AREX Feed Article
ainta 独立复现确认 Claude 黎曼 zeta 证明:至少 67.25% 零点为单零点且在临界线上
8 月 11 日,独立研究者 ainta 在 X 上确认,Anthropic 由 Claude 撰写的黎曼 zeta 论文证明了"至少 67.25% 的零点为单零点且位于临界线上",并公开了复现仓库;他把论文交给 GPT-5.6-Sol Pro 后,下界被进一步推到 67.30%。当天 23:37(UTC),确认,Anthropic 为这一结果准备的 Lean 形式化证明通过了 Comparator 校验。
8 月 10 日,Anthropic 在中公布了这一结果:未发布的研究版 Claude 把黎曼 zeta 函数满足黎曼猜想的零点比例下界从 41.6% 提高到 67.2%,博客同时称 Claude 与 Anthropic 员工 Eric Easley 合作产出了一份可校验的形式化证明。
ainta 复现确认 67.25%,GPT-5.6-Sol Pro 给出 67.30085%
8 月 11 日 04:41(UTC),账号 写道:"Anthropic published a paper today, written by Claude, proving that at least 67.25% of the zeros of the Riemann zeta function are simple and on the critical line."
他的 X 简介标注为 QED_Audit 研究员、IOI 金牌、两届 ICPC World Finals 金牌(第 2、第 3 名),现为密码学博士生。截至抓取,这条帖子显示 110 万次浏览、2040 个赞、107 次转发。
帖子第二句交代了 67.30% 的来源:"I gave Claude's paper to GPT-5.6-Sol Pro. It's now 67.30%."他链接的于 8 月 11 日创建,README 说明这是对 Claude 论文《More Than Two Thirds of the Zeros of the Riemann Zeta Function Lie on the Critical Line》()中 Theorem D 的扩展,附完整证明与可复现的有限验证。论文 Theorem D 的原下界是 liminf N₀ˢ(T,2T)/N(T,2T) ≥ 3/2 − (1/√2)cot(1/√2) = 0.672500703679…,即 67.25%。
仓库在 Theorem D 之上加了两条新论证:取 3 个连续零点时得到 ε₄ ≥ 221/10⁶,对应下界 67.2519767%;取 7 个连续零点时得到一个六变量下界 ≥ 19/5000,对应 67.3008528%。主结果写为 liminf ≥ 0.673008527927…。
验证器使用 Arb 区间算术和穷举细分,运行 zeta-zero-verify three 与 zeta-zero-verify seven 即可重跑,每次从公式重建全部超越数包络。仓库状态栏写的是:"Research draft generated by GPT-5.6 Sol. Independent verification and peer review are welcome."这条 67.30% 扩展是 GPT-5.6 Sol 生成的研究草稿,作者自己写明欢迎独立验证与同行评审。
Lean 官方确认:形式化证明通过 Comparator
8 月 11 日 23:37(UTC), 发帖:"Great to see that these results include not only a Lean formalization, but one that passes validation in Comparator."帖子同时附上了仓库 。
仓库于 8 月 6 日创建,是对论文 Theorems A–E 的完整 Lean 4/Mathlib 形式化,不含 sorry。所有解析输入都有证明:Weil 显式公式、Riemann–von Mangoldt 零点计数、Γ′/Γ 的 Stirling 型估计、Chebyshev–Mertens 素数估计、Montgomery–Vaughan 广义 Hilbert 不等式。
顶层定理没有假设,仓库不声明任何公理,#print axioms 只报告 propext、Classical.choice、Quot.sound 三条标准公理。Theorem D 的 0.67250 下界以 ≥ 2 − 1/c₁* 的形式收录其中(c₁* = √2·tan ϑ/(1+ϑ·tan ϑ),ϑ = 1/√2)。
Comparator 指仓库的 comparator/ 目录:15+12+6 条受信任的题目声明配上解,最强的检查是语句相等性核对加 Lean 内核重放证明。仓库标注为研究产物、不维护、不接受贡献,主要贡献者账号 eeeasley 对应博客中提到的 Eric Easley。截至抓取,仓库有 45 颗星;Lean 官方这条帖子显示约 7254 次浏览。
67.2% 是博客口径,67.25% 是定理常数,67.30% 还在等评审
67.2% 是 Anthropic 博客给出的数字,论文 Theorem D 的精确常数是 0.672500703679…,即 ainta 复现确认的 67.25%。两者是同一个定理。
"单零点且在临界线上"比"在临界线上"更强:N₀ˢ 只数临界线上互不相同的单零点,N 数的是带重数的全部零点。ainta 仓库的主结果写为 liminf N₀ˢ(T,2T)/N(T,2T) ≥ 0.673008527927…。
67.30% 是三者中唯一尚未经同行评审的一环:它来自 ainta 仓库对 Theorem D 的扩展,仓库明确标注草稿由 GPT-5.6 Sol 生成。Anthropic 的 Lean 形式化覆盖论文原定理(含 0.67250 下界),不含 ainta 仓库的这条 67.30085% 扩展。
黎曼猜想本身仍然没有被证明
Anthropic 在里写明了边界:"We don't expect that the techniques Claude used will lead to proving the Riemann hypothesis."公司不认为这套技术路径能最终证明黎曼猜想。
博客还披露,Anthropic 的数学家 Levent Alpöge 与 Ralph Furman 验证了论文,Brian Conrey 与 Dan Goldston 两位该领域专家在短时间内审阅了论文。实验发起者、Anthropic 员工 Jarred Sumner 在 8 月 10 日的里写的是个人评价:"Still not sure what that means, but some analytic number theorists seem excited."
目前公开可核验的进展停在两处:论文的 67.25% 下界有通过 Comparator 的 Lean 形式化与 ainta 的独立复核;67.30085% 的扩展仍是等待同行评审的研究草稿。
参考链接
- ainta 的确认帖:
- ainta 的复现仓库:
- Lean 官方确认帖:
- Anthropic 的 Lean 形式化仓库:
- Anthropic 研究博客:
- 论文 PDF:
- Jarred Sumner 的帖子: