Did the neural network prove the Riemann hypothesis: what is known about OpenAI's results and the verification of proofs

Based on the materials available for this analysis, it cannot be claimed that the neural network solved the Riemann hypothesis. The materials discuss narrower claims, individual mathematical results, and proofs that, according to the description, can in some cases be checked using the Lean system. This information does not provide grounds for declaring one of the most famous unsolved problems in mathematics solved.

These are different levels of claims.

One distinction lies between the statements “the model produced an interesting result” and “the mathematical community recognized a proof of the original hypothesis.” The former describes the system's work; the latter requires verification, reproduction, and substantive evaluation by specialists. This article does not draw the second conclusion.

What exactly is known about OpenAI's claims

One of the areas discussed in connection with OpenAI concerns applying reasoning models to mathematical research problems. The materials available for this analysis contain no report of a complete proof of the Riemann hypothesis. They describe the system's ability to find ideas, construct chains of reasoning, and correct individual errors. They also mention the possibility of formalizing a result in a language suitable for computer verification, but the relevant documents are absent from the set of materials. Therefore, these details are presented here as a retelling of claims, not as an independently confirmed conclusion.

A suggestion has appeared online that model GPT-7 may be behind such results. There is no official confirmation of this in the available materials, so it should be regarded as a rumor. Nor are reports confirmed that OpenAI supposedly solved 90 of the world's 500 most important mathematical problems. The order of the problems in the published list does not make it possible to assess their scientific significance, and the view that approximately one-fifth of the results are refutations is not official OpenAI statistics.

A cautious formulation for the available materials would be as follows: they attribute mathematical results to OpenAI, some of which, according to the descriptions, are accompanied by formal proofs or materials for subsequent verification. A list of such works and supporting documents is not attached here. The evaluation of each result depends on the theorem's exact formulation, the completeness of the proof, its novelty, and its connection to the original scientific problem.

What is the Riemann hypothesis

Within the scope of this analysis, the Riemann hypothesis is described as a statement about the nontrivial zeros of the Riemann zeta function. The article uses the connection between this function and the distribution of prime numbers, while the zeros are considered in the complex plane, whose coordinates include real and imaginary parts.

The formulation used here states that every nontrivial zero of the zeta function has a real part equal to the same value. This vertical line is called the critical line. In the text, this formulation is linked to the description of deviations in the distribution of prime numbers, although it itself concerns the properties of a complex function.

Checking a finite number of zeros within the accepted framework does not replace a statement about all possible zeros. Even confirmation of the hypothesis within the ranges checked does not rule out a counterexample beyond the calculations. The transition from finite verification to a statement about all cases remains a separate logical step.

In the provided materials, the Clay Mathematics Institute is named as the organization that classifies the Riemann hypothesis among the Millennium Prize Problems and promises a one-million-dollar prize for its proof. A link to the institute's page is absent, so in this article the problem's status and the prize amount are given only as information from the source text, not as information verified against the primary source.

Why the quasi-Riemann hypothesis does not solve the original problem

One of the works about which reports circulated claims to prove the so-called quasi-Riemann hypothesis. In the original description, this term is associated with a weaker statement, functions, and methods used in the study of prime numbers. The precise mathematical relationship cannot be established without the work itself. Within the limits of the description provided, a proof of a particular or weakened property should not be considered a proof of the full Riemann hypothesis.

The difference can conveniently be represented in terms of scope. The full hypothesis, in the formulation used here, requires control over all nontrivial zeros of the zeta function and their location on the critical line. A weaker statement may describe a particular class of functions, a limited set of conditions, or a consequence of established theory. Without the work itself, the exact strength of the result cannot be determined, but its name alone does not provide grounds for considering the original problem solved.

University of Toronto mathematics professor Daniel Litt, analyzing one of these works, assessed the result as a narrow special case of a more general problem and noted that it follows from a stronger investigation being conducted by one of his students. In the available materials, this is presented as an expert observation. It does not prove that the work is erroneous, but it shows why a sensational headline must be compared with the precise formulation of the result and with existing research.

What is known about Catalan's constant

The published work also claims to have proved the irrationality of Catalan's constant. In the original text, irrationality is explained as the impossibility of representing a number exactly as the ratio of two integers. This report concerns a separate mathematical problem and, by itself, does not claim to prove the Riemann hypothesis.

Within the available materials, this should be regarded as an unconfirmed report. Even if the proof turns out to be correct, evaluating the result will depend on the completeness of the argument, the reproducibility of every step, and its connection to already published theory. No documents for such verification are provided here.

