The Riemann hypothesis, often considered the most important unsolved problem in mathematics, may be the latest conquered by artificial intelligence.
Levent Alpöge, a mathematician at Anthropic, posted on Friday a four-page note on his personal website saying that he had completed a proof of the hypothesis by collaborating with Claude, Anthropic’s A.I. system. He did not include a traditional mathematical paper to describe the proof, but instead linked to computer code that he claimed verified the result.
“Claude and I have had a great time together with such breadth in recent months,” Dr. Alpöge wrote, “but this time Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line.”
If the claim is verified, it is the latest remarkable advance by artificial intelligence this year. For months now, mathematicians have been repeatedly stunned to see machines conquer mathematical problems that eluded human geniuses for years — in the case of the Riemann hypothesis, for more than a century and a half.
“This is the Mount Olympus,” said Steven Strogatz, a professor of mathematics at Cornell University. Dr. Strogatz said it seemed clear that Dr. Alpöge played a central role in telling Claude which mathematical ideas to explore to pull together the proof.
“It is a remarkable thing we might see a lot more of in the future, where there is a very skillful jockey riding a really fast horse.” Dr. Strogatz said.
The Riemann hypothesis was formulated in 1859 by Bernhard Riemann, a German mathematician. The problem involves prime numbers (a whole number greater than 1 that can only be divided evenly by itself and 1, such as 2, 3, 5, 7, 11, 13 and so on) and the question of how they are distributed along the number line as it stretches toward infinity.
As one moves further along the line, primes become rarer: Up to 100, some 25 percent of the numbers are prime, but up to one million, only about 7.8 percent are. Riemann hypothesized a formula for estimating, or counting, how many primes exist below any given number.
Other researchers said they were checking Dr. Alpöge’s code, which is written in Lean, a language that turns mathematical statements into a form that can be verified. If the Lean code compiles, that indicates the logic of the proof is solid.
“I think it’s likely to be correct,” said James Maynard, a mathematician at the University of Oxford in England. “I’d be 98 or 99 percent certain.”
Lean is a robust proof checker, Dr. Maynard said, but added: “You have to be a little bit careful about making sure that a Lean verification is proving the thing that you want it to be proving.”
Dr. Alpöge was not the only person working on the Riemann hypothesis. Also on Friday, Alessandro Masullo, an independent mathematician living in Rome, released an incomplete proof that he said he had been working on for years. Brian Conrad, a Stanford University mathematician who was Dr. Masullo’s doctoral thesis adviser, said the proof, which employs a completely different approach than the one used by Dr. Alpöge and Claude, was credible without major gaps.
“It’s a draft version of a work for which the ideas have all been worked out,” Dr. Conrad said. Dr. Masullo was in the process of revising the writing and fixing minor errors, Dr. Conrad said. He added that the plan was to release a polished paper in the coming year. He said it was “frankly completely outrageous” of Dr. Alpöge to release a proof that he did not fully understand.
Dr. Conrad said Dr. Masullo’s proof did not employ A.I., partly because he, like some other mathematicians, does not fully trust the system not touse his ideas and produce a proof without giving credit.
“The security of things put into such systems is not so clear,” Dr. Conrad said.
Dr. Alpöge did not reply to a request for comment about his proof.
Also on Friday, Dr. Alpöge and Tristan Buckmaster, a professor of mathematics at New York University, revealed a proof of a different variation of the Navier-Stokes problem. The Navier-Stokes equations describe the flow of liquids, and in September, OpenAI announced that it had solved a longstanding question: Could the equations produce nonsensical answers where the liquid could accelerate to infinite speeds? The OpenAI proof found such a case using a specially designed driving force applied to a liquid.
In their new work, Dr. Alpöge and Dr. Buckmaster claim to show that even without a driving force, a liquid could still reach infinite speeds.
Navier-Stokes and the Riemann hypothesis are among the seven Millennium Prize problems selected in 2000 by the Clay Mathematics Institute, which offered a $1 million prize for each one solved. A century earlier, in the year 1900, the Riemann hypothesis was among the 23 unsolved problems that the German mathematician David Hilbert named as the most important.
During that entire history, up until this year, a proof of the hypothesis seemed to be far beyond the capability of generations of mathematicians.
The problem attracted mathematicians not only because it seemed to be the biggest challenge, but also because it seemed to describe something basic yet mysterious about numbers.
“Basically, no one doubted that the Riemann hypothesis was true,” Dr. Maynard said. “We really wanted to know why it was true.”
The new proof “will give us great understanding, and so it’s something to be celebrated,” Dr. Maynard said.
In October, OpenAI released a tranche of more than 350 findings that included a proof of the so-called quasi-Riemann hypothesis, a version with looser constraints that is thus easier to prove. That in itself was a surprise, in that A.I. was able to crack the problem but also in that the solution turned out to be a fairly straightforward application of known tools. Indeed, several mathematicians had come close to completing a proof along similar lines.
It seemed that getting from the quasi-Riemann hypothesis to the full hypothesis would require new leaps of understanding. Yet, one month later, the Riemann hypothesis appears to be solved, too.
Alex Kontorovich, a mathematician at Rutgers University, said that after the quasi-Riemann hypothesis proof, he told himself, “Well, that’s it. Now that we know that A.I. can do mathematics at the highest levels, then I can’t be shocked anymore.”
But he said he was still flabbergasted on Friday. “I’m absolutely, absolutely shocked,” he said.
The post Anthropic Mathematician Claims to Have Solved Math’s Most Famous Problem appeared first on New York Times.




