Euclidean Modal Logics

Updated 8 July 2026

Euclidean modal logics are defined as normal extensions of K5, characterized by axiom 5 and its Euclidean frame condition which ensures that if a world relates to two others, then those worlds relate to each other.
They interact with standard frame conditions such as reflexivity, transitivity, seriality, and symmetry, often culminating in systems like S5 where accessibility becomes an equivalence relation.
Recent research in this area spans epistemic settings, proof theory, distributed knowledge, and spatial semantics, highlighting both computational complexities and novel model-theoretic applications.

Searching arXiv for papers on Euclidean modal logics and closely related topics. Euclidean modal logics are normal modal logics extending $K5$ , the logic of axiom $5$, whose Kripke frames satisfy the Euclidean first-order condition

$\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$

In the standard correspondence theory, axiom $5$ is

$\Diamond p \rightarrow \Box \Diamond p,$

and in epistemic notation it yields negative introspection,

$\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$

The family includes $K5$ , $D5$ , $T5$ , $K45$ , $5$0, and $5$1. In parallel, some recent work uses “Euclidean” for modal languages interpreted over Euclidean spaces $5$2; that semantic usage is distinct from the Euclidean frame condition on accessibility relations (Balbiani et al., 14 Aug 2025, Wang et al., 30 Mar 2026, Bezhanishvili et al., 2024).

1. Euclideanity, axiom $5$3, and frame correspondence

A Kripke frame is a pair $5$4 with non-empty $5$5 and a binary relation $5$6; a Kripke model adds a valuation $5$7. Satisfaction for $5$8 and $5$9 is standard: $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 0

$\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 1

Within this setting, axiom $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 2 characterizes the class of Euclidean frames, and $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 3 consists of all theorems of $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 4 plus axiom $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 5 (Balbiani et al., 14 Aug 2025).

Euclideanity interacts in a rigid way with other standard frame conditions. Reflexivity is expressed by $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 6, transitivity by $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 7, seriality by $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 8, and symmetry by $\forall x \forall y \forall z \bigl((xRy \wedge xRz) \rightarrow yRz\bigr).$ 9. The cited literature states that reflexive plus Euclidean implies symmetry and transitivity; in particular, if $5$0 is reflexive and Euclidean, then $5$1 is an equivalence relation, which is the characteristic $5$2-behavior (Wang et al., 30 Mar 2026). A mechanized presentation in PVS gives the corresponding lemmas ref_Euc1, ref_Euc2, and sym_Euc, and also formalizes the soundness direction

$5$3

for the modal axiom $5$4 (Rushby, 2022).

The same correspondence extends to multi-agent epistemic settings. With agents $5$5, formulas may include $5$6 and distributed knowledge $5$7, where $5$8 is nonempty. Distributed knowledge is interpreted by the intersection relation

$5$9

If each $\Diamond p \rightarrow \Box \Diamond p,$ 0 is Euclidean, then $\Diamond p \rightarrow \Box \Diamond p,$ 1 is Euclidean as well (Wang et al., 30 Mar 2026).

A Euclidean modal logic is, by definition in the cited work, a normal modal logic $\Diamond p \rightarrow \Box \Diamond p,$ 2 that is an extension of $\Diamond p \rightarrow \Box \Diamond p,$ 3. The standard examples form a familiar hierarchy of frame restrictions (Balbiani et al., 14 Aug 2025).

System	Frame conditions	Modal presentation in the cited literature
$\Diamond p \rightarrow \Box \Diamond p,$ 4	Euclidean	$\Diamond p \rightarrow \Box \Diamond p,$ 5
$\Diamond p \rightarrow \Box \Diamond p,$ 6	Euclidean + serial	$\Diamond p \rightarrow \Box \Diamond p,$ 7
$\Diamond p \rightarrow \Box \Diamond p,$ 8	Euclidean + reflexive	$\Diamond p \rightarrow \Box \Diamond p,$ 9
$\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 0	transitive + Euclidean	validates $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 1 and $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 2
$\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 3	serial + transitive + Euclidean	belief semantics
$\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 4	reflexive + transitive + Euclidean	equivalence relations

The role of $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 5 is especially central. In the cited papers, $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 6 is presented both as $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 7 and as $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 8. In frame terms, once reflexivity and Euclideanity are present, symmetry and transitivity follow, so each accessibility relation becomes an equivalence relation (Balbiani et al., 14 Aug 2025, Wang et al., 30 Mar 2026).

