A lower explicit threshold for Chen’s theorem under GRH

Get email updates

Steve Hubbard

Assume GRH and the exact fixed-level sieve and analytic premises corresponding to Bordignon and Starichkova’s Theorems 4–7. Then every even N≥exp⁡(exp⁡(13.375))N \ge \exp (\exp (13.375)) is the sum of a prime and a square-free integer greater than one with at most two prime factors.

What was already known

Bordignon and Starichkova’s 2024 journal Corollary 1 gives exp⁡(exp⁡(14))\exp (\exp (14)). Their 2022 preprint records the earlier exp⁡(exp⁡(15.85))\exp (\exp (15.85)). The formal 5–7 hypotheses use a factor-two main-term normalization and version 2 adds nonnegative sieve constants.

AI involvement AI-led

Claude Opus 5.5 independently found the new parameter choice and wrote the Lean check under Steve Hubbard’s direction. Gemini Argon High drafted the manuscript from Claude’s briefs; fresh Claude sessions checked correspondence.

Importance Modest

The threshold improves the cited numerical bound through parameter selection in the same sieve method, with four explicit analytic inputs.

Technical verification Formal (Lean)

Lean checks the exact conditional target, eleven exact ε\varepsilon windows, a one-sided bracket estimate and Chen’s weighted inequality. No human expert verified the four analytic transcriptions. Hexagon v4 discusses the all-level versus fixed-level sieve issue and explicitly retains the fixed-level form as an assumption.

Formal checking

Lean checks the exact conditional target, eleven exact ε\varepsilon windows, a one-sided bracket estimate and Chen’s weighted inequality. No human expert verified the four analytic transcriptions. Hexagon v4 discusses the all-level versus fixed-level sieve issue and explicitly retains the fixed-level form as an assumption.

Checker
Lean kernel and Comparator reported by Hubbard
Scope
For GRH and the exact Lean versions of the linear-sieve Theorem 4 and Theorems 5–7, every even N≥exp⁡(exp⁡(13.375))N\ge \exp (\exp (13.375)) is a prime plus a square-free integer greater than one with at most two prime factors. The manuscript discusses a fixed-level versus all-level sieve issue and retains the fixed-level form as a premise.
Basis
Reported by the linked checker

Human understanding Few

The record credits Claude with the mathematical parameter choice and the Lean proof. It does not identify a human explanation of the improvement or a human expert review of the four analytic transcriptions.

Correction and publication history