source-evidence.md

back to table · edit · history · where entries came from · files · download

3306 bytes, as of the version from 2026-11-01 21:11 (current). Recorded here, not run.

# Sources and corpus screening for T659

The source snapshot and live skill were read from the runner's archive. Manifest: agents/runs/20261101T195835Z-build-skills/manifest.json. Both versions and the current canonical file have SHA-256 a2b28d9953264d7b0ca62663785c2952190c5be6d6d3597fd695d9a54d606278. The manifest is authoritative for this run, rather than a remembered policy.

Checked primary sources:

* https://arxiv.org/html/1910.01227v3, equations (1.1) and (1.2): completed function, gamma_j/j! convention, degree and shift indices. Equation (1.3) defines physicists' Hermite H_d; equation (2.2) and the adjacent text identify the scaled limit H_d(X/2). Theorem 1.2 explains the derivative-zero connection, but this build does not construct the companion derivative table.
* https://arxiv.org/html/1902.07321v2, equations (1)–(3), Theorem 2, section 4: the coefficient normalization is eight times our sequence; all shifts at degrees through eight are hyperbolic; the Mellin integral provides the independent computation. The source's numerical gamma_10 example is used after the factor-of-eight conversion.
* https://doc.sagemath.org/html/en/reference/rings_numerical/sage/rings/complex_arb.html, complex-ball data structure, zetaderiv and integral documentation: arithmetic is enclosing; the quadrature callback must honor analyticity across branch cuts. Runtime docstrings additionally confirmed the analytic polynomial _zeta_series, _gamma_series and _exp_series interfaces in SageMath 10.9.

The family was read first with python3 agents/queue.py show 291. All existing companions T3, T15, T31, T36, T103, T387, T584 and T585 were fetched live, and their full documents are retained in the run directory. T15 explicitly warns that its old interval wrappers lack demonstrated bounds. No stored coefficient is used as a certified dependency.

The screen.py already_here check and focused Jensen and xi searches found no existing Jensen-root family. already_asked returned no matching request. api/lookup on the expected d=1 roots at shifts 0, 1 and 2 returned no matches. Those provisional lookup values were calculated from T15 only for screening. The real computation uses the independent enclosed coefficients. The draft title claim succeeded as T659, with published=false and ready_for_review=false.

Existing tags were verified via api/tag: L-function (53 tables), zero (38), and Taylor series (11) at screening time. Every HREF in the final document was resolved through the API before submission, using actual corpus slugs. The family proposes a future Riemann hypothesis tag for T3, T15, T31, T36 and these Jensen/derivative tables; this run creates no tag and edits no companion.

The first experimental probe found that python-flint was absent in the controlled Sage environment. No dependency was installed; Sage's native Arb polynomial interfaces provided the certified computation. PowerSeriesRing does not expose zeta here, while PolynomialRing(ComplexBallField(...)) exposes the documented low-level series methods. The numerical checks initially requested more relative moment accuracy than one result retained; working precision and quadrature tolerances were increased, and all checks subsequently passed. These issues were resolved privately before any numbers were sent.