For epistemic and doxastic applications, the Euclidean axiom is usually read as an introspection principle. In the distributed-knowledge setting, $\neg K_i \varphi \rightarrow K_i \neg K_i \varphi.$ 9 has each $K5$ 0 transitive and Euclidean; $K5$ 1 has each $K5$ 2 serial, transitive, and Euclidean; and $K5$ 3 has each $K5$ 4 reflexive, transitive, and Euclidean, with $K5$ 5 interpreted by the intersection of the agents’ relations (Wang et al., 30 Mar 2026).

The recent literature also studies Euclidean logics with additional algebraic or probabilistic structure. One paper treats $K5$ 6, the symmetric Euclidean case, in the analysis of almost sure validities on finite frames; another treats Euclideanity inside crisp Gödel modal logic, where the usual Boolean interdefinability of $K5$ 7 and $K5$ 8 is unavailable (Sliusarev, 2024, Rodriguez et al., 2020).

3. Semantics, proof theory, and nonclassical variants

Recent work on proof-theoretic semantics develops base-extension semantics for the classical propositional modal systems $K5$ 9, $D5$ 0, $D5$ 1, and $D5$ 2, with $D5$ 3 as the primary modal operator. In that framework, a base is a countable collection of atomic inference rules; validity is given by an inductive definition generated by provability in such a base. The paper proves appropriate soundness and completeness theorems and establishes the duality between $D5$ 4 and a natural presentation of $D5$ 5. It also proves that the semantics, in its current form, is not complete with respect to Euclidean modal logics, and adds that the formulation makes essential use of relational structures on bases (Eckhardt et al., 2024).

Mechanized correspondence theory gives a different perspective on the same terrain. The PVS embedding of propositional and quantified modal logic defines the Euclidean property by

$D5$ 6

includes five as the schema $D5$ 7, and derives the relational implications from Euclideanity. The report also presents countermodel recipes for failed correspondences and formalizes quantified variants such as constant-domain and varying-domain Barcan formulas, while observing that Euclideanity itself does not drive the Barcan correspondences (Rushby, 2022).

In crisp Gödel modal logic, Euclideanity requires a genuinely bimodal treatment. The cited paper axiomatizes Euclidean frames by two schemes,

$D5$ 8

It states that, unlike the classical Boolean setting, $D5$ 9 and $T5$ 0 are not interdefinable in $T5$ 1; consequently, the Euclidean condition must be captured by both schemes. The paper then proves that $T5$ 2 extended by $T5$ 3 is strongly complete for local consequence over crisp Euclidean frames (Rodriguez et al., 2020).

These results collectively show that Euclideanity is stable across several semantic regimes, but its technical expression depends strongly on the underlying proof theory and truth-value algebra. This suggests that axiom $T5$ 4 is best viewed as a correspondence principle whose exact proof-theoretic realization varies with the semantic setting.

4. Euclidean distributed knowledge, forgetting, and uniform interpolation

In distributed-knowledge modal logics, formulas are generated by

$T5$ 5

with $T5$ 6 and nonempty $T5$ 7. The modal depth $T5$ 8 is the maximal nesting of modal operators. Satisfaction for $T5$ 9 is standard, while distributed knowledge uses the intersection semantics

$K45$ 0

The note $K45$ 1 coincides with $K45$ 2 under this semantics is explicit in the cited work (Wang et al., 30 Mar 2026).

The paper proves the uniform interpolation property for six distributed-knowledge modal logics: $K45$ 3 Uniform interpolation is defined so that, for atoms $K45$ 4 to omit, a formula $K45$ 5 is the strongest consequence of $K45$ 6 over $K45$ 7. The paper interprets forgetting an atom $K45$ 8 by bisimulation-quantifier semantics and shows that the result of forgetting coincides with a uniform interpolant $K45$ 9 (Wang et al., 30 Mar 2026).

The technical core is a refinement of canonical formulas and literal elimination for Euclidean systems. In the non-Euclidean systems $5$00, $5$01, and $5$02, the paper proves

$5$03

for satisfiable canonical $5$04. In the Euclidean systems $5$05, $5$06, and $5$07, literal elimination may fail, and the paper instead proves the depth-bound construction

$5$08

which yields $5$09 when $5$10. The paper explains that this bound captures the cost of ensuring collective $5$11-bisimulation against Euclidean-transitive behavior (Wang et al., 30 Mar 2026).