How Lean verifies a mathematical proof

The article describes Lean as a formal tool for verifying mathematical proofs. In the explanation adopted for this text, a researcher records definitions, initial assumptions, theorems, and transitions in a strict formal language, after which the system checks the recorded chain according to the rules of the chosen formal theory. Lean's documentation is not attached, so this is a working explanation, not a conclusion drawn from a verified primary source.

According to the article's description, such verification differs from ordinary reading. A person may overlook an implicit assumption, an error in an index, or a transition that seems obvious but does not logically follow from the preceding steps. Here Lean is presented as a system that checks the formal correctness of the recorded chain, not the persuasiveness of the exposition or the scientific value of the result.

This approach has limitations stated in the article itself. Lean checks a specific statement in a specific formalization, but it does not decide for the researcher whether the mathematical formulation was chosen correctly. If the wrong theorem, an overly strong axiom, or a statement that does not correspond to the original problem was entered into the system, a successful verification by itself does not mean that the initial problem has been solved.

The article distinguishes several levels of control:

• the model proposes a mathematical statement or line of reasoning;
• the researcher clarifies the formulation and checks that it concerns the intended problem;
• the formal system checks the logical transitions in the recorded proof;
• mathematicians assess the result's novelty, significance, and connection to already published works.

Based on the available materials, it cannot be established which specific OpenAI studies underwent formal verification or to what extent. The article says that proofs suitable for automated analysis were prepared for some results, while they are absent for others, but no list of works or supporting documents is attached. Therefore, this information should be understood as a retelling of the original claims. The possibility of error remains even when the reasoning text appears detailed and consistent.

Why independent expert review remains necessary

Within the logic of this article, a mathematical proof must answer not only the question “does the conclusion follow from the premises?” but also “is the exact result claimed by the author proved?” In such an assessment, an independent specialist compares the definitions, scope of applicability, hidden conditions, and connection to previous research. For this article, this is a description of a possible procedure, not a reference to a separate prescribed review process.

A new result may, during such a review, turn out to be a special case of a known theorem, a renamed version of an existing statement, or a consequence of a stronger work. In that case, the formal proof may still be error-free, but the scientific novelty may be less than suggested by a press release or headline. This is a general evaluation scenario, not an established conclusion about a specific OpenAI work.

And what happens if the computer confirms every step, but the formulation of the problem turns out to be too weak? The formal theorem recorded in the system will be confirmed, but not necessarily the original scientific hypothesis. Peer review and discussion with independent mathematicians remain separate stages of verification in such a case.

Based on the available materials, it cannot be concluded that all the published results have already been definitively recognized by the mathematical community. Such a conclusion would require accessible proofs, reproducible formalization, and independent analysis of how the results relate to the existing literature.

What AI really changes in mathematics

In the article's materials, reasoning models are considered tools for finding mathematical ideas and checking individual solutions. They are credited with exploring approaches, maintaining a long chain of conditions, finding contradictions, and translating part of an argument into a formal language. These capabilities are presented as observations and claims from the original materials, not as the result of an independent comparative study.

Additional computations, specialized tools, and automatic verification, in this description, make systems more useful for tasks involving control of intermediate steps than a chatbot that produces a quick answer without such verification. No direct comparative study is available in the materials, so this is the author's assessment of the described approach, not an established general fact.

This information by itself does not prove that artificial intelligence understands mathematics in the same way as a human. It describes the ability to perform certain kinds of research work that, according to the reports cited, were previously associated with highly qualified specialists. In the article, scientific understanding is linked to posing significant questions, choosing a direction, explaining the result, and assessing its place in the broader theory.

A computer may formally verify a proof that is difficult for a human to understand completely. This is a possibility mentioned in the materials, not an established property of a specific OpenAI system. In such a case, the article's author considers transparency, independent reproduction, and explanation of the result important. The greater the role of AI in mathematical research, the more visible the distinction becomes between generating an idea, formally verifying it, and scientifically interpreting it.

Conclusion: did the neural network prove the Riemann hypothesis

Based on the available materials, there are no grounds for considering the Riemann hypothesis proven. The claims examined in the article concern mathematical research, individual assertions, and formal verification tools. The report of a proof of the quasi-hypothesis describes a result that is presented in the source material as weaker, rather than as a solution to the full Riemann hypothesis. The report concerning the irrationality of Catalan's constant requires independent scholarly verification.

The most cautious conclusion is as follows: in the available materials, AI is presented as a tool for mathematical discovery and formalization, but a sensational claim about a result does not yet confirm the solution of a major unsolved problem. With regard to the Riemann hypothesis, within the scope of this review, a precise formulation, proof of all the claimed cases, and independent evaluation by the mathematical community remain essential.