VibeMathedMath problems solved by AI

Tightness of the Cohn-Elkies Bound in Dimension 36

Can a Cohn-Elkies auxiliary function certify the best known sphere packing in dimension 3636 as optimal? No. An explicit dual-feasible point for the Cohn-Elkies linear program, built from weight-1818 modular forms for Γ0(24)\Gamma_0(24), shows the two-point linear programming bound in dimension 3636 exceeds the density of the Kschischang-Pasupathy packing by a factor of at least 32.9132.91.

Result
Disproved(see note)
Status
Resolved
AI contribution
AI co-developed
Method
Construction
Field
Sphere packing
Posed by
Henry Cohn, Noam Elkies
Year posed
2003
Years open
23y
Solved
2026-07-13
Model
Claude Fable 5, Claude Opus 4.8, Codex (GPT-5.6)
Vendor
Anthropic / OpenAI
Collaborators
Rifat Jumagulov
Verification
Unreviewed
Publication
Preprint
Significance
25 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

rules out the two-point LP method in this dimension; the optimal packing in dimension 36 remains unknown

What the AI did

The disclosure reports substantial assistance: the Claude models were used for the construction of the certificate itself, for the verification tooling and for drafting, and Codex was used as an independent cross-check of the certificate computations.

Verification

The result is an explicit dual-feasible certificate, so it is checkable in principle by evaluating the constructed function; the paper reports an independent cross-check of the computations by a second model. We have not reproduced it. arXiv preprint, not peer-reviewed.

Source

arXiv:2607.11319 - A dual linear programming bound for sphere packing in dimension 36

Discussion