Riemann Hypothesis

Riemann Hypothesis Proof Claim: Anthropic Mathematician Says He and Claude Solved Math's Biggest Open Problem

Illustration of a glowing spiral of abstract mathematical curves and points rising from a dark chalkboard-like surface, no people, logos or readable text
Illustration (AI-generated)

Levent Alpöge, a mathematician at the artificial intelligence company Anthropic, announced on Friday, Nov. 20, that he believes he has proved the Riemann hypothesis, the problem many mathematicians consider the most important open question in their field, working with the company’s AI model, Claude. “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis,” Alpöge wrote, according to Nature.

The claim is extraordinary, and it is not yet accepted by the mathematical community. Early computer checks of a formal, machine-readable version of the proof have passed, but some of the most important verification steps were still running late Friday, and leading mathematicians say the work needs careful scrutiny. Within hours, a second, separate claim emerged: Stanford mathematicians said a young researcher there, Alessandro Masullo, has been working on a human-written proof for months.

Update (8:22 PM ET): The formal checks have now come back clean. On Saturday afternoon, the Lean and Mathlib communities published a joint statement, endorsed by the Lean Focused Research Organization and the Mathlib Initiative, saying: “We have full confidence that Alpöge’s proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement.” The statement says dozens of people have compiled Alpöge’s 17-million-line Lean library without errors, that two independent comparison tools confirmed the theorem it proves is exactly Mathlib’s statement of the Riemann hypothesis, and that the exported proof passed Lean’s official kernel plus several independent checkers, run on an inert export out of reach of any malicious code. “Honestly the chance that anything is fishy here is approximately zero,” MIT’s Andrew Sutherland told Scientific American. The Guardian reported that Alpöge said the work took about six weeks with Claude. The certification covers the machine-checked Lean proof; formal peer review, and any decision by the Clay Mathematics Institute on its $1 million prize, are still to come. The sections below were written before these confirmations.

What is the Riemann hypothesis?

The hypothesis comes from an 1859 paper by the German mathematician Bernhard Riemann, and it concerns the prime numbers, the numbers such as 2, 3, 5 and 7 that cannot be written as the product of two smaller numbers. Primes do not follow a regular pattern, but Riemann noticed that how often they occur is closely tied to the behavior of a function now called the Riemann zeta function, the Clay Mathematics Institute explains.

The hypothesis says that all the “interesting” solutions of the equation zeta(s) = 0, its so-called nontrivial zeros, lie on a single vertical line, where the real part of the number is exactly 1/2. If that is true, Nature notes, it would mean primes are scattered among the whole numbers in a way that is essentially as random as it could be. The Clay institute says the statement has been checked for the first 10 trillion solutions, but checking examples is not a proof that it holds for all of them. That is why a general proof has been sought for more than 160 years, and the problem is one of the Clay institute’s Millennium Problems.

What Alpöge actually released

According to Nature, Alpöge posted a 4-page document that includes a sketch of the proof. The core of the claim, though, is a much larger formal proof written in Lean, a programming language and proof checker that lets computers verify every logical step of a mathematical argument.

Scientific American reported that the claim goes further than the classic problem: it also says the proof settles the much harder generalized Riemann hypothesis (GRH). A verification page run by MIT mathematician Andrew Sutherland clarifies the scope: the formal package contains five headline theorems, including the Riemann hypothesis as defined in Lean’s standard math library, Mathlib, and an analogous statement for all Dirichlet L-functions. “GRH” in this case means Dirichlet L-functions only, the page says, not every possible generalization.

Has the proof been verified?

Partly, by computer, and not yet by people. Sutherland’s page, last updated at 22:05 UTC on Friday (5:05 p.m. ET), reported that:

  • All 175,187 modules of the published Lean code compiled with zero errors.
  • The five headline theorems rely only on Lean’s three standard axioms, and no extra axioms were added.
  • The Mathlib files used were byte-for-byte identical to the official versions.
  • Every module passed a replay through the Lean kernel, and two independently written checkers, called nanoda and con-leche, accepted the full exported proof. Another checker, lean4lean, accepted every compiled module.

At that time, the official Comparator check and a replay of the export through Lean’s own kernels were still running. The page also lists caveats: the build ran on cloud machines without a sandbox, and a comparison with a sandboxed build had not been made. Most importantly, it notes that a passing check shows the formal statements follow from the axioms, but “relies on those kernels and on Mathlib’s definitions being what they are intended to be.”

In plain terms, computer checks can confirm that the Lean code proves what the Lean statement says. Mathematicians still need to understand the argument and confirm that nothing unexpected is hiding in the setup.

How mathematicians are reacting

Reaction has been a mix of excitement and caution. “If this fully checks out, it is clearly a historic milestone in mathematics,” Fields Medal winner James Maynard of the University of Oxford told Nature.

Others were sharply critical of how the claim was released. Andrew Granville, a number theorist at the University of Montreal, told Nature that Alpöge should have prepared a clearer write-up first. “What he released is garbled,” Granville said. He also pointed to the timing, which came just days after Anthropic’s initial public offering and a drop in the company’s stock. Alpöge addressed that in his document: “There is an obvious cynical view, to do with the timing, but it is false,” he wrote, according to Nature. Anthropic did not immediately respond to Nature’s request for comment.

The second claim: Alessandro Masullo

Hours after Alpöge’s post, Ravi Vakil, president of the American Mathematical Society and a Stanford mathematician, wrote in a blog post that Masullo, a young Stanford mathematician, had produced a proof that purportedly did not use AI, Scientific American reported. According to the post, Stanford number theorist Brian Conrad has been giving Masullo feedback for several months. Conrad confirmed the details to the magazine.

Conrad was careful not to endorse the result. “To judge the success of the method requires the paper to reach a stage of more final revision,” he told Scientific American. “It’s undergoing major revision.” He said the choice to go public was between “two unpalatable options”: announce it now, or stay silent and post something later. The magazine noted that Masullo’s proof has not been verified by anyone and that other mathematicians are already skeptical.

Why it matters

A confirmed proof would be one of the biggest results in the history of mathematics. The Clay Mathematics Institute says a proof would “shed light on many of the mysteries surrounding the distribution of prime numbers,” and Nature notes it would land in the bicentennial year of Riemann’s birth. If the AI-assisted claim holds, it would also be the most dramatic example yet of AI systems contributing to research math, in a year Nature describes as full of AI mathematical breakthroughs.

What happens next

Expect the formal checks on Sutherland’s page to finish first, followed by a much slower process of human review, in which experts work through the argument. Neither claim has been through peer review. Until that happens, the safest description is that the Riemann hypothesis has a claimed proof, with promising early computer checks, and not a confirmed one.

FAQ

Has the Riemann hypothesis been solved?

It has been claimed, not confirmed. Computer checks of the Lean proof had mostly passed as of Friday evening, but key checks were still running and human experts have not signed off.

Who is Levent Alpöge?

A mathematician at Anthropic who announced the claimed proof, which he says he found working with Claude.

Who is Alessandro Masullo?

A young Stanford mathematician whose separate, human-written proof attempt was made public by Ravi Vakil. Brian Conrad says it is undergoing major revision.

What is GRH?

The generalized Riemann hypothesis extends the claim to related functions. In the Lean package, it covers Dirichlet L-functions.

Sources