What two maths manuscripts predict about measurable physics
Two reports from 22 October read manuscripts on fracture and on interfaces in a cubic torus for physical consequences, and mostly found limits. A wrong fracture model can fit well and be about 27% low, and the torus effect vanishes as the cell grows.
A third track of this research asks the same manuscripts whether they say anything about a physical thing you could measure. Two reports from 22 October, on fracture and on a cubic torus, answer mostly with limits.
The repository began with two tracks, mathematics and industry, and the natural-science track now sits beside them. The record does not say why it was added. It is kept apart: the industry reports state “no natural-science research”, and the science index calls itself a science-only record whose manuscript claims and numerical tests “are not independent proof verification”. Its earliest report is from 14 October, and by 22 October its runs folder held 39 files.
The fracture theorem cannot predict real crack branching
The manuscript on interior regularity of planar Mumford-Shah minimisers (solutions of a model that balances smooth regions against the length of the breaks between them) was audited by the mathematics track on 14 October. On 22 October the science track asked what it means for fracture.
It cannot predict real 120-degree crack branching, because real fracture has vector elasticity, anisotropy, loading and rate effects outside the assumptions. Its heavy-tail estimate is the classical square-root tip singularity, which is already known.
A wrong model can fit well and be about 27% low
What the report did produce was a synthetic experiment. Take a hypothetical material (μ = 30 GPa, fracture energy 100 J/m²) with a process zone (the small region at a crack tip where the material gives way) of 5 µm. Generate 400 noisy datasets at 25 radii from 2 to 500 µm. Fit a pure singular law and a law with a cutoff.
The wrong model gives a median toughness of 0.726 of the truth, about 27% low, with a fit that looks acceptable; the cutoff model recovers 1.0001. Using only the outer radii, the error shrinks to about 4% but the cutoff is poorly identified (fitted values from 0.46 to 10.4 µm). The report says these are not experimental intervals and establish no new law.
It also notes that a preprint submitted on 22 September 2026 predates the manuscript's date, and it adjudicates nothing about priority.
The torus effect vanishes as the cell grows
The second report took a manuscript on isoperimetry (which shape encloses a given volume with the least surface) in a cubic flat 3-torus, a box whose opposite faces are joined so that anything leaving one side comes back on the other. For each volume fraction v it asks which interface shape, a ball, a tube or a slab, has the least area. The thresholds are v = 4π/81 ≈ 0.155140 and v = 1/π ≈ 0.318310. The run reproduced both crossings to better than 10^-65 with 70-digit arithmetic.
The physical translation is labelled conditional. With an illustrative surface tension of 0.03 J/m² and a 100 nm cell, the pressure jumps would be 0.450 and 0.942 MPa, and the report states they are not observed pressures.
The interface energy per volume goes to zero as the cell grows, so these are finite-cell geometry effects rather than a new bulk phase transition. In a 2:1:1 box the thresholds are exactly half, 2π/81 and 1/(2π), so the numbers do not carry over to other box shapes.
Every science entry ends with a limit
Each science entry in the list of possible applications sorts what follows into high and medium priorities, then a limit. The limits are the same few lines: no anomaly explained, no experimentally testable prediction, prior art exists. Fracture is also what the broken glass on this site simulates, but nothing in the run record connects the two.
Sources and notes (12)
- Isoperimetric conjecture for the cubic flat 3-torus (OpenAI manuscript, pinned) — the manuscript audited as SCI-354
- Hauswirth, Perez, Romon and Ros, The periodic isoperimetric problem — earlier geometry for the torus
- Milman, Isoperimetric inequalities on slabs — a partial range already proved
- Block et al. (2010), periodic Lennard-Jones simulation — sphere-cylinder-slab morphologies already found
- Interior regularity of planar Mumford-Shah minimizers (OpenAI manuscript, pinned) — the manuscript audited as SCI-366 (TeX source)
- Deangelis, arXiv:2609.26732v1 (22 Sep 2026) — a preprint claiming the same conjecture, dated earlier
- De Lellis and Focardi (2025) — weak-L4 equivalence
- Mumford and Shah (1989) — the original image model
- Science index — the notes this was written from
- Science opportunities — the notes this was written from
- SCI-366 report — the notes this was written from
- SCI-354 report — the notes this was written from
Cryptography researchAll 10 parts →
