research.md

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

6280 bytes, as of the version from 2026-11-16 11:55 (current). Recorded here, not run.

# Even-row Bessel period determinants

Checked directly: FSY arXiv:2006.02702v2, Notation 8.2/(8.3) defines I0^(2i) K0^(n-2i) with t^(2j-1), square size floor(n/4); the following paragraph deletes the last row and column at 4|n. Theorem 8.4/Table 1 relates these raw minors, with explicit pi factors, to rational-equivalence critical periods. The table stores raw representatives, not canonical Deligne classes and not a universal proved L-value formula. Domain n>=5; zero-dimensional n=4 excluded. No convention ambiguity remains for this assigned even-row quantity.

Checked directly: Zhou arXiv:2012.03523v3 (1.21), Proposition 4.5 gives the odd-factor reflection multiplier. It does not cover even factor counts. Equations (4.37)–(4.38) were inspected directly: they give M(2,3;1)=pi sqrt(15) C/2 and C=Gamma(1/15)Gamma(2/15)Gamma(4/15)Gamma(8/15)/(240 sqrt(5) pi^2), hence E5=sqrt(3) times that Gamma product divided by 480 pi. Sibling T724 is a relation target and source-location lead, not independent evidence.

DLMF 10.32.1 and 10.32.9 were inspected for endpoint bounds I0(t)<=e^t and K0(t)<=sqrt(pi/(2t))e^-t. The latter also bounds K0(t)<=E1(t/2)<=2-log(t) on (0,1]. DLMF 13.6.10 fixes K0(t)=sqrt(pi)e^-t U(1/2,1,2t). The shared generator's general moment routine is reused with its explicit bounds; new even matrices and independent checks are computed in this run.

Live discovery: authenticated catalogue fetched once for this planning pass (catalogue.json); screen.already_here and already_asked saved in screen.json. No matching even-row title/family found, including drafts. Definitions and current metadata inspected for T521/T647/T648/T720/T721/T724. T521 has scalar overlaps, T647 full consecutive rows, T648 vacuum rows; T724 is the separate odd-row family, now offered by its controller. Only the assigned new draft may be edited. Feynman integrals remains absent from /tags; existing integral/period/special values/number theory tags are suitable. /tags/ returned 404; /tags works.

Provenance: proposal #323 explicitly contributes to wanted #263; #263 inspected and remains a broad open request. Do not close it or claim unrelated family origins. The exact initial_commit_messages mapping in proposal.json is used on first creation and preserved during fill. Request provenance stays out of mathematical prose.

Sources are archived with URLs and SHA-256 in sources-manifest.json; local/live matching instruction archive is skill-manifest.json. Larger numerical artifacts and their hashes will be recorded in evidence-manifest.json and handoff.md. Selection rationale is argument-selection.md. No scientific verdict is supplied by this build.

Endpoint derivations checked: cosh u>=exp(u)/2 gives K0(t)<=E1(t/2); for t<=1, E1(t/2)<=log(2/t)+e^-1<2-log(t). On t<=exp(-A), I0(t)^a<=exp(a exp(-A)); putting u=-log t bounds the origin by the integral of exp(-(r+1)u)(u+2)^b, evaluated by b integrations by parts as in generate.py. At infinity I0^a K0^b t^r <= (pi/2)^(b/2)t^(r-b/2)exp(-(b-a)t). For p>=0 use log(t/T)<=(t-T)/T, and for p<0 use t^p<=T^p, giving the stored explicit exponential tail bound. These are bounds for every selected moment, not estimates based on decimal agreement.

Sign derivation: Andreief applied with positive weight t I0(t)^2 K0(t)^(n-2), row functions h(t)^(i-1), h=(I0/K0)^2, and column functions (t^2)^(j-1), gives a positive ordered multiple integral. I0 increases by its positive power series, while K0 decreases by its positive integral representation. Both Vandermondes have the same positive ordering, so each raw determinant is positive. This derived sign check does not authorize taking absolute values.

Numerical method documents inspected: Sage complex_arb integral callback contract (analytic flag, NaN outside a holomorphic domain) and FLINT acb_calc rigorous finite-enclosure contract. The original algorithm is Johansson, arXiv:1802.07942. The current Sage documentation is version 10.10; actual runtime is measured as 10.9. Direct downloading that Sage documentation returned HTTP 403; it was inspected through the browser tool, while FLINT documentation is archived as Arb.html. These documents support the method, not correctness of an arbitrary callback.

The nonzero controls can also be derived directly. DLMF 10.32.9 and Tonelli give integral_0^infinity t K0(t) dt = integral_0^infinity sech(u)^2 du = 1. For the squared moment, the standard derivatives K0′=-K1, K1′=-K0-K1/t imply d[t^2(K0^2-K1^2)/2]/dt=t K0^2. At zero this antiderivative tends to -1/2 (t K1(t)->1; t K0(t)->0), and at infinity to zero, giving M(0,2;1)=1/2. These are calibration quantities, not stored new entries.

BBBG arXiv:0801.0891v2, equation (7), inspected directly: the formulas for c_(1,k) and c_(2,k) give the two nonzero moment controls at k=1. Source archived as BBBG.html, hash in sources-manifest.json.

Completed production evidence: all 56 defining moments plus two controls carry finite rigorous Arb balls. All 12 raw even determinants have positive finite enclosures, with minimum relative accuracy 388 bits; the largest measured absolute-permutation-sum/determinant cancellation bound is below 0.595 decimal digit. Exact Fraction comparisons show all complete balls fit their 100-significant-digit serialization intervals. Existing dry_run.py/check.py checks pass, with native RealBall results, no suspicious exactness pattern, and a 2.2 KB entries block. Independent mpmath computation is a separate step; no independent scientific verdict is issued here.

Completed independent checks: mpmath 1.4.1 at 130 decimal working digits independently integrated all 12 even matrices and expanded signed determinants by permutations. All 12 values lie inside the proposed stored last-place intervals; maximum midpoint discrepancy is 0.483602 stored ulp. Gamma E5 and all six odd-factor reflection checks at n=5,7,9,11,13,15 pass using freshly integrated odd matrices. The 50-digit calibration controls and deliberately wrong sign, missing-pi, wrong-dimension, divergent-boundary, NaN/infinite and vacuous interval controls pass. This is builder-owned independent implementation evidence, not the fresh independent scientific verdict required from the controller’s critic. Executed log and result JSON are retained.