⏱️ Reading time: 7 min

On November 20, 2026, mathematician Levent Alpöge uploaded a repository containing 17 million lines of Lean code. Inside, according to the community that reviewed it, is a proof of the Riemann Hypothesis, one of the oldest and most widely used conjectures in mathematics [source].

📑 En este artículo
  1. What the Riemann Hypothesis proof claims
  2. Four checks before calling it “verified”
  3. Who this changes something for
  4. Lean is not the only proof assistant: Rocq and Isabelle/HOL
  5. What the verification does not settle
  6. How to audit Alpöge’s library without being a prime number theorist
  7. References

What the Riemann Hypothesis proof claims

The Riemann Hypothesis predicts where the zeros of a function called the Riemann zeta function fall, and much of what is known (or assumed) about the distribution of prime numbers depends on that prediction. Alpöge claims to have proved the hypothesis and wrote it directly as a library for the Lean proof assistant, verifiable by a machine rather than by a committee of human reviewers [source].

The hypothesis is not just an abstract curiosity: many results in number theory, from error bounds in the distribution of primes to constructions in factorization-based cryptography, are stated today as “assuming the Riemann Hypothesis, it holds that…”. If Alpöge’s proof withstands the mathematical community’s scrutiny over the coming months, those conditional statements would become unconditional theorems.

Lean is a language and proof checker: each step of a proof translates into logical rules that a small program, the kernel, can check mechanically. Mathlib is the collaborative library where the Lean community accumulates already-formalized mathematics; the statement of the Riemann Hypothesis already lived there, as RiemannHypothesis, before Alpöge’s proof arrived.

Four checks before calling it “verified”

A joint statement from the Lean and Mathlib communities, backed by the Lean FRO and the Mathlib Initiative, certified the proof against four criteria published before the announcement [source]: that the formal statement is the correct one, that the library compiles, that the code proves exactly that statement, and that the raw proof is valid.

The second point was confirmed, according to the statement itself, by dozens of people who independently reproduced the build and got the same export file [source]. The third was resolved by isolating the Riemann statement in a file that only imports Mathlib and comparing it against the proof’s export with two separate tools, comparator and con-leche --compare; both agreed that the theorem proved is exactly Mathlib’s, not a weaker version [source].

The fourth criterion carries the most weight: the raw proof went through Lean’s official kernel and through several external checkers, including con-leche and con-ron, which in turn have a formal proof, relative to set theory, that they cannot accept a false proof [source]. The idea is that a bug in Lean’s logic alone cannot be enough to slip through a false proof: there are checkers built in other languages reviewing the same export, without access to the original source code.

flowchart TD
    A["RiemannHypothesis statement in Mathlib"] --> B["Compared against Alpoge's library"]
    B --> C["Library compiles without errors"]
    C --> D["Lean kernel checks the raw proof"]
    D --> E["con-leche and con-ron check the export independently"]
    E --> F["Lean and Mathlib community certifies the result"]
Alpöge’s library spans about 17 million lines of Lean code. Foto de Trnava University en Unsplash

Who this changes something for

For a number theorist, the news confirms, if it holds up, something much of the community already took for granted in practice. For anyone working with proof assistants, the case matters for a different reason: it is the largest formalization made public in Lean so far, and it stress-tests the same kernel and the same Mathlib that other formal verification projects use, not just pure mathematics ones.

Lean’s architecture deliberately separates a small, trusted kernel from the rest of the system: anyone can write elaborate tactics or automation, but in the end everything reduces to terms that this minimal core checks step by step. That same separation is what allows, in theory, trusting a 17-million-line proof without reading all of it: there is no need to trust the 17 million lines, only the kernel that checks them at the end, plus the external checkers that reviewed the export on their own.

