Preprint
Article

This version is not peer-reviewed.

Every Natural Strengthening Examined Is False: Type Mismatch in Lossless Reformulations of the Lonely Runner Conjecture

Submitted:

14 November 2026

Posted:

17 November 2026

You are already at the latest version

Abstract
The Lonely Runner Conjecture (\( \mathrm{LRC}_n \)) admits many lossless reformulations—finite, geometric, dynamical, or arithmetic rewrites exactly equivalent to it. This paper reports a structural fact that emerged from a program of attacking the conjecture through such reformulations: every natural strengthening of a lossless reformulation that the program examined—every version that would open a new toolkit—is false. The claim is an observational summary of four witnessed settings, not a meta-theorem over all conceivable strengthenings: a reformulation equivalent to \( \mathrm{LRC} \) carries the full truth-content of \( \mathrm{LRC} \), and a strictly stronger version is false, with the falsity visible at small \( n \) through exact rational witnesses. We exhibit the type mismatch in four settings across two reformulation worlds. The finite pair-sum covering reformulation supports three: its proved inventory contains only closure, invariance, and exhaustiveness statements (no proved margin-type statement exists in it); its single margin-shaped instrument, a transfer-matrix model of the solution cascade, is closed by theorem in its single-anchor scope (the constrained transfer matrices are partial permutations of spectral radius exactly 0 or 1; the general branching strand-census operator is not covered); and its character-rigidity route has orbit-invariance laws falsified exactly, with marginal envelopes saturated precisely where the missing margin would be needed. The fourth setting is independent of the cascade: the Lonely Runner zonotope, whose covering radius satisfies a strictly stronger inequality than \( \mathrm{LRC} \) that is false from \( n \) = 5 on, with \( \rho=16/47>1/3 \) at \( v \) = (1, 2, 3, 4, 5) (certified from both sides by branch and bound), strict witnesses \( \rho \) ≥ 17/47 > 5/14 at \( v \) = (1, 2, 3, 4, 5, 12) , \( \rho \) ≥ 29/80, 4/11, 1793/5000 (all > 5/14), and a proved weight identity among the quotient norm's pair forms that forces the denominators of the exact values (the recurring 47 is \( 3\cdot4+3\cdot5+4\cdot5 \)); its Ehrhart \( h^* \)-polynomials are real-rooted and log-concave through \( n \) = 10, with strictness decaying like \( \Theta(1/n) \) and no correlation with covering margins at the achievable sample sizes. Alongside the negative result the paper states the program's one bounded positive theorem—the sum-only equality \( M(V)=\max_{\{u,w\}}\tau(u,w) \), the rigid-zone translate-class classifications at 7,11,13,17,19,37, and the composite-side structure at \( p^2 \) and \( pq \), with its hypothesis ledger—and a finite witness archive with a certification ladder: exact deepest-hole classification at \( n \) = 3, away-certified and branch-and-bound certificates at \( n \) = 4 (\( \rho \) = 9/34 at \( v \) = (1, 2, 3, 16)), exact refutations at \( n \) = 5,6 (the sharpest two-sided), and certified deepest-hole equality at \( n \) = 5, \( m \) = 10, 20, 25, 30 (\( \rho=\delta \) exactly). Every number is re-asserted by a persisted, independent, exact computation, and the certification method carries a negative control that correctly refuses a false statement. The paper is a negative-results paper and claims no progress on the conjecture itself: it proves no new cases of \( \mathrm{LRC} \), and the surviving equivalent forms are as hard as they ever were. Its contribution is the obstruction's anatomy—why four natural toolkits, each correct and each productive on its own terms, cannot cross the same gap.
Keywords: 
;  ;  ;  ;  ;  ;  
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.