argument-selection.md

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

4598 bytes, as of the version from 2026-11-19 23:07 (current). Recorded here, not run.

# Argument selection: primitive odd modular periods

The single quantity is the signed real scalar Omega_f^- fixed by
r_f^-(X)=Omega_f^-*P_f^-(X). Here r_f is the upward integral from 0 to
imaginary infinity of f(tau)(tau-X)^(k-2), and the odd part includes the
factor 1/2. P_f^- is the primitive integral proportional polynomial with
positive leading coefficient, without rescaling X or removing any factor.
The domain consists of normalized newforms of exact level N, trivial
character, even weight k>=4, rational Fourier coefficients and nonzero odd
part. This restriction permits a unique primitive integral shape; it is not
a claim that coefficient fields of higher degree have no useful periods.

Checked source locations: PARI Modular_forms.html, mfperiodpol; Paşol–Popa,
arXiv:1202.5802v5, definition of rho in Section 1 and Proposition 5.11(a,c).
The formula and rationality fix the natural family. Live companion T696
metadata and its full document were refreshed in this run. Its existing
shape selection is context, not independent evidence of correctness.

The ordered parameters are N,k,i. No continuous free coupling
is present. All embeddings of every newform participate in the lexicographic
ordering of exact (a_2,a_3,...) through a Sturm bound. The index i is retained
when restricting to rational fields. It is not a PARI orbit position or a
renumbered rational-form index.

- N: a discrete exact-level index. The finite reference selection has the
  full-level slice N=1, and every level 2<=N<=10 at weight 12. The latter adds
  prime, composite and nonsquarefree levels, retaining empty and nonrational
  spaces in the completeness checks. The bound 10 is explicitly noncanonical.
- k: a discrete even-weight index. At N=1, every nonempty rational newform
  space in 4<=k<=26 is selected, not only Delta. The second slice holds k=12
  fixed while varying the exact level. These are complementary slices, not
  a rectangular product. Weight 12 is the first full-level cusp weight and
  supplies richer period polynomials while making a level comparison.
  The finite bound 26 is explicitly noncanonical. Weight 24 is examined and
  its degree-two Hecke field is excluded by the domain, not by decimals.
- i: all rational-field newforms in these spaces are retained with their
  indices among all real embeddings. This includes the multiple-form levels
  and any index gaps. Exact decomposition, dimension and Sturm comparisons
  check that no rational form in the selection has been omitted.

The expected roster has six rational forms at level one (weights 12,16,18,
20,22,26), and twelve rational weight-12 forms at levels 2 through 10: 18
forms and 18 entries. This is a prediction to be checked by the executable
checks, not a reason to discard a contrary exact decomposition. The selected
spaces include level 2 and level 7 even if their rational roster is empty.

Breadth is reviewed before depth: both complementary slices restore the
missing absolute scale for every selected primitive odd shape in T696,
rather than increasing precision for a few famous examples or constructing
an arbitrary grid. Extending farther in level or weight remains possible.
General Hecke fields need further integral-lattice and embedding conventions
and are outside this quantity's stated rational domain. The even scalar is a separate quantity in live sibling T765, whose values
are not independent evidence. No extra tables are built here. The default enumeration, Programs and
complete-note must describe these same slices. The actual measured size and
roster are in numerical-evidence.json; complete is no for the natural family.

The coordinated skill manifest is manifest.json, copied unchanged from
agents/runs/20261119T222759Z-build-skills/manifest.json. Every archived resource
hash was verified and the local/live bundles match, bundle SHA-256
b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2.
The independent critic must assess these reasons and selection interactions;
this build document is not the independent argument-review verdict.

Actual complete roster, checked by exact decomposition and both constructions:
N=1 weights 12,16,18,20,22,26; weight 12 pairs (N,i)=(3,1),(4,1),(5,2),
(6,1),(6,2),(6,3),(8,2),(9,1),(9,3),(10,1),(10,2),(10,3). There are 18
entries, approximately 3 KB including parameter labels; longest value is
105 characters. All selected odd parts are nonzero, primitive and positive
leading. The full 33-embedding roster, including excluded degree>1 forms,
is retained in roster.json. There are no continuous parameters to justify.