Lean and Mathlib Certified Alpöge's 17 Million Line Claude Proof of the Riemann Hypothesis
On November 21 the Lean and Mathlib communities certified Levent Alpöge's 17 million line Lean proof of the Riemann Hypothesis, posted November 20 with Claude.
A 167-year-old problem in number theory now has a machine-checked proof, and the people who maintain the checker say it is the real Riemann Hypothesis, not a lookalike theorem with the same name.
Levent Alpöge, a mathematician on Anthropic's technical staff, posted a four-page note on November 20: he believes that working with Claude he found a proof of the Riemann Hypothesis, that the same method gives the Generalized Riemann Hypothesis, and that the arguments are formalized in Lean with no extra assumptions. The archive sits in 14 parts on his site. He wrote that Claude "far outpaced me near the finish line."
On November 21 the Lean and Mathlib communities, with the Lean FRO and the Mathlib Initiative endorsing the statement, said they have full confidence the proof is correct. They put the library at 17 million lines. They say it depends directly only on Mathlib, that dozens of people reproduced the same raw proof export, and that two independent tools confirmed the theorem proved in Alpöge's code is exactly Mathlib's statement of the Riemann Hypothesis.
What the checkers actually ran
The statement they checked is Mathlib's RiemannHypothesis, which number theorists have already vetted. The exported proof was then run through Lean's official kernel plus separate checkers, including con-leche and con-ron, which come with a formal argument that they cannot accept a false proof. Those checks used an inert export, so leftover scripts in the source could not interfere. Andrew Sutherland at MIT published an overview of that work.
Michael Rubinstein at Waterloo ran a separate pass on university machines: all 175,187 modules compiled, the five theorems depend only on Lean's three standard axioms, Lean's kernel replayed every module, and an independent kernel, nanoda, accepted the 1,484,729 declarations those theorems rest on. No typechecker errors.
The Riemann Hypothesis is one of the Clay Mathematics Institute's seven Millennium Prize problems. Clay has not awarded anything. A kernel accepting a file is not the same as mathematicians agreeing they understand why the argument works. The Lean statement says the community still has to explain the new ideas.
Posted the morning after the IPO
Alpöge's own footnote is blunt about the calendar. The work finished late Sunday night. He messaged Anthropic CEO Dario Amodei before announcing, and he held the note until after the listing. "The precise timing is horrible. If I could have, I'd have finished this all much earlier." Anthropic's Class A shares started trading Thursday as ANT, a debut Finpresso already walked through.
The note is not an Anthropic press release. It is one employee publishing Lean. His timeline inside the PDF runs from a Stepanov determinant on October 6 to GRH on November 16. He says he guided the setup and that Claude supplied the technical steps he could not have written at that speed, including a chain of theta lifts whose length, for a von Mangoldt sum of length X, Claude put at the square root of X.
A proof this long is a research program, not a chalkboard argument. Rubinstein is already writing a human-readable account for the zeta function, and he is careful: the inequality he calls (M) still lives in the Lean development. The machine said the file proves the theorem. The next job is making that file a piece of mathematics people can read.
Some offers on this page may be paid placements or contain affiliate links.
Subscribe to Techpresso
Free daily newsletter, read in 5 minutes.
Subscribe free