Claude 的黎曼猜想证明获 Lean 官方认证,马斯克直呼惊人成就Riemann Hypothesis Proof by Claude Officially Verified by Lean


AI free token/AI免费积分

Claude 的黎曼猜想证明获 Lean 官方认证,马斯克直呼惊人成就

困扰人类数学界 167 年的黎曼猜想,迎来最新进展。

就在刚刚,Lean 和 Mathlib 社区发布联合声明,宣布正式认证 Anthropic 数学家 Levent Alpöge 提交的黎曼猜想形式化证明,并表示对其正确性具有充分信心。

从数学家本人宣布成果,到形式化证明社区公开认证,整个过程仅仅过去一天。

更引人关注的是,Alpöge 此前透露,这项研究由他与 AI 模型 Claude 合作完成。在证明接近终点时,Claude 展现出的数学推理能力,甚至超出了他的预期。

11 月 20 日,Alpöge 在个人网站公布研究成果,宣称解决了著名的黎曼猜想(Riemann Hypothesis),同时取得了涉及广义黎曼猜想(Generalized Riemann Hypothesis)的结果。

黎曼猜想由德国数学家伯恩哈德·黎曼于 1859 年提出,是克雷数学研究所公布的七大千禧年数学难题之一。

它涉及黎曼 ζ 函数非平凡零点的分布,核心命题是所有非平凡零点的实部是否都等于 1/2。由于这一问题与素数分布密切相关,过去一个多世纪,无数数学家尝试攻克,却始终未能找到严格证明。

马斯克也发文感叹,证明黎曼猜想能够让大量此前依赖该猜想成立的数学结论摆脱条件限制,成为无条件成立的定理,并称其为超级智能取得的又一项惊人成就。

Alpöge 此次采用了一种相当特殊的成果发布方式。

他曾尝试让 Claude 将证明整理成人类能够阅读的文字版本,结果生成的草稿超过 25 万页,庞大到连他本人都认为难以阅读。如此惊人的篇幅,甚至被部分网友戏称为一场针对数学界的「DDoS 攻击」。

最终,Alpöge 决定不公开这份文字草稿,而是发布了一份 4 页的说明文件,以及分成 14 个压缩包的 Lean 形式化证明代码,总规模达到约 1700 万行。

Lean 是一种交互式定理证明工具,可以将数学命题及其推理过程转化为计算机能够严格检查的形式。研究者需要把复杂的数学逻辑表达为机器可验证的代码,再由证明检查器判断推理是否符合形式逻辑规则。

然而,面对如此庞大的代码,外界最关心的问题也随之出现。

1700 万行代码究竟有没有真正证明黎曼猜想,还是仅仅让计算机完成了一次看似成功的运行?

为回答这一问题,Lean 和 Mathlib 社区迅速组织了大规模验证工作,来自社区的众多成员共同参与检查。

其中一个关键环节,是确认程序最终证明的数学命题与原始猜想完全一致。毕竟,即使代码能够正常运行,也可能因为定义存在细微差异,最终证明了另一个问题。

社区为此采用 Mathlib 数学库中经过专家审查的黎曼猜想定义,并专门创建独立文件,使用 comparator 和 con-leche 两套工具进行比较,最终确认 Alpöge 证明的定理与 Mathlib 中的黎曼猜想定义完全一致。

与此同时,数十名社区成员成功编译整套代码,并复现出相同的原始证明导出文件。

麻省理工学院数学家 Andrew Sutherland 公布的验证记录显示,整套证明涉及 175187 个模块,全部成功编译,错误数量为零。

不过,代码能够成功编译,并不意味着其中的数学推理必然正确。

因此,社区进一步使用 Lean 官方内核,以及 con-leche、con-ron 等证明检查器,对导出的证明进行审查,同时引入由第三方开发、采用不同编程语言实现的独立检查器。

其中,con-leche 和 con-ron 还具备相对于集合论基础的形式化可靠性证明,能够降低检查工具自身存在逻辑漏洞的风险。为了排除潜在的代码干扰,相关检查针对的是导出后的静态证明数据,避免源代码中的恶意程序影响结果。

经过上述检查,Lean 和 Mathlib 社区认为此前提出的四项验证标准已经全部满足,并在联合声明中明确写道: 「我们完全相信 Alpöge 对黎曼猜想的证明是正确的,并祝贺他取得这一成就。」