The Euclidean obstruction is explicit. Counterexamples are given, even in $5$12 with $5$13 and modal depth at least $5$14, where the Back condition of collective $5$15-bisimulation fails for $5$16. Two positive special cases are isolated: if $5$17, then $5$18 for the Euclidean systems; and for the two-agent Euclidean systems $5$19, $5$20, and $5$21, the equality $5$22 holds for all satisfiable canonical $5$23 (Wang et al., 30 Mar 2026).

5. Complexity, definability, and asymptotic validity

For satisfiability, Euclideanity lies on the NP side of the universal-Horn dichotomy. The cited complexity classification treats Euclidean frames as the Horn clause

$5$24

written there as $5$25. It states that satisfiability over Euclidean frames, hence $5$26, is NP-complete and has the polynomial-size model property. The same source states that $5$27 is NP-complete by the result of Halpern and Rêgo that any normal modal logic including axiom $5$28 has NP-complete satisfiability, and it restates Ladner’s theorem that $5$29 satisfiability is NP-complete (0802.1884).

For modal definability, recent work gives a structural classification internal to Euclidean extensions of $5$30. Every Euclidean frame is decomposed into galaxies, and finite rooted test frames are given by flowers $5$31. For a Euclidean modal logic $5$32, the crucial set is

$5$33

The paper proves that the following are equivalent: decidability of the modal definability problem over $5$34, decidability of the correspondence problem over $5$35, decidability of the first-order theory $5$36, and finiteness of

$5$37

In the decidable case, all three problems are in EXPSPACE; otherwise they are undecidable. The same paper also gives a meta-decision result in NEXPTIME for formulas of the form $5$38 (Balbiani et al., 14 Aug 2025).

The asymptotic theory of finite Euclidean frames behaves differently. For the classes $5$39, $5$40, $5$41, and $5$42, the logic of almost sure validities coincides exactly with the base logic: $5$43 The same paper proves analogous coincidence results for the transitive non-branching logics $5$44 and $5$45, while explicitly noting that $5$46 and $5$47 are not treated in the main results (Sliusarev, 2024).

Taken together, these results show that Euclidean modal logics exhibit sharply different computational profiles depending on the metatheoretic problem. Satisfiability for standard Euclidean systems is NP-complete, modal definability across Euclidean extensions has an EXPSPACE/undecidable boundary controlled by flower validation, and almost sure validity in several finite-frame classes adds no principles beyond the underlying logic.

The term “Euclidean modal logic” is also used in semantic settings where “Euclidean” refers to the ambient space rather than to the relation $5$48. In the logic of polyhedral reachability, formulas are interpreted as piecewise-linear subsets of $5$49, $5$50 is interpreted as interior, and a binary reachability modality $5$51 is defined via continuous paths. The paper emphasizes that this sense of “Euclidean” is semantic and should be distinguished from the Euclidean property of Kripke frames; it proves soundness and completeness of $5$52 with respect to finite posets and polyhedral models (Bezhanishvili et al., 2024).

A related topological line studies the modal logic of planar polygons. There the closure algebra $5$53 is generated by polygons in $5$54, with $5$55 interpreted as interior and $5$56 as closure. The resulting logic $5$57 is finitely axiomatizable, complete with respect to crown frames, not first order definable, does not have the Craig interpolation property, and has PSPACE-complete satisfiability (Gabelaia et al., 2018).

Distance logics of Euclidean spaces provide yet another usage. For each $5$58, the cited work defines three relations on $5$59: $5$60 All three relations are symmetric, so $5$61 is valid for all three; $5$62 is reflexive, so $5$63 holds for $5$64; and neither transitivity nor Euclideanity in the modal sense holds for any of the three relations, so both $5$65 and $5$66 fail for all three. The paper further proves that the farness and nearness logics of $5$67 strictly contain those of $5$68, while the constant-distance logics agree, and that the farness logic of the reals is not finitely axiomatizable and does not have the finite model property (Agnew et al., 8 Jan 2025).

These spatial and metric developments do not alter the standard Kripke-theoretic meaning of Euclidean modal logic, but they do show that the adjective “Euclidean” now designates two distinct research programs: one centered on axiom $5$69 and Euclidean accessibility, the other on modal semantics over Euclidean spaces. A plausible implication is that current work increasingly treats the shared terminology as a point of comparison rather than of identification.