research.md

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

5855 bytes, as of the version from 2026-11-20 06:18 (current). Recorded here, not run.

# Checked family research: rank-zero central L-values

T771 is this run's private populated draft for the first proposal in #353,
contributing partially to wanted #266. No offer, publication, human review,
or whole-request completion is claimed. The period sibling has no assigned
identifier in this build. Only the selected central-value table was built.

The matching archived canonical skill was read from the local archived
bundle, including all six relevant references. Its complete local/live
resource hashes were verified. See skill-manifest.json and
../20261120T044104Z-build-skills/manifest.json; identical bundle SHA-256:
b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2.

Sources actually inspected:
- Cremona ecdata page, Tables One and Two, for reduced global minimal models,
  source classes, torsion structure lists and projective generators.
  https://johncremona.github.io/ecdata/
- Sage elliptic L-series documentation, at1 and L_ratio, for the split
  central Mellin series and the semistable 11a1 ratio control. Inspected
  through the web tool; the direct HTML download failed and is not a cache.
  https://doc.sagemath.org/html/en/reference/arithmetic_curves/sage/schemes/elliptic_curves/lseries_ell.html
- PARI elliptic-curve documentation, ellap, ellan, ellrank, ellrootno,
  elllocalred, ellminimalmodel and ellperiods. ellrank supplies algebraic
  lower/upper bounds; analytic Sha outputs were never used.
  https://pari.math.u-bordeaux.fr/dochtml/html-stable/Elliptic_curves.html
- The current LMFDB JSON record for 11.a2 confirms that its exact model is
  Cremona 11a1, not Cremona 11a2. The frozen alllabels source supplies the
  entire explicit label crosswalk. Download details: source-downloads.json.

The full frozen allgens and alllabels bytes at commit
25cec5ecfec8b9f016eb1631ac633194c2bed39f are retained here, with URLs,
paths and hashes in source-manifest.json. Their N<=100 extracts are attached.
roster.py parses all source records rather than starting from exceptional
Sha examples: 306 curves through 100, 284 of source rank zero, 76 classes.
roster.json retains both label systems, exact models and source line numbers.

Checked normalization: primitive uncompleted Hasse–Weil L(E,s), center 1,
Lambda=N^(s/2)*(2*pi)^(-s)*Gamma(s)*L; root number +1 on every selected
class. Ordered N,c4,c6 identify the reduced global minimal source model.
No factorial adjustment occurs at rank zero. Periods remain a distinct
quantity. The full BSD period is not used to generate central values.

The independent-input checks in build_check.py found algebraic rank bounds
[0,0] from PARI 2-descent for all 284 curves. Their exact conductors, minimal
models, c4/c6/Delta, torsion structures, source torsion-point orders and bad
local factors were checked. Source class membership is sourced; prime traces
through 97 agree throughout each class. No standalone descent-covering
certificate verifier or independently computed Sha order is claimed.

The production method counts projective points exactly, builds good and bad
Euler factors and uses exact multiplicative recurrences. Hasse yields
|a_(p^k)|<=(k+1)*p^(k/2); multiplicativity yields |a_n|<=d(n)*sqrt(n).
Divisor pairing gives d(n)<=2*sqrt(n), hence |a_n/n|<=2 and the omitted
central Mellin tail is <=4*q^(M+1)/(1-q). Arb encloses q, the partial sum,
and this tail at 640 bits. The selected term counts range from 208 to 624;
minimum effective relative accuracy is 559 bits. Each 100-significant-digit
stored decimal interval contains the actual non-vacuous production enclosure.

All 76 outputs passed PARI lfun comparisons at 832 and 1024 bits, all
35,584 used Euler coefficients matched ellan, and a separate double point
count checked primes through 97. Both wrong signs and omitted factors of
two were rejected. The published positive 11a1 control and semistable ratio
L(E,1)/Omega=1/5 passed. PARI comparisons are heuristic implementation
checks, not the source of rigorous error bounds. Conductor/root/descent
routines share PARI with that implementation; the producer's point-count
and Arb sum code are separate. Evidence: check-results.json, calculation-final.jsonl,
build_check.py, generate.py and program-check.log. All failed probes and
initial calculation outputs remain preserved, with their identified causes.

Fresh catalogue discovery reached all 767 tables and both accessible offered
drafts. already_here, already_asked, numeric lookup probes and all four exact
tag lookups were repeated. No matching rank-zero central-value family was
found. The 11a1 and 37b1 numeric probes had no hit; the primitive-period
control found T72 as expected. This is evidence of discovery, not a proof
from failed number searches. Existing T69–T71 store derivatives at ranks
1–3; T72–T74 store primitive real periods. T211 has the conductor<=100
Faltings-height population. T413 is p-adic; T683 currently has weights
3,4,6; T745 is at s=2 (and its reflected derivative). T765/T766 currently
have level-one and weight-12 modular-period slices. Their actual metadata
were inspected; none duplicates this quantity and selection. See catalogue.json,
candidate-metadata.json and discovery.json. The four registered tags are
retained without departures or alias proposals: elliptic curves, number
theory, L-function, special values.

Live readback verified all 76 actual stored intervals against the production
enclosures and both saved independent approximations; generator.verify also
matched 76/76. See stored-check-results.json and fill-2.log. Final rendered
presentation, revision citations, attachment hashes and audit are recorded in
the handoff after final attachment upload. No independent verdict is written
by this builder. A fresh scientific critic and argument-selection review
remain required before the controller may offer the draft.