值得注意的是,MIT 公布的验证记录还涉及五项主要定理,涵盖经典黎曼猜想,以及针对狄利克雷 L 函数的广义结果。不过,这并不意味着所有类型 L 函数对应的广义问题都已得到解决。

相比技术验证的结果,另一个受到广泛关注的问题,是 Claude 究竟在此次研究中扮演了什么角色。

Alpöge 在最初的声明中透露,黎曼猜想一直是他从学生时代就希望解决的问题。为了攻克这一难题,他长期积累相关理论知识,甚至做好了投入数十年研究的准备。

按照他的描述,研究初期主要由自己确定方向,将两种不同的数学思想结合起来,并引导 Claude 探索适合推广的数学情境。

随着工作逐渐深入,双方的分工也发生了变化。Alpöge 表示,在研究接近终点时,Claude 的推导速度远远超过了自己,部分数学思路甚至令他感到惊讶。

他还分享了一个具体例子。在尝试推广 Stepanov 方法时,研究遇到了如何在整数环境中模拟某种重数结构的问题,Claude 提出可以将 theta lifts 的链长度设定为 √X。

Alpöge 认为,这一建议成为推动研究取得进展的重要思路之一。

从研究者的描述来看,Claude 已经超越了文献检索、代码补全等传统辅助工作,开始直接参与前沿数学问题的探索与推导。

MIT 的验证记录还显示,独立检查工作也使用了 Claude Code。不过,目前尚无法准确量化人类研究者与 AI 在最终证明中的贡献比例,相关数学创新仍有待进一步梳理。

对于数学界而言,形式化证明获得认证,无疑是一次重大进展,但它也带来了一个此前很少有人需要面对的难题。

计算机可以严格检查数千万行代码中的逻辑关系,却无法直接向人类解释,其中究竟隐藏着怎样的新数学思想。

Alpöge 本人也承认,目前的证明规模过于庞大,而且极其难以阅读。他希望未来能够大幅简化推理过程,将其整理成适合数学家理解、讨论和传播的传统论文。

菲尔兹奖得主邓煜也表示,「至少拿出 500 页版本前看都不看」,足以说明如此庞大的证明给人类阅读带来的压力。

当 AI 生产知识的速度远超人类理解知识的能力,数学家可能逐渐沦为 AI 的审稿人,耗费大量时间消化机器生成的成果。科学进步的速度与人类认知能力之间,或许也将出现一道新的鸿沟。

本文来源:https://www.ifanr.com/1690820
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Riemann Hypothesis Proof by Claude Officially Verified by Lean

The Riemann Hypothesis, which has perplexed the mathematical community for 167 years, has reached a new development.

Just now, the Lean and Mathlib communities issued a joint statement announcing the formal certification of the formalized proof of the Riemann Hypothesis submitted by Anthropic mathematician Levent Alpöge, and expressing full confidence in its correctness.

Only a single day passed from the mathematician's own announcement of the result to the public certification by the formalized-proof community.

Even more notable is that Alpöge previously revealed that this research was conducted in collaboration with the AI model Claude. As the proof neared its end, Claude demonstrated mathematical reasoning abilities that even exceeded his expectations.

On November 20, Alpöge published his research findings on his personal website, claiming to have solved the famous Riemann Hypothesis while also obtaining results related to the Generalized Riemann Hypothesis.

The Riemann Hypothesis was proposed by the German mathematician Bernhard Riemann in 1859 and is one of the seven Millennium Prize Problems announced by the Clay Mathematics Institute.

It concerns the distribution of the nontrivial zeros of the Riemann zeta function, with the central proposition being whether all nontrivial zeros have real parts equal to 1/2. Because this problem is closely related to the distribution of prime numbers, countless mathematicians have attempted to crack it over the past century and more, yet a rigorous proof has never been found.

Elon Musk also posted a comment, remarking that proving the Riemann Hypothesis could free a large number of mathematical conclusions that previously depended on the hypothesis from their conditional status, turning them into unconditionally valid theorems, and calling it another astonishing achievement of superintelligence.

Alpöge adopted a rather unusual way of releasing his results this time.

He once tried having Claude organize the proof into a text version readable by humans, but the resulting draft spanned more than 250,000 pages, so enormous that even he found it hard to read. Such astonishing length was even jokingly called a 'DDoS attack' on the mathematical community by some netizens.

In the end, Alpöge decided not to release this text draft, but instead published a 4-page explanatory document, along with the Lean formalized proof code split into 14 compressed packages, totaling roughly 17 million lines.

