Unimodal Logic of Density

• Unimodal logic of density is a modal system defined via dense Kripke frames where every accessibility edge admits a midpoint, captured by the axiom □□p → □p.
• Selective filtration constructs finite models, leading to an EXPTIME upper bound for satisfiability while establishing PSPACE-hardness.
• The approach contrasts with bimodal weak-density systems, highlighting the structural costs of enforcing density in relational frameworks.

The unimodal logic of density is the propositional modal logic determined by dense Kripke frames, that is, frames $(W,R)$ satisfying

$$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$

In this setting, “density” means that every accessibility edge can be split through an intermediate world. The corresponding modal axiom is

$\square\square p \to \square p,$

and recent work gives a finite-model construction by selective filtration yielding an EXPTIME upper bound for satisfiability, while also establishing PSPACE-hardness (Balbiani et al., 15 Jul 2025).

1. Semantic basis

The language is the usual propositional modal language

$$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$

with $\lozenge\phi$ defined as $\neg\square\neg\phi$. A Kripke frame is a pair $(W,R)$, and a model is $(W,R,V)$ with valuation $V:At\to\wp(W)$. Truth is standard: $$s\models \square\phi \iff \forall t\,(sRt \Rightarrow t\models\phi), \qquad s\models \lozenge\phi \iff \exists t\,(sRt \text{ and } t\models\phi).$$ The distinctive semantic clause is the density condition on $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$0: every $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$1-edge admits an intermediate point. This makes the logic a frame-defined unimodal system rather than a metric, probabilistic, or statistical theory of density (Balbiani et al., 15 Jul 2025).

The frame condition is tightly matched to a modal axiom. The least modal logic containing $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$2 is exactly the set of formulas valid on all dense frames. The decision problem can therefore be formulated either as validity over dense frames or as satisfiability of the negation in dense frames. This equivalence is central to the complexity analysis developed for the logic (Balbiani et al., 15 Jul 2025).

2. Axiomatization and logical character

The axiom

$$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$3

is the defining principle of the unimodal logic of density. Semantically, it encodes the possibility of refining each accessibility step by a midpoint; proof-theoretically, it identifies the least modal logic valid on dense frames. The paper on modal density treats this as the core unimodal case and contrasts it with a separate bimodal logic of weak density, whose frame condition is

$$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$4

and whose characteristic axiom is

$$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$5

The bimodal system is therefore a nearby but distinct formalism rather than an alternative axiomatization of unimodal density (Balbiani et al., 15 Jul 2025).

A useful conceptual point is that the density logic is not introduced through algebraic or topological shape constraints on numerical densities. Its primitive semantics is relational. Worlds, accessibility, and subformula preservation are the relevant objects; “density” names a structural property of $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$6, not a property of a probability density or a unimodal function. This distinction matters because the same words occur in several unrelated arXiv literatures.

3. Decision problem and complexity

The principal computational question is: given a formula $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$7, determine whether $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$8 is valid on all dense frames, or equivalently whether $$\forall s,t\in W\;\bigl(sRt \rightarrow \exists u\in W\,(sRu \wedge uRt)\bigr).$$9 is satisfiable in some dense frame. The published upper bound is:

The satisfiability problem for the unimodal logic of density is in EXPTIME (Balbiani et al., 15 Jul 2025).

The same work also shows PSPACE-hardness. Accordingly, the exact complexity is not fixed by the reported results, but the problem is at least PSPACE-hard and has an EXPTIME upper bound. The upper bound is nontrivial because ordinary filtration only yields an easier coNEXPTIME upper bound for validity: the least filtration of a dense model remains dense, but that alone does not exploit the density condition efficiently enough for the stronger analysis (Balbiani et al., 15 Jul 2025).

The paper further contrasts the unimodal situation with the bimodal weak-density case. There, a tableau-like method gives PSPACE-completeness. The stated contrast is that unimodal density is handled by selective filtration, bimodal weak density by a tableau/window construction, and the unimodal case lands in EXPTIME while the bimodal weak-density case is tighter at PSPACE. The paper explicitly interprets this as showing that the unimodal density condition is structurally more expensive to enforce via finite-model construction (Balbiani et al., 15 Jul 2025).

4. Selective filtration

The EXPTIME upper bound is obtained by a finite construction built from subformulas of the input formula. Let $\square\square p \to \square p,$0 be the set of subformulas of $\square\square p \to \square p,$1, with $\square\square p \to \square p,$2, ordered compatibly with syntactic dependency. A $\square\square p \to \square p,$3-tip is a bit-vector

$\square\square p \to \square p,$4

satisfying the Boolean constraints forced by the syntax: $\square\square p \to \square p,$5 is coded by $\square\square p \to \square p,$6, negation by complement, and disjunction by coordinatewise maximum. For a prime theory $\square\square p \to \square p,$7 in the canonical model, the associated vector $\square\square p \to \square p,$8 is a $\square\square p \to \square p,$9-tip (Balbiani et al., 15 Jul 2025).

The initial finite structure $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$0 is defined by taking $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$1 to be the set of all $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$2-tips and setting

$$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$3

iff whenever $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$4 and $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$5, then $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$6. The paper then works with $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$7-clips, finite substructures preserving the canonical tips and the canonical accessibility patterns relevant to $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$8. Since there are at most $$\phi ::= p \mid \bot \mid \neg\phi \mid (\phi\wedge\phi)\mid \square\phi,$$9 tips, the ambient family of such structures is finite (Balbiani et al., 15 Jul 2025).

The key technical move is the pruning operator $\lozenge\phi$0. For a clip $\lozenge\phi$1, it produces $\lozenge\phi$2 where

$\lozenge\phi$3

and

$\lozenge\phi$4

Thus a world is retained only if every negative box has a successor witnessing failure of the boxed formula, and an edge is retained only if it already has a midpoint. Iterating $\lozenge\phi$5 must stabilize because the family of clips is finite and each application shrinks or preserves the current structure. At the fixed point $\lozenge\phi$6, the frame is dense by construction: any remaining edge has a two-step witness, and an edge without such a witness would be deleted (Balbiani et al., 15 Jul 2025).

Truth of subformulas is then recovered by defining a valuation coordinatewise on the fixed-point structure and proving the expected truth lemma: $\lozenge\phi$7 Because the initial structure has at most $\lozenge\phi$8 states and each pruning step is polynomial in the size of the current clip, the fixed point can be computed in exponential time. This is the basis of the EXPTIME upper bound (Balbiani et al., 15 Jul 2025).

5. Related density formalisms in logic

The phrase “logic of density” also appears in other parts of logic, but usually with a different technical meaning. Two cases are especially relevant.

Setting	Density notion	Main result
Unimodal modal logic	Every $\lozenge\phi$9-edge has a midpoint	satisfiability in EXPTIME (Balbiani et al., 15 Jul 2025)
Semilinear substructural logics	admissibility of a density rule	density elimination for $\neg\square\neg\phi$0 (Wang, 2015)
Natural-density quotient structures	$\neg\square\neg\phi$1-additivity of density on $\neg\square\neg\phi$2	extension theorem for natural density (Talponen, 2015)

In semilinear substructural logics, density is proof-theoretic rather than relational. The central object is the admissibility of the strong density rule

$\neg\square\neg\phi$3

with $\neg\square\neg\phi$4 fresh. A uniform hypersequent method proves density elimination for $\neg\square\neg\phi$5, and for $\neg\square\neg\phi$6 this yields standard completeness via the cited result of Metcalfe and Montagna (Wang, 2015). Despite the common terminology, this is not the unimodal logic of dense Kripke frames.

A different adjacent line of work studies natural density modulo null-density sets. If $\neg\square\neg\phi$7 is the family of subsets of $\neg\square\neg\phi$8 with natural density, $\neg\square\neg\phi$9 the null-density sets, and $(W,R)$0 iff $(W,R)$1, then one can extend suitable Boolean substructures of $(W,R)$2 to a $(W,R)$3-algebra $(W,R)$4 on which density is $(W,R)$5-additive (Talponen, 2015). This provides a measure-theoretic semantics for density, but not a modal logic of dense frames.

6. Terminological boundaries and other uses of “unimodal density”

A recurrent source of confusion is that “unimodal” and “density” are also central terms in statistics, topological data analysis, and probability, where they mean something entirely different from dense accessibility relations.

In topological statistics, a function $(W,R)$6 is called unimodal when every positive super-level set $(W,R)$7 is contractible or empty, and the unimodal category counts the smallest number of unimodal summands in a decomposition of $(W,R)$8 (K et al., 7 Oct 2025). On finite metric trees this decomposition is constructively computable by a greedy sweeping algorithm (Baryshnikov et al., 2018), whereas on general graphs it is NP-hard even for fixed $(W,R)$9, with additional hardness for planar restrictions, inapproximability, and higher-dimensional generalizations (K et al., 7 Oct 2025). This is a theory of topological modes of functions, not a modal logic.

In probability theory, isotropic unimodal Lévy processes are processes whose one-dimensional distributions have radial, radially nonincreasing densities. For such processes, sharp two-sided estimates for the transition density $(W,R,V)$0, the Lévy density $(W,R,V)$1, and the tail function are governed by weak scaling of the Lévy–Khintchine exponent $(W,R,V)$2, with the canonical estimate

$(W,R,V)$3

under the stated scaling hypotheses (Bogdan et al., 2013). Related work obtains two-sided Dirichlet heat kernel estimates in $(W,R,V)$4 domains for pure jump isotropic unimodal Lévy processes even when the intensity of small jumps is low (Cho et al., 2019). Again, “unimodal density” here concerns radial monotonicity of probability densities, not modal operators.

In Bayesian deconvolution, a symmetric unimodal latent density is represented as a mixture of symmetric uniforms,

$(W,R,V)$5

and the mixing density $(W,R,V)$6 is modeled by a Dirichlet process location-mixture of Gamma distributions (Su et al., 2020). The resulting method enforces symmetry and unimodality of a latent density under measurement error. This literature is about shape-constrained density estimation rather than the semantics or complexity of modal logics.

The unimodal logic of density is therefore best understood as a specific modal system associated with the frame condition that every edge has a midpoint. Its central contemporary results concern axiomatization by $(W,R,V)$7, selective-filtration model construction, and the placement of satisfiability between PSPACE-hardness and an EXPTIME upper bound (Balbiani et al., 15 Jul 2025).