What was already known
Rupert Li’s discrete dual method gave a lower bound of 1/24 on the linear-programming optimum in both dimensions. Cohn, de Laat and Salmon obtained the packing upper bounds used for comparison.
AI involvement
The authors define the optimization targets and evaluation rules. Épi searched for the retained dual witnesses and the project used a separate implementation to check the exact inequalities.
- Épi: Towards a Persistent, Self-Improving Agent for Autonomous ResearchÉpi: Towards a Persistent, Self-Improving Agent for Autonomous Research
Importance
The two certificates prove the Cohn–Elkies linear program cannot attain the true sphere-packing density in dimensions 10 and 11.
Technical verification
The artifact reports exact support and discrete Fourier inequality checks for the released dual vectors. The comparison uses the published three-point upper bounds in the same normalization.
Formal checking
The artifact reports exact support and discrete Fourier inequality checks for the released dual vectors. The comparison uses the published three-point upper bounds in the same normalization.
- Checker
- Épi authors, reporting a separate exact implementation
- Scope
- Support and discrete Fourier inequalities for the final dual vectors in dimensions 10 and 11.
- Basis
- Reported by the linked checker
- Versioned exact dual certificatesVersioned exact dual certificates
Human understanding
The source explains the comparison and the certificate workflow. It does not document a human explanation of how Épi found the retained witnesses.
Correction and publication history
Historical curation based on the cited claim, AI attribution and checking scope.
Read this published revision