Mathematician Ravi Vakil on November 20, 2026 released ahead of schedule a proposed proof of the Generalized Riemann Hypothesis by Alessandro Maria Masullo. The 940-page draft, The Music of the Primes, took about 12 years to put together, and according to Vakil, no AI was used in the argument. The same day, Anthropic mathematician Levent Alpöge also announced a proposed proof of the Riemann Hypothesis that he says was obtained jointly with Claude.
The Riemann zeta function is tied to the distribution of prime numbers, and the points where it takes the value zero are called zeros. The Riemann Hypothesis states that all of one group of these, the non-trivial zeros, lie on the line where the real part is one half. The Generalized Riemann Hypothesis extends the same question to other L-functions. Masullo's target is the L-functions corresponding to unitary Hecke characters over all number fields. A note remains on page 522 of the public draft asking that the statement and proof of the main theorem about zeros be added.
A Draft Meant to Be Refined Over a Year
In the draft's acknowledgments, Masullo writes that he began the research toward the end of his first year of graduate school at Stanford University, under the guidance of Brian Conrad, and that putting it together took about 12 years. According to Vakil's announcement, in the months before release Conrad had been reading the manuscript closely and Masullo had been incorporating his advice.
An overview of the whole method was sent by letter to mathematician Peter Sarnak in mid-October 2026. Sections 1.7 and 1.8 of the public draft's introduction are excerpts from that letter. On the Sunday just before the release, Masullo and Conrad visited the Institute for Advanced Study (IAS) in Princeton and spent four and a half hours discussing the work with Sarnak at a blackboard. Readers of the public draft can see the overview of the method that was shared before that discussion.
Conrad and Vakil had been preparing a plan to host Masullo at Stanford University for a year to refine the draft. Vakil had envisioned an announcement only after the text had been polished during that time and checked by a sufficient number of experts. Instead of waiting for that plan, Vakil posted the current version on November 20 and called for the argument to be checked and formalized.
In the draft's acknowledgments, Masullo explains that, given the circumstances at the time, he was encouraged to release the current version. Neither Vakil's announcement nor the acknowledgments states the direct trigger for moving the release forward. Conrad told Scientific American that judging whether the method works will require further progress on the revisions now under way.
The Unfinished Parts on Page 522
Masullo's draft covers all number fields, meaning all finite extensions of the rational numbers. Hecke L-functions are defined by choosing mathematical data called a character on each number field.
The argument starts from a trace formula, which expresses the same quantity in two different forms. Pages 519–520 of the draft contain Theorem 12.5.1.1, which states this formula for test functions satisfying specified vanishing conditions, along with its argument. On pages 29–30 of the introduction, the formula is used to study the logarithmic derivative of Hecke L-functions. The logarithmic derivative is the derivative of a function divided by the function itself, and poles appear where the original function is zero. The idea is that determining the range in which the logarithmic derivative is holomorphic, with no poles, narrows down where the zeros can be. The introduction then outlines a route that uses the functional equation to confine the real part of non-trivial zeros to one half.
For the conclusion that the real part of non-trivial zeros is one half, the author refers in Section 14.1.2 on page 528 to Theorem 13.3.1.2 in Chapter 13. A footnote on page 522 of the public draft asks that the statement of Theorem 13.3.1.2 and the proofs of the main group of theorems be added. In the footnotes on page 522, footnote 2, attached to the theorem heading, asks for the statement of the theorem to be inserted, and footnote 3, immediately after it, asks for the main theorem to be stated and proved.
On the same page, Section 13.1, "Explicit formulas," and Section 13.2, "Holomorphic families of test functions," remain only as headings, and the text moves on to Section 13.3. Theorem 13.3.1.1 beyond that contains a statement of the trace formula, but footnote 1 asks for that statement to be corrected. In the text that follows, at the place where a second theorem to be combined with the formula should be given, there is a reference whose number has not yet been fixed. Readers can follow the outline in the introduction and the argument in Chapter 12 in the public draft, but the step connecting them to the conclusion about the zeros is still waiting for its statements to be completed.
In the acknowledgments, Masullo himself says that the mathematics is complete and that what remains is filling in the text. He also notes that the introduction is due to be replaced and that he and Conrad are aware of several passages that need correction. For outside readers to test that assessment, the question is how the actual arguments are written at the places the footnotes point to in the released version.
The Range of L-Functions Covered by the Two Proposals
Andrew Sutherland of MIT used Alpöge's public code to verify five theorems in total, including the ordinary Riemann Hypothesis and statements that say, in the same way, where the zeros of all Dirichlet L-functions lie. Masullo's draft states as its target the Hecke L-functions corresponding to all unitary Hecke characters of every number field. Although the two share the name Generalized Riemann Hypothesis, the target stated by the draft's author differs from the target of the statements verified at MIT.
Alpöge's announcement points to a 14-part archive containing code, written in the Lean proof assistant, that records the theorems and their arguments. Sutherland, who built that code, used version 4.33.1 of both Lean and the Mathlib mathematics library. According to Sutherland's verification report, updated at 7:05 a.m. on November 21 (Japan time), all 175,187 modules compiled without errors, and all modules were also rechecked in Lean's kernel.
The alternative kernel implementations nanoda and con-lechе have also accepted the entire exported proof. con-leche additionally confirmed that the statements of the five main theorems match the statements of the published verification targets. The alternative implementation lean4lean accepted all of the built modules. Five of these modules hit the computation-step limit under default settings, and they passed when the limit was raised and they were rerun.
The five main theorems depend on three axioms: propext, Classical.choice, and Quot.sound. Passing the checks shows that, under the kernel and Mathlib definitions used, the formalized statements follow from those axioms. As of the same record, work was still in progress on rechecking the exported proof with the Lean kernel in standard and strict settings and on the series of checks using lake comparator.
What the Revision Opens Up for Third-Party Verification
The Music of the Primes, which Vakil released, is a draft that Masullo is still revising, and Alpöge has released an announcement and Lean code. For neither proposal can a formal peer-review result be confirmed. For Masullo's draft, completion of independent verification of the whole argument has also not been confirmed.
The Clay Mathematics Institute lists the Riemann Hypothesis as an unsolved problem. Under the rules for the Millennium Prize Problems, the conditions are publication in a qualifying outlet, at least two years after publication, and general acceptance in the global mathematics community.
For Masullo's proposal, the points readers will check after the revision are the full statement of Theorem 13.3.1.2 and the argument along the way from the trace formula to the location of the zeros. Once those are in place, third parties will be able to trace that connection and examine whether the conclusion holds within the range of number fields and characters the author has stated.