If you maintain a library in Lean or Rocq to verify a compiler, a cryptographic protocol or a consensus algorithm, this case is a concrete reference point: it shows what happens when a 17-million-line proof goes through cross-checking with external tools before anyone calls it “verified”, and which part of that process (comparing the isolated statement, running independent checkers on the raw export) you can replicate in your own project even if it is much smaller.

Lean is not the only proof assistant: Rocq and Isabelle/HOL

Lean and Mathlib compete, in the sense that they offer an alternative path, with two other proof assistants the academic community uses to formalize mathematics and critical software: Rocq (the project formerly known as Coq) and Isabelle/HOL.

AssistantApproachWhere it is used most
Lean + MathlibDependent types, minimalist kernel, a single community math libraryRecent formalized mathematics (this case) and, increasingly, software verification
Rocq (formerly Coq)Calculus of inductive constructions, Mathematical Components library separate from the standard oneCritical software verification, with CompCert as the project’s historical flagship
Isabelle/HOLClassical higher-order logic, with the Sledgehammer automation engineLarge-scale mathematical formalizations and verification in European industry

The difference that matters most for a newcomer is not which one proves things fastest, but how large and active the shared library is: Mathlib grows every day with community contributions, which explains why the Riemann statement was already there waiting for a proof before Alpöge published his. All three assistants are free software, so the barrier to entry is not cost but the learning curve and available documentation.

What the verification does not settle

In the Hacker News thread about the announcement, user hodgehog11 made the distinction that matters most after the initial excitement: “it is almost certainly proved, but it is not ‘proven’. For that, you need a reasonably deep human-language explanation of how and why the proof works” [HN]. Lean can confirm that every logical step is valid relative to Mathlib’s axioms without any human having yet understood the full strategy of the proof, and that work of mathematical exposition in prose does not exist yet.

Another comment in the same thread, from Certhas, takes the opposite view: it argues the result “is already verified to a higher standard than the vast majority of peer-reviewed mathematics from the last decade” [HN], precisely because it went through independent checkers in addition to the Lean kernel. A third comment, from feverzsj, takes the urgency out of the matter from another angle: for the number theory community, no one was acting as if the Riemann Hypothesis could be false, so in daily practice “it doesn’t change much” [HN].

None of these three positions is a verified fact: they are different readings of the same formal verification. What is concrete is that the certification covers the statement and the logical validity of the proof, not a traditional peer review or publication in a journal with human referees, which has not happened yet.

Mathlib is the collaborative math library where the Riemann statement already lived. Foto de CDC en Unsplash

How to audit Alpöge’s library without being a prime number theorist

For a developer who wants to check something personally without reading 17 million lines, the path the statement itself describes is to clone Alpöge’s public repository, install Lean and Mathlib following the official documentation, and build the library locally to try to reproduce the result that dozens of people have already reported [source]. The file that isolates the Riemann statement, the one that only imports Mathlib, is the shortest starting point to confirm that what gets proved is the real statement and not a variant.

Anyone who wants to go a step further can review the extended technical analysis linked from the statement itself, with details of how the checks on the raw export were built [source], before citing this result as a settled fact in a paper, a class or an article of their own.

📌 Note: the certification comes from the Lean and Mathlib community itself, not a ruling from a peer-reviewed journal or a formal consensus of the broader number theory community.

References

Featured image: Foto de Peaky Frames en Unsplash

📬 Get new articles by email

We only email about big articles (1-2 a month).

How each article is made (sources, testing and editorial criteria): How we work.

Did it work for you? Got a different error? Say so below: questions get answered and help the next reader.

Leave a comment
Categories: Tech News

El Solitario Editorial

Article prepared by the El Solitario editorial team from cited sources, under the editorial responsibility of Luis Navarro. More details in "How we work".

0 Comments

Leave a Reply

Avatar placeholder

Your email address will not be published. Required fields are marked *

You can include code inside <code>…</code> or, for several lines, <pre><code>…</code></pre>.

This site uses Akismet to reduce spam. Learn how your comment data is processed.