Lean is an interactive theorem-proving tool that can convert mathematical propositions and their reasoning processes into a form that computers can rigorously check. Researchers need to express complex mathematical logic as machine-verifiable code, after which a proof checker determines whether the reasoning conforms to the rules of formal logic.

However, faced with such a massive body of code, the question of greatest concern to the outside world also emerged.

Did the 17 million lines of code truly prove the Riemann Hypothesis, or did they merely make the computer complete a seemingly successful run?

To answer this question, the Lean and Mathlib communities quickly organized large-scale verification work, with numerous community members participating in the checking.

A key step was confirming that the mathematical proposition ultimately proved by the program was completely identical to the original hypothesis. After all, even if the code runs normally, subtle differences in definitions could mean it ended up proving a different problem.

For this purpose, the community adopted the expert-reviewed definition of the Riemann Hypothesis in the Mathlib library and created a dedicated separate file, using two tools, comparator and con-leche, for comparison, ultimately confirming that the theorem proved by Alpöge was completely identical to the definition of the Riemann Hypothesis in Mathlib.

At the same time, dozens of community members successfully compiled the entire body of code and reproduced the same original proof export file.

Verification records published by MIT mathematician Andrew Sutherland show that the entire proof involved 175,187 modules, all of which compiled successfully with zero errors.

However, the fact that the code compiles successfully does not mean that the mathematical reasoning within it is necessarily correct.

Therefore, the community further used the official Lean kernel, along with proof checkers such as con-leche and con-ron, to review the exported proof, while also introducing independent checkers developed by third parties and implemented in different programming languages.

Among them, con-leche and con-ron also possess formalized soundness proofs relative to a set-theoretic foundation, reducing the risk of logical flaws in the checking tools themselves. To rule out potential code interference, the checks targeted the exported static proof data, avoiding any influence on the results from malicious programs in the source code.

After the above checks, the Lean and Mathlib communities considered all four previously proposed verification criteria to have been met, and stated explicitly in the joint statement: 'We fully believe that Alpöge's proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement.'

It is worth noting that the verification records published by MIT also involve five major theorems, covering the classical Riemann Hypothesis as well as generalized results for Dirichlet L-functions. However, this does not mean that the generalized problems corresponding to all types of L-functions have been resolved.

Compared with the results of the technical verification, another widely discussed question is what role Claude actually played in this research.

In his initial statement, Alpöge revealed that the Riemann Hypothesis had been a problem he wanted to solve since his student days. To crack this difficult problem, he accumulated relevant theoretical knowledge over the long term and was even prepared to devote decades of research to it.

According to his account, in the early stages of the research, he mainly set the direction himself, combining two different mathematical ideas and guiding Claude to explore mathematical contexts suitable for generalization.

As the work progressed, the division of labor between the two also changed. Alpöge said that as the research neared its end, Claude's speed of derivation far surpassed his own, and some mathematical ideas even surprised him.

He also shared a specific example. When trying to generalize the Stepanov method, the research encountered the problem of how to simulate a certain multiplicity structure in an integer setting, and Claude proposed setting the chain length of the theta lifts to √X.

Alpöge believes that this suggestion became one of the important ideas that drove the research forward.

Judging by the researcher's description, Claude has gone beyond traditional auxiliary tasks such as literature retrieval and code completion, and has begun to directly participate in the exploration and derivation of frontier mathematical problems.

MIT's verification records also show that Claude Code was used in the independent checking work. However, it is not yet possible to accurately quantify the proportional contributions of the human researcher and the AI to the final proof, and the relevant mathematical innovations remain to be further sorted out.

For the mathematical community, the certification of a formalized proof is undoubtedly a major advance, but it also brings a difficult problem that few people previously had to face.

Computers can rigorously check the logical relationships among tens of millions of lines of code, but cannot directly explain to humans just what new mathematical ideas are hidden within them.

Alpöge himself also admits that the current proof is overly large and extremely difficult to read. He hopes that in the future it will be possible to greatly simplify the reasoning process and organize it into a traditional paper suitable for mathematicians to understand, discuss, and disseminate.

Fields Medalist Deng Yu also said that he 'wouldn't even look at it before a 500-page version is produced,' which suffices to show the pressure that such a huge proof places on human reading.

When the speed at which AI produces knowledge far exceeds the human ability to understand it, mathematicians may gradually be reduced to reviewers for AI, spending large amounts of time digesting machine-generated results. A new gap may also emerge between the pace of scientific progress and human cognitive ability.

本文来源:https://www.ifanr.com/1690820
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry