科技
人工智能声称找到了解决已有167年历史的黎曼假设的方案——人类也提出了自己的解法。
中文译文
对莱文特·阿尔珀格来说,这是非常重要的一周:这位数学家看到自己的雇主Anthropic成为了首个公开上市的前沿人工智能实验室。他似乎还解决了数学领域中最重要的问题之一。
“时机实在糟糕,”他在一篇宣布自己研究成果的文章中写道。文章中他还描述了上周日夜晚他如何给Anthropic的首席执行官达里奥·阿莫代发消息,询问该如何宣布自己找到了一个可用于Lean编译的黎曼假设解决方案。黎曼假设可追溯到1859年,属于千禧年数学大奖难题之一。
该假设认为,素数的出现背后存在着某种隐藏的结构。如果这一成果属实,那我们将进入宇宙奥秘的深层领域。或许要理解这一成果对数学界而言有多令人兴奋,最好的方式就是看看专家们在Zulip上的讨论中是如何试图理解这一结果的。像东安格利亚大学的数论学家克里斯·伯克贝克这样的数学家就发表了这样的看法……
上同调界界的证明似乎是基于自同构提升的论证方法。这一证明确实运用了与GL_n在CM域上的正规代数极化尖点自同构表示相关的伽罗瓦表示的存在,同时还涉及基变换、雅凯-沙利卡定理、克洛泽尔纯度定理、伯恩斯坦-泽列文斯基定理、CM域的主要定理、德利涅-里贝特定理等等。
至于我,大学时在微积分课程后就不再接触数学了,因此无法准确理解这些概念的含义。
但这条评论之后有20位读者作出了反应🤯。数学界还需要一段时间才能完全掌握这一证明的精髓,但大多数人似乎都相信经过形式化处理的Lean结果是可靠的。Lean是一种开源编程语言,可用于将数学概念转化为代码,从而实现对证明的形式化验证。在Zulip聊天群中,包括麻省理工学院的数学家安德鲁·萨瑟兰在内的几位参与者表示,他们已成功确认阿尔珀格的解法可以成功编译。
在文章中,阿尔珀格对自己在没有给出真正易于人类理解的解释就公布该证明而感到些许内疚,但自童年起对这一问题的执着使他觉得,如果推迟公布这一成果,自己“将无法继续生活”。阿尔珀格描述了与Claude一起在极度混乱中逐步找到论证各部分的过程,字里行间流露出对人工智能的兴奋之情,他形容该模型“在接近终点时飞舞、跳跃、闪闪发光”。然而,要让所有这些努力有所意义,我们还得等待人类能够理解研究结果并相互解释,而这正是阿尔珀格致力于追求的目标。
阿尔珀格与他的合作者、纽约大学数学家特里斯坦·巴克马斯特还发布了另一项与纳维-斯托克斯方程相关的重大证明。这些方程引发了今年夏天各前沿实验室与数学家们为解决重大难题而展开的竞争。
与此同时,斯坦福大学教授兼美国数学协会主席拉维·瓦基尔发布了由斯坦福大学博士后亚历山德罗·马苏洛撰写的广义黎曼假设的证明,他指出:“这完全是人类完成的成果,没有用到人工智能。原本的计划是让亚历山德罗到斯坦福大学待一年,进一步完善证明文本,然后在征得足够多人的意见后再公布。但现在看来,直接发布目前的草稿为佳。”由于该证明尚未用Lean语言表达出来,现在还无法判断其是否正确,更不用说是否有人能够理解它了。
在对自己历经12年才完成的证明的简短介绍中,马苏洛写道:“如果这项工作想要向那些因当前数学界的节奏和氛围而感到气馁的数学家传达任何信息,那就是:我们依然在这里。我们依然充满活力。我们不会离开。”
此报道已更新,补充了阿尔珀格的Lean证明已由独立机构进行过验证的信息。
原文
It’s a big week for Levent Alpöge: The mathematician saw his employer, Anthropic, become the first publicly traded frontier AI lab. He also appears to have solved one of the most important problems in mathematics.
“The precise timing is horrible,” he wrote in an essay announcing his findings and describing how he messaged Anthropic CEO Dario Amodei last Sunday night to ask how to announce that he had found a Lean-compilable solution to the Riemann hypothesis, which dates back to 1859 and is among the Millennium Prize math problems.
The hypothesis conjectures that there is a hidden structure underlying the appearance of prime numbers. If the claimed result holds true, we’re entering sublime-mysteries-of-the-universe territory.
Maybe the best way to understand how exciting this is to the mathematics community is to watch the experts puzzling through the result in a chat on Zulip. Mathematicians like Chris Birkbeck, a number theorist at the University of East Anglia, are posting thoughts like this …
The cohomology bound is proved by what looks like an automorphy-lifting argument. It genuinely uses the existence of Galois representations attached to regular algebraic, polarized cuspidal automorphic representations of GL_n over CM fields, along with base change, Jacquet–Shalika, Clozel’s purity, Bernstein–Zelevinsky, the main theorem of CM, Deligne–Ribet and more.
For my part, I tapped out of mathematics at college calculus, so I can’t help decode exactly what that means.
But that comment was followed by 20 readers reacting 🤯.
It will be a while before the mathematics community feels like it has a handle on the proof, but most appear confident that the formalized Lean result will hold. Lean is an an open-source coding language used to express mathematical ideas as code, allowing proofs to be formally verified. Several participants in the Zulip chat, including MIT mathematician Andrew Sutherland, say they have successfully confirmed that Alpöge’s solution can be compiled successfully.
Alpöge appeared in his essay to feel a bit guilty about releasing the proof without a truly human-intelligible explanation, but his obsession with the problem since childhood made him feel he “could not live” if he waited to release the finding.
Alpöge described a “stream-of-consciousness jaunt through the experience of finding all the pieces of the argument through incredible confusion with Claude” with more than a hint of AI euphoria, describing how the model “flew, and danced, and sparkled near that finish line.”
However, for all that action to mean anything, we’ll have to wait for humans to be able to understand the result and explain it to each other, something Alpöge has committed to pursue.
Alpöge and a collaborator, NYU mathematician Tristan Buckmaster, also released another important proof related to the Navier-Stokes equations, problems that helped kick off the race between frontier labs and mathematicians to solve outstanding problems this summer.
Meanwhile, Ravi Vakil, a Stanford professor and the president of the American Mathematics Association, posted a proof of the General Riemann hypothesis written by Stanford postdoc Alessandro Masullo, noting that “it is all human, with no AI. The hope was to bring Alessandro to Stanford for a year to polish the writing and then announce it after running it by enough people. Now it seems best to put out the current draft.”
Without being expressed in Lean, it is too early to say whether the proof checks out, much less if anyone understands it.
In a brief introduction to his proof, which he said is the result of 12 years of work, Masullo wrote that “if any message at all this work is meant to send to the mathematician who feels discouraged by the pace and tone of the current climate, it is this: we are still here. We are very much alive. We are not going anywhere.”
This story has been updated to include that Alpöge’s Lean proof has been checked by independent sources.
