Fine's Conjecture: Modal Logic & Beyond

• Fine's Conjecture is a multifaceted topic originally in modal logic, questioning whether every canonical normal modal logic is first-order complete.
• It extends into temporal Leggett–Garg theory, analytic number theory, and polyhedral adjunction, with each field adapting the concept to its own framework.
• Diverse methodologies—including ultrapower constructions, CHSH inequality analyses, Ramanujan sum evaluations, and computational lattice classification—underscore its broad impact.

Fine's Conjecture is not a single universally fixed statement. In modal logic it denotes Fine’s converse canonicity question: whether every canonical normal modal logic is first-order complete (Goldblatt, 2016). In several neighboring literatures the name is reused more loosely: in temporal Leggett–Garg theory it denotes a claim now established as a theorem (Halliwell et al., 2019), in Fine polyhedral adjunction theory it denotes a Fujita-type finiteness statement that is also already proved (Mora et al., 6 Jan 2026), while in analytic number theory the associated problem is usually called Fine’s query rather than a conjecture (Patkowski, 2020). In quantum foundations, the central established result is Fine’s theorem on joint distributions and CHSH/Bell inequalities, and the cited papers explicitly state that they do not formulate a separate conjecture under that name (Halliwell et al., 2012, Kunjwal, 2014).

1. Modal logic: the original formulation

In its historically central sense, Fine’s Conjecture asks for the converse of Fine’s Canonicity Theorem. The ambient setting is that of normal modal logics: a logic $L$ contains all propositional tautologies and all instances of

$\Box(A \to B) \to (\Box A \to \Box B),$

and is closed under modus ponens, uniform substitution, and necessitation. A logic is first-order complete if there is an elementary class $K$ of Kripke frames such that $L = \mathrm{Log}(K)$. It is canonical if, for every infinite cardinal $\kappa$, its $\kappa$-canonical frame validates $L$ (Goldblatt, 2016).

Fine’s Canonicity Theorem states that first-order completeness implies canonicity. Fine’s Conjecture asked whether the converse holds: if $L$ is canonical, must $L$ be first-order complete? The conjecture was plausible because many natural modal systems are both elementarily determined and canonical. The paper identifies, as standard examples, $K$ on all frames, $\Box(A \to B) \to (\Box A \to \Box B),$0 on transitive frames, $\Box(A \to B) \to (\Box A \to \Box B),$1 on preorders, $\Box(A \to B) \to (\Box A \to \Box B),$2 on equivalence relations or universal relations, and $\Box(A \to B) \to (\Box A \to \Box B),$3 on linear orders (Goldblatt, 2016).

A concise way to situate the main usages of the expression is the following.

Domain	Statement associated with “Fine’s Conjecture”	Status in cited work
Modal logic	Canonicity $\Box(A \to B) \to (\Box A \to \Box B),$4 first-order completeness	False in general (Goldblatt, 2016)
Temporal LG theory	Augmented LG inequalities are necessary and sufficient for a joint $\Box(A \to B) \to (\Box A \to \Box B),$5	Theorem, not conjecture (Halliwell et al., 2019)
Fine polyhedral adjunction	$\Box(A \to B) \to (\Box A \to \Box B),$6 is finite for fixed $\Box(A \to B) \to (\Box A \to \Box B),$7	Proven (Mora et al., 6 Jan 2026)
Analytic number theory	Existence of continuous $\Box(A \to B) \to (\Box A \to \Box B),$8-periodic $\Box(A \to B) \to (\Box A \to \Box B),$9 with $K$0	Posed as Fine’s query (Patkowski, 2020)

Within modal logic, the conjecture concerns the relation between semantic definability by first-order frame conditions and algebraic/canonical validity. That question became a central organizing problem because Fine’s theorem had explained why first-order frame conditions were so effective in completeness and canonicity arguments, while the conjecture asked whether this success was in fact exhaustive (Goldblatt, 2016).

2. Refutation, ultrapowers, and the exact point of failure

The conjecture is false in general. Goldblatt–Hodkinson–Venema showed that there are uncountably many canonical modal logics that are not first-order complete, so canonicity and first-order completeness do not coincide (Goldblatt, 2016).

The structural explanation given in Goldblatt’s analysis is that the two properties are controlled by different ultrapower constructions. Let $K$1 be the countably generated free algebra in the variety validating $K$2, and let $K$3 denote canonical frame formation. Then canonicity is equivalent to validity of $K$4 in all frames $K$5, where $K$6 ranges over ultrafilters. By contrast, first-order completeness is equivalent to validity of $K$7 in all ultrapowers $K$8 (Goldblatt, 2016).

