# OpenAI Proves The Rational Hodge Conjecture For CM Abelian Varieties: Understand What This Means > Source: > Published: 2026-10-07 07:51:58+00:00 *OpenAI has published a paper claiming a proof of the rational Hodge conjecture for a special but important class of geometric objects. It is a partial result on one of mathematics’ most famous unsolved problems, and it comes with several big downstream consequences.* OpenAI has [released](https://officechai.com/ai/openai-releases-over-300-mathematical-results-produced-by-an-internal-model/) a paper titled “The rational Hodge conjecture for CM abelian varieties,” dated September 30, 2026. It claims to settle one of the most celebrated open questions in mathematics for an important family of cases. The paper lists OpenAI as its author rather than naming individual researchers. The result does not solve the full [Hodge conjecture](https://officechai.com/ai/openai-releases-ai-generated-proofs-of-quasi-riemann-hypothesis-unique-games-conjecture-hodge-conjecture-for-cm-abelian-varieties/), which remains one of the seven Millennium Prize Problems and carries a $1 million prize from the Clay Mathematics Institute. But if the proof holds up, it is a substantial step, and it knocks down two other long-standing conjectures as a bonus. Here is what it all means, starting from scratch. ## The Hodge conjecture, explained without the jargon Mathematicians who study shapes have two very different ways of looking at the same object. **The algebraic view.** Some shapes are cut out by polynomial equations. A circle is the set of points satisfying x² + y² = 1. A sphere, a cone, and a curved surface defined by a cubic equation all work the same way. These are the “algebraic” shapes, and they are built from precise formulas. **The topological view.** Here you ignore formulas and look only at the overall structure of a shape: how many holes it has, how it loops back on itself, what kinds of surfaces can sit inside it. Mathematicians capture this information with an invariant called cohomology, which you can think of as a catalog of the shape’s “features.” Now consider a complicated, smooth shape defined by polynomial equations (technically, a smooth projective complex variety). Its catalog of topological features is enormous. Some entries in that catalog are special: they pass a technical test, named after the British mathematician W.V.D. Hodge, that says they are compatible with the shape’s underlying complex-number structure. Features that pass the test (and are described with rational numbers) are called **Hodge classes**. Separately, you can take smaller shapes that sit inside the big one and are themselves defined by polynomial equations. Think of a curve drawn on a surface, or a surface inside a four-dimensional space. These are **algebraic cycles**, and each one leaves a footprint in the catalog. **The Hodge conjecture says that these two things are really the same.** Every feature that passes the Hodge test should be buildable by combining footprints of algebraic cycles. In plain terms: if something looks like it could come from polynomial equations when viewed through topology, then it really does come from polynomial equations. A loose analogy: imagine a vast library where some books are labeled “could have been written by this one author” because the style matches. The Hodge conjecture says every such book really was written by that author, or can be assembled from the author’s known works. Nobody has ever found a book that matches the style but can’t be traced back. Nobody has been able to prove that none exists, either. ### Why the “rational” part matters The word “rational” in the title refers to the kind of arithmetic allowed when combining algebraic cycles: you may use fractions, not just whole numbers. This matters because the stricter integer version of the conjecture is known to be false. Mathematicians discovered counterexamples decades ago. The rational version is the one that is still believed true and still open. ### Where things stood before The conjecture was proposed by Hodge in 1950 and has resisted proof for more than seven decades. It is known in a few easy situations, such as the lowest and highest-dimensional cases, but is unproven in general, even though it is widely believed. Progress has come in scattered special families, and general attacks have repeatedly stalled. ## What OpenAI says it proved The paper targets **CM abelian varieties**. Two terms need unpacking. **Abelian varieties.** These are higher-dimensional cousins of the donut shape. An elliptic curve, the kind that appears in cryptography and in the proof of Fermat’s Last Theorem, is a one-dimensional abelian variety. Higher-dimensional versions have rich structure and are central to modern number theory and geometry. **CM (complex multiplication).