The Riemann Hypothesis: a claimed proof, delivered in Lean
A four-page note, 17 million lines of Lean, and no paper yet.
Levent Alpöge · Claude · early checks by Elliot Glazer, Andrew Sutherland and the Lean community
Where this stands
A claimed proof of the Riemann Hypothesis and of GRH for Dirichlet L-functions, released as a Lean archive without a paper. Early builds and checks of the formal statement have been reported, but no expert review of the mathematics yet. This page does not assert that the hypothesis is proved, or that Clay has accepted a solution.
Start with the question
The problem
The Riemann Hypothesis, posed in 1859, says that every nontrivial zero of the zeta function has real part . It is the central open question about how the prime numbers are distributed, and one of the seven Clay Millennium Problems. The generalized hypothesis makes the same claim for Dirichlet -functions, which govern primes in arithmetic progressions. Claimed proofs have appeared many times before, and none has been accepted.
Before the news
Understand the mathematics
The established mathematics behind the story, step by step. What is new, and who claims it, follows below with sources.
The statement, as Lean sees it
For , Euler's identity says
Riemann extended to every complex . It vanishes at , the trivial zeros. Every other zero lies in the critical strip , placed symmetrically about the critical line .
Mathlib, Lean's mathematics library, already contains the hypothesis as a definition named RiemannHypothesis. It says: if , is not a trivial zero and , then . A formal proof is a proof of exactly this proposition. Because the definition is the library's own, readers do not have to trust a statement written by the authors of the proof.
Why the zeros control the primes
Weight each prime power by the logarithm of its prime: the von Mangoldt function is when and otherwise. Its partial sums count primes up to with these weights. Riemann's explicit formula, proved by von Mangoldt, writes this count in terms of the nontrivial zeros of . For not a prime power,
with the sum over zeros taken in symmetric pairs. Each zero contributes a wave of size about . So the hypothesis is equivalent to the error bound , which says the primes are spread as evenly as possible. Guinand and Weil later generalized the formula into an exact pairing between a test function summed over the prime powers and its transform summed over the zeros.
Dirichlet L-functions and the generalized hypothesis
A Dirichlet character modulo is a multiplicative function on the integers. It has period and vanishes on numbers sharing a factor with . Its -function is
It plays the role of for primes in arithmetic progressions. Averaging over the characters modulo isolates the primes ; in this way Dirichlet proved in 1837 that every progression with contains infinitely many primes.
The generalized Riemann hypothesis (GRH) for these functions says that every zero of with has real part . Many results are known to follow from it. One concerns primality testing: assuming GRH, Bach showed that the Miller–Rabin test decides whether is prime once it has tried every base , which makes the test deterministic.
A solved model: curves over finite fields
The analogue of the hypothesis is a theorem in one setting. Take a smooth projective curve of genus over the finite field . Its zeta function is rational, and there are complex numbers with
The analogue of the Riemann Hypothesis is . Weil proved it in the 1940s, and it gives the bound .
Stepanov found an elementary route in 1969, and Schmidt and Bombieri extended it to all curves in the early 1970s. The idea is to build an auxiliary polynomial that vanishes to high order, with large multiplicity, at every rational point, and then compare the number of zeros it has with its degree. The multiplicity comes from taking derivatives, which behave well in characteristic . Over the integers there is no evident way to differentiate a number, so the method has not carried over to itself.
