Has the Riemann Hypothesis Fallen? 17 Million Lines of Code Say Yes, Mathematicians Aren't So Sure

Categories: Editorial
Has the Riemann Hypothesis Fallen? 17 Million Lines of Code Say Yes,  Mathematicians Aren't So Sure

ON 20 November, Anthropic mathematician Levent Alpöge released a four-page note and an archive of nearly 17.4 million lines of Lean code, saying that, working with the company’s Claude AI, he had proved the Riemann hypothesis and its generalised cousin. A feat that many experts placed outside the realm of what could be achieved with today’s technology.

Reaction was electric. Brian Conrey, a leading expert on the problem, called it possibly the biggest thing ever to happen in mathematics, though he had yet to see a full manuscript. Ohio State’s Ghaith Hiary said he never expected to live to see this day. One of the most influential mathematicians of all time, David Hilbert once said: “If I were to awaken after having slept for a thousand years, my first question would be: Has the Riemann hypothesis been proven? If you have been following the series of Millennium prize breakthroughs driven by artificial intelligence, you can be forgiven for being a little shocked to see a 200-year problem fall so quickly, but then we may still be bamboozled by machine-based discovery.

Where the Lean proof stands

Lean is a proof assistant that checks every logical step mechanically. Mathlib, its main maths library, already contained a formal statement of the Riemann hypothesis, built on earlier work formalising zeta and L-functions. Within a day, the Lean and Mathlib communities reported that two independent tools confirmed the theorem proved in Alpöge’s library is exactly Mathlib’s statement of the hypothesis. Their joint statement expressed full confidence that the proof is correct. But the final sentence says — “We look forward to the mathematical community’s follow-up work to help everyone clearly explain what new ideas are in this proof.”

Why it matters

For over 160 years, one question has haunted mathematics: is there a hidden order in the prime numbers? In 1859, Bernhard Riemann linked the primes to a function called the zeta function and guessed that all its important “zeros” lie on a single straight line. If he was right, the primes are spread along the number line about as evenly as they possibly could be.

No one has been able to prove it, and the Riemann hypothesis became the most famous unsolved problem in math.

A multitude of mathematical papers simply assume it is true. A proof would convert those assumptions into theorems, and would guarantee that deviations from the prime number theorem’s average stay small. Carlo Pagano of Concordia University believes the new ideas could open entirely new avenues in number theory. Some coverage points to implications for cryptography, but the payoffs are more likely to be theoretical than immediate.

The case for caution

The difficulty is understanding. Alpöge says the human-readable version runs past 250,000 pages and admits he does not yet fully understand it. Proof-correct is not the same as knowing how it is proven. And the result is more in keeping with scientific discovery than human invention. Nobody complains that telescopes are being used to study stars.

The Association for Human Mathematics argues however that no human can reasonably claim to be responsible for, or even to understand, such a document, and that the announcement reflects a ‘corporate model valuing technical validity over accountability and understanding’.

An article on Emerald Book, goes further, arguing that compiling code is not the same as solving the problem, and noting that OpenAI’s September Navier–Stokes claim was later found to have been subtly mistranslated into Lean. It stresses that a human audit of the definitions is still needed. The Clay Mathematics Institute’s rules require peer-reviewed publication and at least two years of scrutiny before the $1 million prize can be awarded.

Context matters too. Alpöge works for Anthropic, and he announced the result a day after the company’s record-setting IPO. Meanwhile, independent mathematician Alessandro Masullo has published a 940-page human-written proof, which also awaits scrutiny. A human may have thus produced just such a proof months ago, purportedly without any assistance from artificial intelligence.

The bottom line

Mathematicians have rarely had this much formal evidence so quickly, and the claim is not yet settled. Whether the Riemann hypothesis is truly solved will be decided by human experts over months and years. Either way, the field of mathematics has fundamentally changed, leading to what Hannah Fry terms an ‘existential crisis’. You can read about the so-called mathpocalypse here.

Disclosure: Claude, the AI used in the proof, was developed by Anthropic, as was the digital assistant that aided the author in researching and drafting this post. I guided the assistant to produce working text from sources on both sides of the current debate, polishing and supplementing the result. My human contribution is thus editorial in nature. In 2009 the author compiled Mathlib for the Ubuntu OS project as part of an Ubuntu Day event.

Further note: I should also add that the Google search tool was used to initially find sources such as Nature and New Scientist, and I manually include these urls. Other voices arrived on my daily stream via social media such as x.com and facebook. The assistant enables me to rapidly piece together the story as though I am writing from what was once a busy newsroom, but of course, I am simply at my desk at home. If anyone wishes to have a discussion on a subject that Stanford refers to as compcivi (computers for civics), please inbox me.