Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)

Mathematician claims he proved the Riemann hypothesis with Claude, $1 million prize

Mathematician Alexander Polishchuk says that he and Claude proved the Riemann hypothesis, the question about prime numbers that has resisted solution for 167 years and carries a $1 million prize from the Clay Mathematics Institute. The proof is written in Lean, a kind of compiler for mathematics in which a program checks every step. But there is no independent verification yet, and it has not been confirmed that the formalization in Lean matches the actual hypothesis. For now, the accurate wording is "claims" rather than "proved."

AI-processed from TG: @ai_machinelearning_big_data (Machinelearning); edited by Hamidun News
Mathematician claims he proved the Riemann hypothesis with Claude, $1 million prize
Source: TG: @ai_machinelearning_big_data (Machinelearning). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @ai_machinelearning_big_data (Machinelearning) ↗
Editor-in-Chief
Key metric #1According to the author, Claude proposed a chain of theta lifts of length √X
Key metric #2Instead of a paper, the author released Lean archives and a four-page description of the work
Key metric #3The author says the intermediate materials amount to 250,000 pages

Bold claims of a "proof of the Riemann hypothesis" appear regularly, and nearly all of them fall apart on first inspection. This story is unusual: the author is a professional mathematician, Alexander Polishchuk, Claude took part in the work, and according to the author, a machine verified the result. That does not mean the problem is solved. Still, this example is a useful way to look at how sensational mathematical claims are checked today and what role AI can play.

What exactly is claimed

The Riemann hypothesis was formulated in 1859, so the problem has been open for 167 years. It is one of the seven Millennium Prize Problems of the Clay Mathematics Institute, and a $1 million prize is offered for solving any of them. Roughly speaking, the hypothesis describes how "evenly" prime numbers are distributed along the number line. Many estimates in number theory depend on it.

Polishchuk claims to have obtained a proof not only of the hypothesis itself but also of its generalized version (GRH). According to him, Claude played a key role in the final stage: the model suggested using a chain of theta lifts of length √X, and this became one of the central elements of the argument. A theta lift cannot be explained in two lines. It is a technical device from advanced mathematics, and only specialists can evaluate it by reading the document.

The format of the publication is also unconventional. There is no ordinary paper. The author posted archives with the formalization and a four-page document describing the course of the research. According to him, the volume of intermediate material exceeds 250,000 pages. No person could read that much in full.

What Lean is, and why "verified" does not yet mean "correct"

Lean is a system in which a proof is written as a program, and the compiler checks every logical step. For a developer, the closest analogy is automated tests. If everything builds without errors, the conclusion follows from the premises within the system. That is why "verified in Lean" carries more weight than "I reread it and it seems to hold."

But tests have a well-known problem, and Lean has the same one. Imagine a project where every test is green, yet the tests check a function different from the one the product needs. Green results prove nothing. In mathematics, this looks like the following:

  • Formulation. If the definition of a prime number or of the zeta function is written in code slightly differently from the original, the theorem that gets proved is a different one.
  • Extra assumptions. One must confirm that the archive contains no added axioms or placeholders that make the proof trivial.
  • Scale. A huge archive is hard to audit, and one has to check both the conclusion and whether the problem was posed correctly.

The original report says this directly: for now, it is the author's claim, not a proof recognized by the mathematical community. Independent verification of the correctness of the formalization and of its match with the original hypothesis is still needed.

What developers and product managers can take from this

Even if the proof does not hold up, the workflow is instructive. AI here does not replace the human; it suggests an idea to an expert. The model proposes a step, a person evaluates it, and a formal verifier enforces rigor. This is the same pairing already used in code: an LLM generates a variant, and tests and the compiler filter out errors. In mathematics, Lean plays the role of the tests.

The practical takeaways are these:

  • Do not repeat headlines as fact. Until number theorists and Lean experts have weighed in, the correct wording is "claims," not "proved."
  • Read the primary source. The document is on the author's site (alpo.ge/remark.pdf), and specialist mathematicians will be analyzing it.
  • Check not only the result but also the problem statement. This rule applies to your own projects too: if you trust a model's conclusion, first make sure the thing being checked is what you actually need.
  • Follow the audit of the formalization. If the definitions match the canonical ones and no hidden assumptions turn up, this will become one of the most notable examples of LLMs contributing to science. If not, it will remain another lesson in how easily a plausible result can be mistaken for a correct one.

For now, the reasonable stance is cautious interest. Verification will take time, and its outcome will show whether this work enters the history of mathematics or remains a curious episode.

Story

Key Questions Answered

What exactly is claimed?

The Riemann hypothesis was formulated in 1859, so the problem has been open for 167 years. It is one of the seven Millennium Prize Problems of the Clay Mathematics Institute, and a $1 million prize is offered for solving any of them. Roughly speaking, the hypothesis describes how "evenly" prime numbers are distributed along the number line. Many estimates in number theory depend on it. Polishchuk claims to have obtained a proof not only of the hypothesis itself but also of its generalized version (GRH).

What Lean is, and why "verified" does not yet mean "correct"?

Lean is a system in which a proof is written as a program, and the compiler checks every logical step. For a developer, the closest analogy is automated tests. If everything builds without errors, the conclusion follows from the premises within the system. That is why "verified in Lean" carries more weight than "I reread it and it seems to hold." But tests have a well-known problem, and Lean has the same one.

What developers and product managers can take from this?

Even if the proof does not hold up, the workflow is instructive. AI here does not replace the human; it suggests an idea to an expert. The model proposes a step, a person evaluates it, and a formal verifier enforces rigor. This is the same pairing already used in code: an LLM generates a variant, and tests and the compiler filter out errors. In mathematics, Lean plays the role of the tests. The practical takeaways are these: Do not repeat headlines as fact.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…