Sharp continued-fraction bounds from exact digit runs

JD Jones

Gives a sharp global inverse bound for the maximum of three consecutive ordinary continued-fraction approximation coefficients. In natural-extension coordinates Ga(x,y)−cG_{a}(x,y)-c is at least (aβ/(a2+4))(a\beta /(a^{2}+4)) times the maximum distance of x or y from β\beta , for real a≥2a \ge 2, with β=((a2+4)−a)/2\beta =(\sqrt{(a^{2}+4)}-a)/2 and c=1/(a2+4)c=1/\sqrt{(a^{2}+4)}. The coefficient is also optimal over actual convergent triples for integer a≥2a \ge 2 with variable index.

What was already known

Tong’s classical maximum threshold and Adhin, Bakhtawar and Kraaikamp’s digit-sensitive minimum refinement and suggestion to extend it. Earlier geometry is credited.

AI involvement AI-centered

JD Jones’s pinned Disclosure.md explicitly credits GPT-6 Pro for proposing the maximum inverse estimate and optimal window cutoffs, with Codex formalization and checking.

Importance Modest

The sharp stability estimate refines the three-convergent maximum under the stated digit restrictions.

Technical verification Formal (Lean)

Project reports Lean kernel checks of the exact inverse and actual-convergent optimality results. Exact finite arithmetic replays are separate.

Formal checking

Project reports Lean kernel checks of the exact inverse and actual-convergent optimality results. Exact finite arithmetic replays are separate.

Checker
JD Jones’s project via Lean 4
Scope
Global maximum inverse bound for real a≥2a \ge 2 and optimal coefficient over actual convergent triples for integer a≥2a \ge 2 with variable finite index.
Basis
Reported by the linked checker

Human understanding Few

A human explanation of the central model-produced argument is undocumented.

Correction and publication history