White's choice of R is already at his method's ceiling (simplified-program gain 2.9e-6 from R=20 to R=320), one exact dual point 0.37399241331 was accepted, and the full program was not reproduced to 0.379005 because the Parseval cone needs a conic solver this run did not have.
hunts/overlap_lower (Record 63 of 98 in chronological sequence)
Guiding Question
is the published lower bound on Erdos's minimum overlap constant the limit of its own convex program, or the point at which the search stopped
Method & Verification
White's simplified and full programs were reconstructed and swept in R, with one dual point accepted in exact directed-rounded rationals and four cutting-plane attempts failing to enforce the Parseval cone.
Lineage & Relationships
Primary Sources (at pin 8fa46e134)
Editorial Notes
Three opening-brief premises were wrong at the source: the lower bound has moved (Kim-Pilanci 0.37912, 2026-06-30 preprint), the catalogue entry has no asterisk, and 0.000059 of upper-bound movement is 2016 human minus 2026 AI, not twelve months of AI. White discretises M and never f, so the unsafe direction hunt r_828c8b worried about does not arise. No bound on C is claimed.
Date Provenance
commit 57a3166cf4c0d62cf32001e378912204f560ee54, hunts/overlap_lower/RESULTS.md, author 2026-08-23T22:24:47-05:00