argument-selection.md

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

3400 bytes, as of the version from 2026-11-10 11:25 (current). Recorded here, not run.

# Argument selection for T705

The named object is O'Sullivan's polynomial
\(P^{d,n}(X)=\sum_{j=0}^d\binom dj\gamma_{n+j}H_{d-j}(X)\),
with the completion \(\xi(s)=s(s-1)\pi^{-s/2}\Gamma(s/2)\zeta(s)/2\),
\(\gamma_m=m!\xi^{(2m)}(1/2)/(2m)!\), and physicists' Hermite polynomials.
The normalization is fixed; there is no free Gaussian deformation or coupling.

Checked primary source: C. O'Sullivan, arXiv:2007.13582v2, equations
(1.2), (1.4), section 2 and Corollary 3.8. The family is a published variant
of the Jensen criterion. Its roots differ from the ordinary Jensen roots
in T659 and from the roots of the individual Hermite basis polynomials.
The coefficient convention also agrees with arXiv:1910.01227v3, equations
(1.1)–(1.3). The shift differentiates the entire central-series function
Theta, not xi in its original argument.

Every free parameter has the role `discrete_index`:

* **d**, degree, has domain d >= 1. We take the consecutive degrees 1 through
  10. This includes linear and quadratic formulas, both parities, and several
  higher-degree root configurations before extending degree depth. It is a
  bounded small-degree reference for this particular finite-polynomial
  construction, not a family chosen merely because a library evaluates it.
* **n**, coefficient shift, has domain n >= 0. We take 0 through 10. These
  are the central case and the first ten derivatives of Theta within the
  fixed construction, so no continuous input grid is hidden in this index.
  The consecutive prefix permits comparison of early shift dependence.
  This shorter prefix balances degree breadth against the increasing count
  of roots per polynomial, independently of T704's deeper shift range.
* **k**, real root index, starts at 1 and orders real roots increasingly,
  counting multiplicity. For every selected (d,n) we retain all k=1,...,d.
  Certification must establish d disjoint sign-changing intervals. Thus no
  asymmetric subset of roots or particular parity is favored. The general
  definition only indexes real roots that exist; the selected block is
  certified to have d simple real roots without assuming RH.

The block has 110 polynomials and 605 roots. Neither count is a target.
Bounds on d and n are noncanonical computational reference cutoffs,
independent of decimal notation. The rectangle here enumerates two genuine
indices of one published polynomial, not unrelated simple real arguments.
There are no exploratory continuous inputs and no mixed-role groups.
The leading cases remain first: d, then n, then increasing k.

Breadth review: both parities, the central case, early shifts, and every root
are covered. Adding a Gaussian scale, a Laguerre/Bessel basis, monic
polynomial coefficients, or discriminants would change the selected object
or add another quantity and is outside this table. Deep shifts and high
degrees are left for extensions with a separate cost and relevance review.
The finite block and its rationale are repeated in the completeness note;
the parameter domains are not restricted to this run's cutoffs.

The archived governing instructions are recorded in
`agents/runs/20261110T105402Z-build-skills/manifest.json` and the attached
`skill-evidence.json`. Both archived skills have SHA-256
`886390324dd1971e381a43cf3b46b822ab0bc67d62f31d8481e29ec79556746b`.
This build rationale is not the required independent critique or acceptance.