief bases (Falakh et al., 2021).

7. Applications and Algorithmic Aspects

The belief base approach finds application in:

• Symbolic model checking of multi-agent epistemic properties, including only-believing and dynamic extensions, with QBF-based algorithms demonstrating practical efficiency for committee and voting protocols (Lima et al., 2023).
• Multi-agent epistemic frameworks that require explanations in terms of explicit assumptions, introspection, or awareness, as belief bases naturally restrict logical omniscience and provide a basis for resource-bounded reasoning (Lorini, 2018).
• Rational belief change in AI and knowledge representation, with explicit formal and computational connections to evidence theory, partial meet contraction, and preference orderings (Ktari et al., 2020, Falakh et al., 2021).

Belief base semantics thus constitute a foundational, expressive, and computationally tractable alternative to traditional Kripkean frameworks, anchoring epistemic logic, belief revision, and distributed reasoning in explicit, agent-centric structures.