Pull to refresh
Logo
Independent mathematicians certify AI-assisted Riemann hypothesis proof

Independent mathematicians certify AI-assisted Riemann hypothesis proof

New Capabilities

Levent Alpöge's 17-million-line Lean library proves the correct statement; peer review and human understanding lag.

Overview

Updated 1 hour ago

For 167 years, the Riemann hypothesis has been a central open question about prime numbers. On November 20, 2026, Levent Alpöge posted a four-page paper and a 17-million-line Lean library that he said proved it with Claude's help. Within a day, the Lean and Mathlib communities certified that the library compiles, proves the same statement as Mathlib's RiemannHypothesis, and passes independent proof checkers.

The result does not end the story. The proof is far too large for any person to read line by line, and a separate 940-page human draft by Alessandro Masullo and Brian Conrad appeared the same day, unfinished. The code settles whether the theorem is true. It does not yet explain why, or turn the proof into mathematical insight.

Why it matters

A machine-checked proof can settle a 167-year-old question while leaving mathematicians unable to explain it.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

17 million
Lines of Lean code
Size of Alpöge's library, released as formal proof.
5
Headline theorems checked
Each depends only on three standard axioms, per MIT verification page.
940
Pages in human draft
Masullo and Conrad released the unfinished manuscript the same day.
250,000
Pages in unreleased AI explanation
Alpöge says this is the length of Claude's explanation.
167 years
Time since Riemann posed hypothesis
Riemann proposed it in 1859.

Voices

Curated perspectives — historical figures and your fellow readers.

Sign up to generate historical perspectives on this story.

People Involved

Organizations Involved

Timeline

4 events Latest: Today
Tap a bar to jump to that date
  1. Lean community certifies proof

    Today Verification

    Statement and proof meet four community criteria.

  2. Independent kernels accept export

    Today Verification

    MIT page says Lean source compiles; comparator run pending.

  3. Alpöge posts AI-assisted proof

    Announcement

    Releases four-page paper and 17-million-line Lean library.

  4. Human proof draft released

    Announcement

    Masullo and Conrad post unfinished 940-page manuscript.

Scenarios

1

Peer review accepts the Lean proof

Possible Resolves by Nov 21, 2028

Discussed by: Scientific American and AFP note formal verification is not the same as peer review.

Lean and Mathlib have certified the code, but AFP reported that one institute requires peer review and two years of scrutiny. A journal would have to accept a proof whose human-readable explanation runs 250,000 pages. Publication follows if editors treat kernel checks as sufficient evidence; a demand for conventional exposition could stall it.

2

Mathematicians turn the code into a readable proof

Unlikely Resolves by Nov 21, 2029

Discussed by: Science quotes mathematicians who expect it could take years for the code to be digestible.

The proof settles the theorem but not human understanding. Science quotes mathematicians who expect it could take years for the code to be digestible. If researchers isolate intermediate lemmas and rewrite them in conventional mathematics, a readable proof could follow and validate hundreds of results that assume the Riemann hypothesis.

3

Masullo-Conrad draft collapses under scrutiny

Possible Resolves by End of 2027

Discussed by: Science and TechCrunch report the draft was released before its authors finished checking it.

The 940-page manuscript was released before its authors finished checking it. Conrad said they had planned another year to address errors. A fatal gap would leave the Lean proof as the only verified route; a clean review would give mathematicians two independent proofs to compare.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

1976

Four color theorem (1976)

Kenneth Appel and Wolfgang Haken announced a proof that every planar map can be colored with four colors. It relied on a computer checking hundreds of cases, which no human could verify by hand.

Then

Some mathematicians objected that a computer-assisted proof was not a proof.

Now

The result was eventually accepted, and computer-assisted proofs became a standard tool.

Why this matters now

It established the same tension now facing the Riemann proof: a machine can check what humans cannot.

1998-2014

Kepler conjecture (1998-2014)

Thomas Hales announced a proof that no arrangement of spheres packs more tightly than the face-centered cubic arrangement. The proof was long and complex, and a team spent years formalizing it.

Then

The formal verification, called Flyspeck, finished in 2014.

Now

It showed proof assistants could certify a major result that peer review had struggled to check.

Why this matters now

Alpöge's Lean proof follows the same path: machine verification substitutes for line-by-line human checking.

1993-1995

Wiles's Fermat proof (1993-1995)

Andrew Wiles announced a proof of Fermat's Last Theorem in 1993. A referee found a gap months later.

Then

Wiles and Richard Taylor repaired it in 1994; the proof was published in 1995.

Now

The episode became a case study in how expert review catches errors that authors miss.

Why this matters now

It shows that even a celebrated proof can need repair, as the Masullo-Conrad draft may.

Sources

(7)