You are attacking Simon Lieb–Thirring (2000 #15) on https://github.com/wustep/maths. Read AGENTS.md and problems/simon-lieb-thirring/ first (q1 is on this branch / PR #93). Covering stays frozen. Do not start a new problem.
q1 dent, not independently replayed by Maths: a certified C1 trial pair gives L_{1,d}/L^{cl} ≤ 1.45576, below FHJN published 1.456 (arXiv:1808.09017 / JEMS 2021). Python 1.455719, Rust 1.455756. Their Lemma 11 second pair does not convert below 1.456. Sobolev 2/√3 untouched. Conversion L/Lcl = (9√3/4) C1. Replay: cd problems/simon-lieb-thirring/compute/q1 && ./run_all.sh
This is q2. Orchestrate up to 8 workers on light search, cap concurrent heavy numeric at 3–4. Aim for a LARGE jump, not another 0.0002 shave.
Targets:
- Replay q1 first (run_all.sh). If it fails, fix the verifier, do not claim 1.45576.
- Search a better C1 pair / a new test-function class that moves L/Lcl by a real amount (toward 1, or at least clearly below 1.45). Write the imagined certificate first.
- Fetch any paper after FHJN that might have already beaten 1.456 before you claim priority.
- Dent = verified finite improvement of the published record with verifier+cert. Residue if you cannot beat 1.45576.
Update ATTACK.md, WALKTHROUGH.md, PROBLEM.md, README table/ledger. Open one PR on a q2 branch. If #93 is not yet on main, branch from this q1 branch and do not fight the merge agent. Rust/C/Python/Lean as they fit.