Quick Answer: Alessandro Maria Masullo has circulated a draft Generalized Riemann Hypothesis proof, and Ravi Vakil says Brian Conrad has reviewed it closely and found it promising. Several readers have since reported problems, and the community has not accepted the argument. Until every reported issue is resolved, treat the draft as unverified.
Nothing rattles the number theory world like a credible-looking proof of a famous conjecture. On November 20, 2026, Vakil posted a "hurried announcement" on his Math 216 blog about a Generalized Riemann Hypothesis proof by Masullo. The post was written from a meeting, and within hours it had changed. Here's what the announcement says, what the claim would mean, and how you should judge a proof like this before anyone else has checked it.
What the Generalized Riemann Hypothesis Actually Claims
The classical Riemann Hypothesis concerns the nontrivial zeros of the Riemann zeta function. Riemann conjectured in his 1859 paper that they all lie on the line where the real part of s equals 1/2. The Generalized Riemann Hypothesis extends that claim to a much larger family of objects, the Dirichlet L-functions and related L-functions, and it asserts the same critical-line behavior for their zeros.
Why does this matter beyond pure curiosity? GRH controls how evenly primes are spread across arithmetic progressions. Under GRH, Gary Miller's primality test becomes deterministic and runs in polynomial time. Many papers in analytic number theory are conditional: they say "assuming GRH, we get the following bound." Those results would become unconditional if GRH were proved.
Here's the detail that trips up non-specialists. The Clay Mathematics Institute's Millennium Prize concerns the classical Riemann Hypothesis, not GRH. Since GRH generalizes RH, a proof of GRH would settle RH too, but the reverse is not automatically true. Keep the two statements separate when you read headlines.
Numerical evidence is strong. Computations going back to Xavier Gourdon's work in the 2000s have confirmed that more than 10^13 zeros of the zeta function lie on the critical line. That is a useful sanity check, but numerical evidence can never prove the statement for all zeros. Several mathematicians have argued that the real obstacle is conceptual, not computational.
The Timeline, Step by Step
The post itself is short, but the sequence of events matters for judging the claim. In order:
- Vakil reports that Masullo has a proposed proof, and that Conrad has been examining it for months, giving "extensive feedback" that is being incorporated.
- The original plan was for Conrad to work with Masullo for about a year to polish the writing, then announce the result after wider review.
- Vakil decided instead to release the current draft early, asking people to check it, formalize it, and fill any small gaps.
- About eight hours later, Vakil updated the post. Several people had reported problems in the proof, identified with the help of different AI models.
- Vakil explicitly says he makes no claim about the correctness of the proof strategy, and that all issues must be resolved "to the satisfaction of the community" before the proof is accepted.
Step 5 is the one to remember. Vakil is not endorsing the result. He's describing a process that hasn't finished.
Why "All Human, No AI" Doesn't Settle Anything
Vakil stresses that the draft is "all human, with no AI." That's a statement about how the text was produced. It is not a statement about whether the argument is correct.
This is the counter-intuitive part. Most people assume a proof written by a careful human expert is probably right. Historically that assumption is incomplete. Human-written proofs of major conjectures have contained fatal errors, sometimes in a single lemma that everything else depends on. When a key step fails, the rest of a long argument can be correct and still prove nothing.
The irony in this case is sharp. The reported problems were found with AI models. That's a practical lesson: AI tools are increasingly useful as error-hunters, even if you wouldn't trust them to write the proof. Using them as adversarial checkers can surface gaps quickly, but a flagged gap still has to be confirmed by a human expert before anyone treats it as settled.
So when you see "human-written" or "reviewed by a respected expert" in a claim like this, read those as provenance and credibility signals. They tell you who to watch. They do not tell you the proof works.