The crucial point is that canonical frame formation does not commute with ultrapowers. In general,

$K$9

Goldblatt makes this failure concrete by showing that if $L = \mathrm{Log}(K)$0 is a nonprincipal ultrafilter on a countable index set and $L = \mathrm{Log}(K)$1, then

$L = \mathrm{Log}(K)$2

so the two structures can differ drastically even in cardinality (Goldblatt, 2016).

There is always a natural injective map

$L = \mathrm{Log}(K)$3

which makes $L = \mathrm{Log}(K)$4 isomorphic to a subframe of $L = \mathrm{Log}(K)$5. However, the subframe need not be generated. Since modal validity transfers cleanly along generated subframes but not arbitrary subframes, canonicity of $L = \mathrm{Log}(K)$6 on all $L = \mathrm{Log}(K)$7 does not force first-order completeness (Goldblatt, 2016).

The failure is not universal. For subframe logics, validity is closed under all subframes, not merely generated ones. In that case the injection above becomes sufficient to transfer validity downward, and every canonical subframe logic is first-order complete. This yields a positive resolution of Fine’s conjecture for that subclass (Goldblatt, 2016).

The same paper also isolates residual open problems. One is whether validity of $L = \mathrm{Log}(K)$8 in the canonical frame built from a countable language forces validity in canonical frames built from larger languages. Another is whether $L = \mathrm{Log}(K)$9 and $\kappa$0 are elementarily equivalent for all infinite $\kappa$1; the paper states that this remains unknown, although they are equivalent for the quasi-modal fragment (Goldblatt, 2016).

3. Fine’s theorem in quantum foundations and conjectural overextensions

In quantum foundations, the central object is Fine’s theorem rather than Fine’s conjecture. In the CHSH scenario, one is given four pairwise distributions $\kappa$2, $\kappa$3, $\kappa$4, and $\kappa$5 for dichotomic observables $\kappa$6, together with consistent one-site marginals. Fine’s theorem states that there exists a non-negative joint distribution $\kappa$7 reproducing those four marginals if and only if all eight CHSH inequalities hold (Halliwell, 2014).

The theorem is exact: the CHSH inequalities are not merely necessary but sufficient. Halliwell’s analysis of quasi-probabilities uses this fact to distinguish viable quasi-probabilities, whose positive marginals can be matched by some genuine probability distribution, from non-viable quasi-probabilities, whose marginals admit no such extension. In the CHSH case, viability is therefore reducible to CHSH/Bell-type feasibility via Fine’s theorem (Halliwell et al., 2012).

The same paper explicitly states that it does not introduce or rely on a separate “Fine’s Conjecture.” What it does raise is a speculative question: whether there exists a general ordered-projector quasi-probability

$\kappa$8

that is positive if and only if the CHSH inequalities hold. The paper states that the authors “suspect there is no such general formula,” because the existence of a matching joint probability when CHSH holds does not determine a unique quasi-probability; rather, one typically obtains a family parameterized by unfixed higher-order correlators. The EPRB example in the paper provides counterevidence to such an identification: there are regions where CHSH holds and hence a non-negative joint $\kappa$9 exists, but the natural component $\kappa$0 is negative and linear positivity fails (Halliwell et al., 2012).

In contextuality theory, a related but distinct overextension also fails. For Specker’s scenario, the paper on noncontextuality proves that within the ambit of Fine’s theorem the following are equivalent: existence of a joint distribution, existence of a measurement-noncontextual and outcome-deterministic model, and existence of a measurement-noncontextual and factorizable model. Outside that ambit, if factorizability is dropped, the equivalence collapses. The paper therefore rejects a natural conjectural extension according to which general noncontextuality would still be equivalent to existence of a joint distribution; Liang–Spekkens–Wiseman inequalities are needed precisely because non-factorizable measurement-noncontextual models need not admit any joint distribution (Kunjwal, 2014).

A recurrent misconception in this area is thus that Fine’s theorem extends unchanged from CHSH-type compatibility structures to broader notions of contextuality or quasi-probability positivity. The cited papers treat this as false: Fine’s theorem is sharp in its own domain, but its hypotheses, especially consistency and factorizability, are not dispensable (Halliwell et al., 2012, Kunjwal, 2014).

