Abstract:
According to technology media The Decoder, a "proof competition" between artificial intelligence and human mathematicians is taking place around one of the most famous unsolved problems in mathematics - the Riemann Hypothesis. Anthropic mathematician Levent Alpöge announced that
he and Claude found the proof of the generalized Riemann Hypothesis
. Just hours later,Stanford University researchers also published a draft of a competitive proof completed entirely by humans
. However, both results currently await further verification by the mathematical community.

Alperger announced on Friday local time that he had completed relevant research with the assistance of Claude and published the paper and supporting computer code on his personal website. He said that
the research results not only involve the Riemann Hypothesis proposed in 1859, but also cover the broader Generalized Riemann Hypothesis (GRH).
It is reported that Alperger has been working with Claude over the past few months to study this problem. Initially, he was responsible for defining research directions and integrating different methods on simplified mathematical models. However, after entering the actual problem, Claude made breakthroughs in some key derivation that exceeded his own understanding.
An important discovery in the research is that by connecting special mathematical transformations in series, the effect that is difficult to achieve by traditional differential methods in integer problems can be simulated. Alperger said that Claude came up with the key idea: for a sequence of length X, a number of X square root transformations need to be concatenated. He believes that this discovery itself is enough to shock mathematicians.
This proof of AI participation would exceed 250,000 pages if printed.
Alperger explained that because computer formal verification requires clearly writing out every logical derivation, even extremely basic mathematical rules cannot be omitted, so the scale of the proof is very large.Currently, what he has disclosed is a formal proof written in the Lean programming language, which is divided into 14 data packages. Lean can step by step check whether mathematical derivation conforms to established rules, but this does not mean that the proof has been recognized by the mathematical community. Experts still need to confirm whether the formal expression accurately corresponds to the original mathematical proposition.
Just hours after the news was announced, Ravi Vakil, professor of mathematics at Stanford University and president of the American Mathematical Society, urgently released another draft proof of the generalized Riemann hypothesis on his personal blog, written by mathematician Alessandro Masullo.
Vakil particularly emphasized: “This was done entirely by humans, without the use of AI.”
It is reported that Brian Conrad, a well-known number theorist at Stanford University, has reviewed Masulo's research for several months and provided a lot of feedback. The research team originally planned to spend a year to further improve the paper, but Anthropic suddenly announced the news of the AI proof, prompting them to disclose the existing draft in advance.
However, Masulo's proof was quickly questioned. Wakil updated his blog about eight hours after the first announcement, saying that many people have used different AI models to discover possible problems in the proof. He emphasized that all problems must be solved by the mathematical community before the proof can be officially accepted. He does not guarantee that this proof strategy is correct.
This means that neither the proof that Claude participated in nor the purely human proof published by Stanford University can currently be regarded as the Riemann Hypothesis has been finally overcome.
The Riemann Hypothesis was proposed by German mathematician Bernhard Riemann in 1859. It is one of the most important unsolved problems in modern number theory. It involves the mathematical laws behind the distribution of prime numbers. It has been listed as one of the seven "Millennium Mathematics Problems" of the Clay Institute of Mathematics since 2000, and a successful solution will receive a US$1 million prize.
The importance of this conjecture lies not only in its own mathematical value. Hundreds of mathematical studies are based on the premise that the Riemann Hypothesis is established. If the conjecture is finally rigorously proven, a large number of conditional conclusions relying on this hypothesis will be expected to be transformed into unconditionally established mathematical theorems. Its impact may also extend to fields such as cryptography, computer science and theoretical physics.
It’s worth noting that this mathematical proof competition occurred after Anthropic completed its record-breaking initial public offering. Alperger said the breakthrough actually occurred late Sunday night. He acknowledged that the timing of the announcement so closely following the company's IPO was not ideal, but denied that it was a PR stunt.
As for the role of artificial intelligence in mathematical research, Alperger said frankly that Claude's derivation ability in some complex fields has exceeded the scope of his current complete understanding. He plans to put together a more human-readable version of the proof in the coming months.
At present, the final fate of the two proofs still depends on independent review by the mathematical community. But this competition, which started almost simultaneously, has put a more realistic question before the scientific world: When AI can participate in solving basic scientific problems that have troubled mankind for hundreds of years, will the research methods and roles of mathematicians also change?
Comments