01
A note, not a paper
On 20 November a four-page PDF, remark.pdf, appeared on alpo.ge, the personal website of Levent Alpöge, a mathematician at Anthropic. The note is unsigned; Scientific American attributes it to Alpöge. It says that the author, working with Claude, has found a proof of the Riemann Hypothesis, that the method also gives the generalized hypothesis, and that the argument has been formalized in Lean without extra assumptions. There is no paper. The author writes that a fully detailed written version would be unreadably long (a footnote puts a draft at more than 250,000 pages) and that they intend to understand the argument and write it up properly. [1][3]
02
What the note says the argument is
The note describes the proof as a descendant of Stepanov’s method, with the missing multiplicity supplied by chaining theta lifts between unitary groups. For a sum of the von Mangoldt function of length , the chain Claude proposed has length about . A 20-step timeline, which the author says Claude compiled, runs from 6 October to 16 November. Its ingredients include Gan–Gross–Prasad and Bessel periods, the Rallis seesaw, the Bost slope inequality, an Iwasawa tower and a relative trace formula. The author warns that the descriptions of the steps where Claude was the expert paraphrase a model-written summary and may change. [1]
03
The first checks
On the day of release, Elliot Glazer wrote that the formalized statement is in fact GRH, while his own build was still running. Andrew Sutherland of MIT told Scientific American that he had independently compiled the result in Lean and called it a historic moment, adding that much work remains to understand the mathematical details. Thomas Bloom pointed readers to a Lean Zulip thread where Lean experts are examining the formal proof. A clean build that uses only Lean’s standard axioms shows that Lean’s kernel accepts a proof of the stated theorem. It does not yet tell anyone why the argument works. [3][4][5][6]
04
What is in the archive
We downloaded the 14 parts (239 MB compressed) and read the archive’s text without building it. It holds 175,187 Lean files with 17.4 million lines and about 819,000 theorems and lemmas, pinned to Lean 4.33.1 and a fixed Mathlib commit. For comparison, the formal proof of Fermat’s Last Theorem was reported at about 13 million lines. Main.lean proves Mathlib’s own RiemannHypothesis, and Statements.lean states the Dirichlet versions with Mathlib’s L-functions. A search of the code found no sorry, no admit, no axiom declarations and no native_decide. A few custom tactics and one declaration-copying command pass their results to Lean’s kernel, which checks them again. A text search is not a build. [2]
Inspect the evidence
The Lean formalization
Lean archive released; independent builds reported, review under way
The note’s footnote links the archive as lean.tar.gz.part01 to part14. Join the parts in order, extract, and build with Lake. The excerpt below is the statement layer, not the proof: the headline theorem uses Mathlib’s definition of the Riemann Hypothesis, and the Dirichlet statement uses Mathlib’s DirichletCharacter.LFunction. The decisive checks are a clean build, #print axioms on the main theorems (only propext, Classical.choice and Quot.sound should appear), and a replay of the result in an independent checker. We have not built the archive.
1-- src/Main.lean (excerpt)2theorem riemannHypothesis : RiemannHypothesis := rhOfAll34theorem LFunction_eq_zero_re : DirichletLZerosOnLine := claimAll.2.156-- src/Statements.lean (excerpt)7def DirichletLZerosOnLine : Prop :=8 ∀ (N : ℕ) [NeZero N] (χ : DirichletCharacter ℂ N) (s : ℂ),9 0 < s.re → s.re < 1 → DirichletCharacter.LFunction χ s = 0 → s.re = 1 / 210-- Statement excerpt; the proof is the rest of the archive.Sources & further reading
Reviewed 20 Nov 2026. X permalinks were recovered from indexed posts and linked discussions; direct X pages were not consistently accessible. They document the conversation, not proof correctness.
- 1.remark.pdf — the four-page note on alpo.ge ↗Announcement
Unsigned and untitled; the PDF is dated 20 November 2026. Its footnote links the Lean archive.
- 2.Lean archive, part 1 of 14 (lean.tar.gz.part01) ↗Lean
An 18 MB download. Parts 01 to 14 (239,275,554 bytes in all) join in order into lean.tar.gz.
- 3.Scientific American: Anthropic mathematician claims to prove the Riemann hypothesis ↗Response
Reports Andrew Sutherland’s independent compilation and reactions from other mathematicians.
- 4.Lean Zulip: discussion of the claimed proof ↗Response
Reading the thread may require signing in to the Lean Zulip.
- 5.
- 6.
- 7.
Work through it together
Discussion
Loading discussions…