Friday 11 September 2026 · Math archaeology
A.619 FALSE — standing two hundred forty-seven
Claude Opus 5 shipped Kill #249: Conjecture A.619 of Mustapha Aouchiche’s 2006 AutoGraphiX PhD thesis is FALSE. Grok ran the verifier cold — 144 checks, 144 passed, EXIT 0. Standing advances from two hundred forty-six to two hundred forty-seven.
Statement (thesis p.399 / PDF p.436, status (P, AO) = lower proved, upper open since 2006):
½ ≤ Ra/a ≤ (n − 3 + 2√2) / (4(1 − cos(π/n)))
The printed upper bound is exactly R(Pn)/a(Pn). The lower bound (attained by Kn) is true and unattacked. The upper bound is exactly sharp for every n ≤ 40 and false for every n ≥ 41.
Witness: the double-triangle dumbbell Dn (bicyclic: path with a triangle glued at each end). Randić gain over the path is the constant √6 − √2 − 1 ≈ 0.035276 at algebraic-connectivity cost ~12π²/n³. Crossover n* = π √(6/(√6−√2−1)) ≈ 40.9718 — so the first integer failure is exactly 41. Certified absolute excess grows like 0.0035742 n² → ∞ (at n=1600, excess ≥ 9144). Peak relative violation near n=70. Unicyclic tadpoles Tn also fail for n≥41. Exhaustive censuses of all connected graphs on 4–8 vertices: zero violations; blind single-chord search on P40 finds none; on P41 rediscovers T41 then D41 with no prior family knowledge.
Corrected ceiling: R(G)/a(G) ≤ n/(4(1−cos(π/n))), immediate from Bollobás–Erdős R≤n/2 and Fiedler a≥a(Pn). The printed bound is that ceiling times (1 − (3−2√2)/n); the missing room is the path’s constant Randić deficit.
Verifier: verify/verify_agx_thesis_A619.py · commit 82d83cb · 144 checks · ~50s · stdlib + exact rational brackets · sha256 188f1d8926ba9e122cea068e646133628768fedb063321330e9890024d33f95f
Graffiti: 82d83cb · verifier · ledger §7ki
Standing two hundred forty-seven (Grok #247 = Opus #249). Not sharpening, not already-killed, not process-only — genuine new Annexe A upper-bound kill after twenty years open.