How a Claimed Proof of This Size Gets Vetted
Major proofs get checked in layers, and each layer catches different failures. The table below shows the usual stages and what each one tends to miss.
| Stage | What it checks | Typical failure mode | Who usually does it |
|---|---|---|---|
| Expert reading | Logical flow and key lemmas | Hidden assumptions in "routine" steps | A handful of specialists in the subfield |
| Targeted gap-filling | Unproven or sketched steps | Gaps that need new techniques, not just details | The author plus collaborators |
| Formalization (e.g., Lean 4 with Mathlib) | Every inference, machine-checked | Missing background theory the library doesn't yet contain | Formalization specialists |
| Refereed journal publication | Overall validity and significance | Errors that survive informal reading | Editors and referees over months or years |
Formalization deserves special attention. Lean 4 with the Mathlib library lets a proof be checked step by step by software, which removes a whole class of human oversight errors. The catch is that formalizing an argument at this level requires building a lot of foundational material first. That's why Vakil asks people to "formalize" the draft: it's a real test, and it's slow.
The Clay Mathematics Institute's rules for the Millennium Prize require publication in a refereed journal of worldwide repute, followed by a two-year waiting period. Grigori Perelman's proof of the Poincaré conjecture was posted in 2002–2003, and the prize was not awarded until 2010. Even a correct proof takes years to become accepted. A proof that goes out in a hurry is not a shortcut through that process.
Where the Strategy Stands, and Why Nuance Matters
Vakil's disclaimer, "I make no claim as to the correctness of the proof strategy," is deliberately narrow. The strategy could be sound while a specific implementation fails. It could also be the other way around: the details could be fine, but the overall approach might not reach the conclusion.
Here's a concrete way to think about it. Suppose a proof of 40 pages depends on a single intermediate estimate on page 27. If that estimate needs a bound that no one has yet established, the claim fails, even if the 39 other pages are beautifully written. That's cause and effect in proofs: one unsupported input breaks the chain.
The honest position right now is this. There's a credible reviewer who has spent months with the draft, a public process for finding errors, and at least one set of reported problems that have not been ruled out. Those facts can coexist with the proof eventually being correct, or with it failing. Anyone who tells you which outcome is certain at this stage is guessing.
A Practical Checklist for Reading Proof Claims
You don't need to be a number theorist to read a claim like this sensibly. Ask these questions, in roughly this order:
- Who is checking it, and is the checking public? Named reviewers who have commented on a draft carry more weight than an anonymous preprint.
- Has anyone found a specific error? A reported gap with a location is far more informative than a general sense of doubt. Look for the exact step.
- Is the claim being presented as settled? Official statements from a respected source will say "proposed," "draft," or "under review." Watch for upgrades to "proved" in secondary coverage.
- Has formalization started? A machine-checked version changes the risk profile dramatically.
Most people stop after the headline. Don't. The update section is where the real information lives, and it's the part that tells you whether the claim is holding up.
The Related Claim Circulating Alongside This One
The Hacker News listing that linked to this post also referenced a separate claim by Levent Alpöge to have proved GRH, shared as a PDF. I haven't reviewed that document, so I can't assess it. Treat it the same way: check who is vetting it, look for specific reported gaps, and wait for formal review before forming a view. Two claims on the same conjecture in a short window is a reminder that interest in GRH is high, and that high interest tends to outpace careful verification.
Frequently Asked Questions
what is the generalized riemann hypothesis proof claim about?
It's a draft argument by Alessandro Maria Masullo asserting that the zeros of Dirichlet and related L-functions all lie on the critical line, where the real part is 1/2. Ravi Vakil posted the draft on his Math 216 blog on November 20, 2026, asking mathematicians to check and formalize it. The author has not yet won acceptance from the community.
is the masullo grh proof accepted by mathematicians?
No. Vakil says that all reported issues must be resolved to the community's satisfaction before the proof would be accepted as true. Brian Conrad has reviewed the draft and found it promising, but reviewer feedback is not the same as acceptance. Acceptance requires further checking and, normally, refereed publication.
how do mathematicians check a proof of the generalized riemann hypothesis?
They read it closely, fill in gaps, and try to formalize it in a proof assistant such as Lean 4 with the Mathlib library, which checks each inference by software. Serious claims then go to refereed journals. For the Clay Millennium Prize, the proof must also survive a two-year waiting period after publication.
why does the riemann hypothesis matter for prime numbers?
RH and GRH control how regularly primes are distributed. They give sharp error bounds on prime counting and on how primes spread across arithmetic progressions. Many results in number theory and cryptography-related algorithms are proved only conditionally on GRH, so a proof would make those results unconditional.