One bound state, always, proved at last?
Two AI preprints on the bound states of the BFSS matrix model
Covers result 270 in OpenAI’s catalog: all 2 papers. See its connections to other results and the work it cites.
3:41. English captions available. The video loads from YouTube when you press play.
Chapters
The claim, in plain words
In 1996, Banks, Fischler, Shenker and Susskind proposed that M-theory, a candidate theory of quantum gravity, could be described by quantum mechanics of very large matrices: nine N×N matrices for N particles called D0-branes, with sixteen supersymmetries. For the idea to work, the N particles must form exactly one bound state with zero energy, standing for a graviton carrying N units of momentum.
That is delicate. The potential has flat valleys where the matrices commute, so particles can drift apart for free, and the energies form a continuum starting at zero. The bound state has to sit exactly at the edge of that continuum. The original paper said a direct proof had not yet been given. Index calculations in the late 1990s made it convincing, 5/4 − 1/4 = 1 state for two particles in ten dimensions and a net count of zero in four and six, but a full proof stayed open; a 2021 paper still called it an open problem.
Result 270 is two OpenAI preprints, 57 pages, dated September 24 and October 5, 2026. The first claims that for every N of two or more, the model with the centre of mass removed has exactly one normalizable zero-energy state, and that it is rotation invariant. For two particles, existence and uniqueness had already been argued; the new content is three or more. The second claims that for two particles there are also infinitely many normalizable states with positive energy, sitting inside the continuum. That contradicts a sentence of the original paper: “No other normalizable bound states can occur.” The preprint notes that the large-N conjecture is a separate question.
What it doesn’t claim
Anything about large N, gravity or M-theory itself, other numbers of dimensions, or the values of the positive energies.
Where it stands
- Posted as preprints
- OpenAI, two papers, 57 pages, dated September 24 and October 5, 2026
- Peer reviewed
- Not yet
- Checked in Lean
- No
Last checked October 19, 2026. The preprints have not been revised. Neither theorem is formalized in Lean; the release contains about 18,000 lines of Lean for building blocks of the first proof, but not the theorem. OSVerify has not reviewed them, and we found no comment from specialists.
How the proofs go
- Counting with partitions. Switching on a mass deformation gives exactly one ground state per partition of N, a way of writing N as a sum of whole numbers: 5 for N = 4, 42 for N = 10. As the deformation is removed, every partition with two or more parts drifts apart into separate clusters, each a smaller bound state known by induction. That leaves room for exactly one state at the centre, and forces it to exist.
- Trapped by spinning. To escape, the matrices must line up along one direction, which cannot carry rotation in two planes at once, so states spinning fast enough in two planes are trapped. By our estimate of the paper’s constants, the argument covers only very fast spinning sectors, with second spin label above about 37. That says nothing either way about lower sectors.
The numbers in the video
- Ours: a two-variable toy without fermions has a discrete spectrum (lowest energy 1.108); with fermions added, the lowest energy falls toward zero as the box grows, from 0.40 to 0.025.
- Ours: the index arithmetic 5/4 − 1/4 = 1 and 1/4 − 1/4 = 0, and the partition counts 2, 3, 5, 7, 11, …, 42.
- Ours: the second paper’s constants made explicit: C ≈ 13.06 and M = 6, so its argument applies above a second spin label of about 37.
Read the source
- The Unique Threshold Bound State of the SU(N) BFSS Model (PDF), 48 pages.
- Positive Eigenvalues of the Relative SU(2) BFSS Hamiltonian (PDF), 9 pages.
- The collection’s catalog, where this is result 270.
The 2 papers in this result
OpenAI groups related papers, such as alternative proofs, supporting results and consequences, into one result; we label the first paper it lists the main paper and the others companions.
Transcript
Read the full script
0:00The question
In 1996, Banks, Fischler, Shenker and Susskind proposed that M-theory, a candidate theory of quantum gravity, could be described by quantum mechanics of very large matrices: nine N×N matrices for N particles called D0-branes, with sixteen supersymmetries. For the idea to work, the N particles must form exactly one bound state with zero energy, standing for a graviton carrying N units of momentum. That is delicate. The potential has flat valleys where the matrices commute, so particles can drift apart for free, and the energies form a continuum starting at zero. Our computer shows the knife edge in a two-variable toy. Without fermions, flat valleys still trap: the lowest energy stays at 1.11 however large the box. With fermions added the trapping cancels, and the lowest energy falls toward zero as the box grows, like a continuum. So the bound state has to sit exactly at the edge of that continuum. The original paper said a direct proof had not yet been given. Index calculations in the late nineties made it convincing: 5/4 − 1/4 gives one state for two particles in ten dimensions, and a net count of zero in four and six. But a full proof stayed open; a 2021 paper still called it an open problem.
1:19The claims
Result 270 is two OpenAI preprints, 57 pages, dated September 24 and October 5, 2026. The first claims that for every N of two or more, the model with the centre of mass removed has exactly one normalizable zero-energy state, and that it is rotation invariant. For two particles, existence and uniqueness had already been argued; the new content is three or more. The second claims that for two particles there are also infinitely many normalizable states with positive energy, sitting inside the continuum. That contradicts a sentence of the original paper: no other normalizable bound states can occur. The preprint notes that the large-N conjecture is a separate question. Not claimed: anything about large N, gravity or M-theory itself, other numbers of dimensions, or the values of the positive energies.
2:16How the proofs go
The first proof counts. Switching on a mass deformation gives exactly one ground state per partition of N, a way of writing N as a sum of whole numbers: five for N = 4, 42 for N = 10. As the deformation is removed, every partition with two or more parts drifts apart into separate clusters, each a smaller bound state known by induction. That leaves room for exactly one state at the centre, and forces it to exist. The second uses rotation. To escape, the matrices must line up along one direction, which cannot carry rotation in two planes at once, so states spinning fast enough in two planes are trapped. Our estimate of the paper’s constants: the argument covers only very fast spinning sectors, with its second spin label above about 37. That says nothing either way about lower sectors.
3:08Where it stands
The preprints are unrefereed and unchanged since release. Neither theorem is formalized in Lean; the release contains about 18,000 lines of Lean for building blocks of the first proof, but not the theorem. OSVerify has not reviewed them, and we found no comment from specialists. If they hold, the bound state that Matrix Theory needs exists for every N, exactly once, and the old sentence ruling out other bound states is false for two particles.
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.