One operator for all?
An AI preprint claims every type II₁ factor with separable predual is singly generated
Covers result 296 in OpenAI’s catalog: its one paper. See its connections to other results and the work it cites.
4:19. English captions available. The video loads from YouTube when you press play.
Chapters
The claim, in plain words
One matrix can generate a whole matrix algebra: a generic 8×8 matrix generates all 64 dimensions, while a normal one gives at most 8, and two self-adjoints packed as a + ib make one generator. Kadison’s generator problem, from an unpublished 1967 problem list, asks whether every von Neumann algebra with separable predual is generated by two self-adjoints, and names the free group factor on three generators. Type I, hyperfinite and properly infinite algebras were settled in the 1960s, and in 1974 Willig reduced the problem to type II₁ factors. Later work covered factors with a Cartan subalgebra and with property Γ, but not the free group factors on three or more generators, where many experts expected a counterexample.
Result 296 is one OpenAI paper, 18 pages, dated September 23, 2026, and published October 6. Its main theorem: for an irreducible inclusion of type II₁ factors with separable predual, the unitaries that, together with the smaller factor, generate the larger one form a dense Gδ set in the trace 2-norm. As a consequence, every type II₁ factor with separable predual is generated by two self-adjoint operators, so by a single operator. With the classical reductions, every von Neumann algebra with separable predual would be singly generated, and Voiculescu’s free entropy dimension would depend on the choice of generators for the free group factors on three or more.
What it doesn’t claim
Anything about whether the free group factors are isomorphic, the question result 287 claims to settle; nonseparable preduals; or an explicit generator.
Where it stands
- Posted as a preprint
- OpenAI, 18 pages, dated September 23, 2026
- Peer reviewed
- Not yet
- Formalized in Lean
- Both main statements (docs)
Last checked October 19, 2026. The paper has not been revised. In Lean, both main statements are formalized and match the paper, together in about 36,000 lines, with no unproved steps in our scan (we did not build them). OSVerify has not reviewed it. We found no expert evaluation, though three later preprints already use it as an input.
How the proof goes
- A nudge spread across a long word. Popa’s theorem gives unitaries in the ultrapower of the smaller factor that are free from any given separable piece. A tiny nudge to the unitary, spread across a word with many copies of it, keeps a fixed correlation with any missing direction.
- A typical unitary generates. Haagerup’s inequality keeps the error under control at every word length, and a Baire category argument shows that a typical unitary generates.
- Down to two. Popa’s irreducible hyperfinite subfactor gives two self-adjoint generators and a typical unitary two more; on a half-size corner, a 2×2 matrix trick packs the four into two.
The numbers in the video
- Ours: words in two random 8×8 self-adjoints span 1, 3, 7, 15, 31, 63 and then all 64 dimensions; none of 2,000 random 4×4 matrices fails to generate.
- Ours: a random-matrix model with 400×400 matrices: as the unitary appears 1 to 64 times in the word, the nudge shrinks from about 0.14 to 0.018 while the correlation stays near 0.1; a random nudge of the same size gives almost nothing.
Read the source
- Relative Generation and the Generator Problem for Finite Factors (PDF), 18 pages.
- Hayes on W*-bundles and generation (arXiv, September 2026).
- The Lean formalization notes.
- The collection’s catalog, where this is result 296.
The paper
OpenAI groups related papers into one result; this one has a single paper.
Transcript
Read the full script
0:00The question
One matrix can generate a whole matrix algebra. Our computer finds that a generic 8×8 matrix generates all 64 dimensions, while a normal one gives at most 8. Two random self-adjoint matrices do the same: the words in them span 1, 3, 7, 15 dimensions and onward until the space is full, and packing them as a + ib turns two self-adjoints into one generator. Among 2,000 random 4×4 matrices, none fails. Kadison’s generator problem, from an unpublished 1967 problem list, asks whether every von Neumann algebra with separable predual is generated by two self-adjoints, and names the free group factor on three generators. Type I, hyperfinite and properly infinite algebras were settled in the 1960s, and in 1974 Willig reduced the problem to type II₁ factors. Later work covered factors with a Cartan subalgebra and with property Γ, but not the free group factors on three or more generators. The free group factor on two generators is generated by two self-adjoints from the start; the question was the factors on three or more. Many experts expected a counterexample there, perhaps through Voiculescu’s free entropy dimension, and in September 2026 one wrote that the general belief is that there should be a separable von Neumann algebra which is not singly generated.
1:30The claim
Result 296 is one OpenAI paper, 18 pages, dated September 23 and published October 6, 2026. Its main theorem: for an irreducible inclusion of type II₁ factors with separable predual, the unitaries that, together with the smaller factor, generate the larger one form a dense G_δ set in the trace 2-norm. As a consequence, every type II₁ factor with separable predual is generated by two self-adjoint operators, so by a single operator. With the classical reductions, every von Neumann algebra with separable predual would be singly generated, and Voiculescu’s free entropy dimension would depend on the choice of generators for the free group factors on three or more. Not claimed: anything about whether free group factors are isomorphic, nonseparable preduals, or an explicit generator.
2:27How the proof goes
The proof uses Popa’s theorem that the ultrapower of the smaller factor contains unitaries free from any given separable piece. A tiny nudge to the unitary, spread across a long word, keeps a fixed correlation with any missing direction. Our random matrix model shows it with 400×400 matrices: the nudge shrinks from about 0.14 to 0.018 as the unitary appears 1 to 64 times in the word, yet the correlation stays near 0.1, while a random nudge of the same size gives almost nothing. Haagerup’s inequality keeps the error under control at every word length, and a Baire category argument then shows that a typical unitary generates. Our computer sees a finite stand-in: with the diagonal matrices as the smaller algebra, none of 1,000 random unitaries fails, and every bad unitary we tried is cured by a nudge of 0.001. Finally, Popa’s irreducible hyperfinite subfactor gives two self-adjoint generators and a typical unitary two more; on a half-size corner, a 2×2 matrix trick packs the four into two.
3:39Where it stands
The paper is unrefereed and unchanged since release. In Lean, both main statements are formalized and match the paper, together in about 36,000 lines; our scan finds no unproved steps, but we did not build them. OSVerify has not reviewed it, and we found no expert evaluation, though three later preprints already use it as an input. If it holds, every type II₁ factor with separable predual is generated by a single operator. On its own it says nothing about whether the free group factors are isomorphic, the question result 287 claims to settle.
Spot a mistake?
This episode simplifies on purpose, but it shouldn’t be wrong. If something is, open an issue on GitHub or leave a comment on YouTube. Corrections are noted here with a date.