4. Temporal Leggett–Garg usage: from conjecture to theorem

In the Leggett–Garg literature, “Fine’s conjecture” is a conventional but technically misleading label. The setting is a single dichotomic variable $\kappa$1 measured at times $\kappa$2, with one-time means $\kappa$3 and two-time correlators $\kappa$4. The physical background is macrorealism per se, non-invasive measurability, arrow of time, and no-signaling-in-time (NSIT) (Halliwell et al., 2019).

The paper proves that, for the standard cycle geometry in which only the pairs $\kappa$5 are specified, an underlying joint distribution $\kappa$6 exists if and only if an augmented set of conditions holds. These conditions consist of the two-time Leggett–Garg inequalities, which are exactly the non-negativity constraints

$\kappa$7

together with the $\kappa$8-time cycle inequalities

$\kappa$9

for all choices $L$0 with $L$1. If one works with sequential data, NSIT/AoT compatibility conditions must also be imposed; if one reconstructs pair probabilities from moments,

$L$2

compatibility is automatic (Halliwell et al., 2019).

This is explicitly presented as a theorem, not merely a conjecture. The paper gives a constructive proof by a generalized Fine ansatz,

$L$3

combined with interval constraints ensuring that the intermediate correlator $L$4 can always be chosen so that all required marginals remain non-negative (Halliwell et al., 2019).

The paper also extends the discussion beyond the cycle case. If all pair correlators are measured, the feasible region is governed by the cut polytope $L$5, and additional odd-cycle or hypermetric constraints appear. The $L$6-gon inequalities

$L$7

become relevant. The cited work proves sufficiency of the natural package consisting of all two-time non-negativities, all three-time LG inequalities, and all $L$8-gon inequalities in certain symmetric cases for $L$9, and reports numerical support more generally, while leaving the full nonsymmetric characterization open (Halliwell et al., 2019).

The temporal usage therefore differs sharply from the modal-logical one. In modal logic, Fine’s Conjecture is false. In Leggett–Garg theory, the phrase refers to a statement that the paper formulates as a theorem once compatibility and positivity are augmented correctly (Halliwell et al., 2019).

5. Fine’s query in analytic number theory

The number-theoretic problem discussed in the cited literature is usually called Fine’s query. Fine asked for an explicit continuous $L$0-periodic function $L$1 on $L$2 satisfying a symmetry condition and the vanishing sum property

$L$3

The paper gives new solutions and generalizations of this problem (Patkowski, 2020).

One broad class arises from multiplicative arithmetic functions $L$4 satisfying the Abel-limit condition

$L$5

For each positive integer $L$6, the functions

$L$7

are continuous, $L$8-periodic, even, and satisfy Fine’s vanishing sum condition. The paper lists $L$9 and $L$0 as examples, using the Abel-limit vanishing of $L$1 and $L$2 (Patkowski, 2020).

A second family introduces indicator-type weights $L$3 and imposes the coprimality restriction $L$4. Under hypotheses on a completely multiplicative $L$5 and the value $L$6, the paper constructs functions $L$7 and $L$8 that lie in the corresponding restricted class and again satisfy the rational-point vanishing property for all $L$9 coprime to $K$0 (Patkowski, 2020).

The proofs are driven by Ramanujan-sum evaluations such as

$K$1

which collapse the sum over $K$2, after which the Abel-limit vanishing of the associated Dirichlet series forces the result (Patkowski, 2020).

The paper places this query in a broader analytic framework by connecting Davenport expansions to Popov’s formula. For an arithmetic function $K$3 with $K$4, it proves a general identity relating weighted fractional-part sums to a Davenport-type cosine series with coefficients $K$5, and recovers Popov’s sum when $K$6. This connection supplies asymptotic information through Mellin transforms, Mellin–Perron inversion, and residue calculations, but it is ancillary to the resolution of Fine’s query itself (Patkowski, 2020).

6. Fine polyhedral adjunction theory and the Fine spectrum

In Fine polyhedral adjunction theory, Fine’s Conjecture is a finiteness statement for a spectral invariant of lattice polytopes. If $K$7 is a $K$8-dimensional rational polytope and $K$9 is its support function, the Fine adjoint polytope is

$\Box(A \to B) \to (\Box A \to \Box B),$00

The Fine $\Box(A \to B) \to (\Box A \to \Box B),$01-codegree is

$\Box(A \to B) \to (\Box A \to \Box B),$02

and the Fine spectrum in dimension $\Box(A \to B) \to (\Box A \to \Box B),$03 is

$\Box(A \to B) \to (\Box A \to \Box B),$04

Fine’s Conjecture asserts that for fixed $\Box(A \to B) \to (\Box A \to \Box B),$05 and $\Box(A \to B) \to (\Box A \to \Box B),$06, the set $\Box(A \to B) \to (\Box A \to \Box B),$07 is finite (Mora et al., 6 Jan 2026).

