{ "title": "Reproducing the pentagon calculations", "working_directory": "anc", "commands": [ "python -m pip install numpy==1.26.4 python-flint==0.9.0", "python verify.py" ], "file_names_in_the_manuscript": { "verification.md": "anc/verification.json contains the instructions for this deposit.", "flux-witness.json.gz": "anc/checks/convex/flux-witness.json contains the uncompressed data from the identically named gzip witness in the companion archive." }, "reference_outputs": "anc/reference-replay/*.log (wall times may vary).", "sections": [ "# Reproducing the pentagon calculations", "The paper uses finite calculations in two places. For a nonconvex pentagon, a bound on the Fourier modes gives a strict upper bound for the eigenvalue. For convex pentagons, the calculations enclose the value at the regular pentagon and verify Rayleigh quotient inequalities on the shape regions described in the proof.", "## Running the checks", "From the extracted directory containing `verify.py`, run", "```sh\npython -m pip install numpy==1.26.4 python-flint==0.9.0\npython verify.py\n```", "To run just the two nonconvex calculations, use `python verify.py --quick`. Both use the standard library. The full replay was tested with 64-bit CPython 3.11.8. Leave Python assertions enabled. The driver verifies the SHA-256 digests in `checks/SHA256SUMS.json` before starting any calculation. It runs three jobs concurrently by default. `--jobs 1` runs them in sequence.", "Each calculation writes its output to `replay-logs/`. Use `--logs DIR` to choose another directory. A failed calculation produces a nonzero exit status. The eleven output files in `reference-replay/` come from rerunning these checks for this edition. This replay took 123 seconds with five concurrent jobs on Windows. The witnesses under `checks/convex/` are inputs. The verifiers recompute the inequalities from their trial coefficients and subdivision trees.", "## Comparing the output", "| Part of the proof | Checker | Expected result |\n| --- | --- | --- |\n| Nonconvex pentagons | `check_three_circles.py` | The first 32 modal inequalities and an infinite-tail bound give $F(P)<393329629/38041250<517/50$. |\n| Nonconvex comparison, second implementation | `independent_three_circle.py` | The least surplus in modes 1 through 63 occurs at 23. A separate bound covers every mode from 64 onward. |\n| Regular pentagon | `check_flux.py` | With unit circumradius, μ₁ exceeds 4.354463. The area-normalized value lies strictly between 10.35335162333505933 and 10.35335162333505954. |\n| Local maximum | `certify_local_radius.py` | Checks the mesh inertia counts, the norms and projection errors for the polynomial trial, the two first-variation coefficients, and the relative deficit of at least $r^2/50$ for $0