** Some abelian varieties have an unusually high amount of hidden symmetry. The paper defines the CM condition precisely: the variety’s endomorphisms (the ways it can map to itself) must contain a commutative algebra of dimension twice the variety’s dimension. Mathematicians view these highly symmetric varieties as the most arithmetically structured examples, and they have long been a testing ground for the Hodge conjecture. **Theorem 1.1 of the paper** states that for any complex CM abelian variety and every codimension, every rational Hodge class is a rational combination of algebraic cycle classes. In other words, the rational Hodge conjecture holds for these varieties in every dimension. The paper adds that it also holds for finite products of CM abelian varieties and for powers of such products. Notably, the claim covers *every dimension*, not just low-dimensional cases where earlier researchers made progress. ## The bonus results The more eye-catching part of the abstract is what follows from the main theorem. The paper says it yields three further results: 1. **The generalized Hodge conjecture for CM abelian varieties.** This is a refined version of the conjecture that predicts where algebraic support for a Hodge structure should live. The paper obtains it by building on earlier work of Fumio Hazama and Salman Abdulali, who had shown the generalized version follows from the ordinary one for CM varieties. 2. **The Tate conjecture for abelian varieties over finite fields.** This is the number-theoretic cousin of the Hodge conjecture. It concerns algebraic varieties defined over finite fields (number systems with finitely many elements, the setting for much of cryptography and coding theory). The conjecture says certain symmetric cohomology classes always come from algebraic cycles. The paper reaches it using theorems of James Milne, who had previously linked it to the Hodge conjecture for CM abelian varieties. 3. **The Hodge standard conjecture for abelian varieties in arbitrary characteristic.** This one is a positivity statement about how algebraic cycles intersect. It is a key part of the broader “standard conjectures” program that Alexander Grothendieck launched in the 1960s to build a universal theory of algebraic cycles. Again, Milne’s earlier work provides the bridge. That structure is worth noticing: the new proof is the key that unlocks implications mathematicians had already worked out. Milne’s results said, roughly, “if the Hodge conjecture holds for CM abelian varieties, then these other statements follow.” OpenAI’s paper claims to supply the missing “if.” ## Why this is significant **For mathematics.** The Hodge conjecture is notoriously hard to attack. A complete proof for an infinite family of varieties, plus three corollaries across number theory and algebraic geometry, would be one of the most meaningful advances on the problem in a long time. The Tate and Hodge standard conjectures in particular have been open for decades. **For the Millennium Prize.** This does not win the prize. The prize requires the conjecture for *all* smooth projective complex varieties, and CM abelian varieties are a narrow slice. Still, the CM case is considered one of the natural footholds, because so much theory is known to reduce to it. **For AI.** This may be the bigger story for readers of this publication. OpenAI has been pushing hard to show its models can do frontier research mathematics. In September the company claimed a solution to the Navier-Stokes problem, another Millennium Prize Problem, which drew pushback from some mathematicians over verification and credit. Press reports around mid-September, citing an anonymous source, said OpenAI staff expected to crack the Hodge conjecture soon. A paper on exactly this topic arrived two weeks later. Reports also suggest that OpenAI sees mathematics as a proving ground for AI systems capable of genuine reasoning, with commercial and capability implications well beyond pure math. ## The caveats A few things are worth keeping in mind before declaring victory. - **It is a preprint.** The paper has not been peer reviewed. Proofs of this depth normally take months of scrutiny by specialists before the community accepts them. - **Independent verification is pending.** Mathematicians outside OpenAI will need to check the argument line by line. Whether the paper comes with a machine-checkable formalization is something to watch for. - **Process and credit questions linger.** How much of the proof was generated by AI, how much by human researchers, and how the work builds on existing literature are all questions the community is likely to probe, especially given the earlier Navier-Stokes dispute. - **The scope is limited.** The result covers CM abelian varieties and what follows from them, not the Hodge conjecture in full generality. ## What to watch next If expert reviewers confirm the argument, expect rapid follow-up work on whether the techniques extend beyond CM abelian varieties, since that is the path toward the full conjecture. Also watch for statements from number theorists, whose field would be most directly affected by the Tate result. For now, the headline is straightforward: OpenAI is claiming a genuine, nontrivial theorem on a problem that has defeated mathematicians for three-quarters of a century. Whether it stands will be decided the way these things always are, by experts reading the proof.