The cited paper treats that finiteness statement as established, citing Garzón Mora–Haase for the proof. Its own contribution is structural classification. The main uniform theorem classifies the top of the spectrum: for a lattice $\Box(A \to B) \to (\Box A \to \Box B),$08-polytope with $\Box(A \to B) \to (\Box A \to \Box B),$09, one has $\Box(A \to B) \to (\Box A \to \Box B),$10 if and only if $\Box(A \to B) \to (\Box A \to \Box B),$11 is unimodularly equivalent to the standard simplex $\Box(A \to B) \to (\Box A \to \Box B),$12, there are no spectrum values in $\Box(A \to B) \to (\Box A \to \Box B),$13, one has $\Box(A \to B) \to (\Box A \to \Box B),$14 if and only if $\Box(A \to B) \to (\Box A \to \Box B),$15 projects onto $\Box(A \to B) \to (\Box A \to \Box B),$16, one has $\Box(A \to B) \to (\Box A \to \Box B),$17 if and only if $\Box(A \to B) \to (\Box A \to \Box B),$18 is an exceptional simplex, and one has $\Box(A \to B) \to (\Box A \to \Box B),$19 if $\Box(A \to B) \to (\Box A \to \Box B),$20 projects to $\Box(A \to B) \to (\Box A \to \Box B),$21 and none of the previous cases holds. No other values occur in $\Box(A \to B) \to (\Box A \to \Box B),$22 (Mora et al., 6 Jan 2026).

The paper also gives complete low-dimensional spectra. In dimension $\Box(A \to B) \to (\Box A \to \Box B),$23,

$\Box(A \to B) \to (\Box A \to \Box B),$24

because a segment $\Box(A \to B) \to (\Box A \to \Box B),$25 satisfies $\Box(A \to B) \to (\Box A \to \Box B),$26 exactly when $\Box(A \to B) \to (\Box A \to \Box B),$27. In dimension $\Box(A \to B) \to (\Box A \to \Box B),$28,

$\Box(A \to B) \to (\Box A \to \Box B),$29

The standard triangle $\Box(A \to B) \to (\Box A \to \Box B),$30 has $\Box(A \to B) \to (\Box A \to \Box B),$31, while the unit square has $\Box(A \to B) \to (\Box A \to \Box B),$32; homogeneity then generates the full two-dimensional list (Mora et al., 6 Jan 2026).

Dimension $\Box(A \to B) \to (\Box A \to \Box B),$33 is described as almost complete. The paper derives a restricted numerator set

$\Box(A \to B) \to (\Box A \to \Box B),$34

then proves that $\Box(A \to B) \to (\Box A \to \Box B),$35 does not occur, and supplies explicit witnesses for $\Box(A \to B) \to (\Box A \to \Box B),$36, $\Box(A \to B) \to (\Box A \to \Box B),$37, $\Box(A \to B) \to (\Box A \to \Box B),$38, $\Box(A \to B) \to (\Box A \to \Box B),$39, $\Box(A \to B) \to (\Box A \to \Box B),$40, $\Box(A \to B) \to (\Box A \to \Box B),$41, and $\Box(A \to B) \to (\Box A \to \Box B),$42. The full denominator sets for several of these numerators remain open (Mora et al., 6 Jan 2026).

Methodologically, the paper develops a Fine mountain construction, determinant formulas for numerators, reduction lemmas based on positive linear relations among core normals, and mixed-integer linear programming searches implemented in polymake and SCIP. This computational framework is presented as a general approach to higher-dimensional classification, although exhaustive enumeration becomes difficult as $\Box(A \to B) \to (\Box A \to \Box B),$43 grows (Mora et al., 6 Jan 2026).

Taken together, these results show that the polyhedral “Fine’s Conjecture” is no longer conjectural in its finiteness aspect. What remains active are classification problems below the top range, the exact next gap below $\Box(A \to B) \to (\Box A \to \Box B),$44, the global accumulation behavior of the shifted spectrum, and exhaustive higher-dimensional descriptions (Mora et al., 6 Jan 2026).