Uniform cubic bounds for irrational values of the dilogarithm

Hu Tan, Ying Zhang

For any positive integers p and q with q>(4e4/27)p3q>(4e^4/27)p^3, Tan and Zhang prove that both Li⁡2(p/q)\operatorname{Li}_2(p/q) and Li⁡2(−p/q)\operatorname{Li}_2(-p/q) are irrational. They give a slightly lower sufficient threshold for negative arguments. The integers p and q need not be coprime.

What was already known

Rhin and Viola used double integrals and a permutation group to obtain computable denominator thresholds for fixed positive numerators. Earlier work extended the method to negative arguments. Prellberg used a recurrence for three adjacent moments to treat specific negative rational points.

AI involvement AI-led

Tan and Zhang report that ChatGPT through Codex developed most of the new arguments, including parameter choices, intermediate lemmas, and exact certificates. The authors supplied the problem and sources and directed successive rounds of mathematical development and checking.

Importance Notable

A single explicit cubic bound works for every numerator and both signs. It replaces a numerator-by-numerator threshold calculation in that range.

Technical verification Partial

The authors report Lean 4 checks of the new estimates and deductions. The uniform theorems take seven published or classical inputs as parameters, including prime-number estimates, two Rhin–Viola identities, and Legendre-function facts. The checks cover the deductions from those inputs. The proofs of the inputs are outside that formal development.

Formal checking

The authors report Lean 4 checks of the new estimates and deductions. The uniform theorems take seven published or classical inputs as parameters, including prime-number estimates, two Rhin–Viola identities, and Legendre-function facts. The checks cover the deductions from those inputs. The proofs of the inputs are outside that formal development.

Checker
Lean 4 with Mathlib, reported by Tan and Zhang
Scope
New uniform estimates and deductions from seven explicit inputs
Basis
Reported by the linked checker

Human understanding Few

The paper describes the positive-argument prime estimate and the negative-argument moment recurrence. The disclosure attributes the proof development and manuscript drafting chiefly to Codex. It does not identify a human explanation of the new estimates.

Correction and publication history

2026-11-07 · publication

Newly source-reviewed cubic sufficient denominator bound, with central Codex work and source-reported Lean checking from explicit inputs.

Read this published revision