back to table · edit · history · where entries came from · files · download
4063 bytes, as of the version from 2026-11-20 00:19 (current). Recorded here, not run.
# Quartic Mahler pencil: checked build leads
T768 is the new private draft for B(k)=m(x^4+y^4+z^4+1+kxyz), actual coefficient k>=0, natural logarithms, Haar probability measure, ordinary integrable torus zeros. Rogers f4(u)=4B(u^(1/4)); hence u=k^4. No interior continuation is supplied. The only nonintegral selected coefficients are positive fourth roots 648^(1/4) in (5,6), 2304^(1/4) in (6,7). Evaluation uses exact fourth powers, not copied root approximations.
Sources actually inspected: Rogers 0704.2438v4 Eq. (2.6), Prop. 2.2 Eq. (2.9), Eqs. (2.19),(2.21), Cor. 2.6 Eq. (2.26); Samart 1205.4803v2 Prop. 1.3(iii) and Theorem 1.4 Eqs. (1.9)–(1.12), including the normalized eta products and q-expansions; DLMF 16.12.2 and 15.8.4; Smyth 1981 p.50 (explicit exponential of 7*zeta(3)/(2*pi^2), DOI 10.1017/S0004972700006894). Local raw sources, source-passages.json and source-manifest.json retain locations/hashes. Rogers/Samart direct urllib retrieval was blocked, so their exact versioned shared source downloads were reused after fresh web inspection. No sibling values or conclusions serve as independent evidence.
Fresh discovery: one lightweight authenticated catalogue pass, 764 records, plus screen.already_here, screen.already_asked, text lookups and numeric lookups. Relevant current candidate definitions were inspected and refreshed immediately before creation. No existing table/draft has this family. B(0) matches T283 and T685; B(4) matches T664's different polynomial context. These coincidences do not duplicate the quartic pencil. Shared issue #350 is this proposal, with #263 as its only direct originating wanted request; #293 is background. Exact initial edit citations were preserved on every same-run document amendment.
Verified registry spellings retained without material departure: Mahler measure, integral, number theory, special functions, special values, hypergeometric functions, K3 surfaces. The K3 tag describes the geometric context, not CM of every selected fibre. Related live IDs/definitions: T761 (template 2.4 derivative period), T731 (marked mirror-quartic ratio), T663 (Rogers/Bertin transform), T767 (distinct f3 measure), T748 (f2=2M(sqrt(u))), T283 (linear Fermat degeneration), T14 (zeta), T145 (primitive character L-values). Their links appear in rendered Similar tables. No companion was edited.
Checked computations: explicit Laurent convolution through power 24; rigorous exact-rational exterior tails; certified Gauss-connection polynomial integration at k=4; 17 finite Arb enclosures, all radii below 1e-120, serialized to 100 significant digits. Independent mpmath Gauss integration checks all 15 exterior rows at 135/175 digits. Independent PARI eta/newform/character L-functions check k=4 and all four CM cases at 140/180 digits; newforms distinguished by q-coefficients. Independent mpmath zeta checks k=0. Nonzero known-answer controls and deliberate wrong-value/normalization/form controls pass. Separate Arb 5F4 checks cover all 15 exterior rows. Stored intervals were read back and all 17 reproduced by verify(sample=None). See mathematical-checks.md for proof details and numerical-evidence.json for actual output paths/hashes.
Argument selection is justified in argument-selection.md; bounds are exploratory and noncanonical, with no famous-only pruning or quantile inputs. Skill manifest: ../20261120T000421Z-build-skills/manifest.json; core b79b7fbc9fa3ce63f2d642cccdac12b0d214757ab01db66c03a2d84264a58e5f; matching bundle b5c3c7985b7c9e178b4899d083a9e3bc4284989dff76f05e5b0e52835d47eed2. A copy is attached as skill-manifest.json.
No unresolved numerical discrepancy. Quadrature and PARI comparisons are numerical cross-checks; rigorous uncertainty comes from generator tail bounds and Arb, not agreement. Scientific critique, the independent argument-selection sidecar, bounded repair if warranted, final family review and offering remain controller stages. This builder issues no independent verdict. Research and all failed diagnostic runs are preserved under this work directory.