back to table · edit · history · where entries came from · files · download
3230 bytes, as of the version from 2026-11-18 10:41 (current). Recorded here, not run.
# T750 numerical evidence
The final generator uses mpmath tanh-sinh on the analytically integrated
Feynman-parameter representation, independently evaluated at 150 and 200
working decimal digits. The table stores 100 significant digits and is labelled
**heuristic (agreement-checked)**. Neither the union intervals nor the guard
widths are claimed to certify the quadrature error. Detailed analytic formulas
and attempted ball strategies are retained in quadrature-diagnostics.md;
those diagnostics do not establish proven rigour for the final values.
All integer inputs are exact. The reusable function also constructs rational
inputs by exact QQ numerator and denominator before mpmath conversion, and
recomputes both precisions. The removable D=0 limit is evaluated by its
convergent atanh series; D<0 uses atan2 to preserve the correct positive branch.
The logarithm avoids subtracting nearly equal B and sqrt(D).
The direct-r Bessel checker uses a genuinely different representation and
Arb ball quadrature. It includes the factor four, right-half-plane analytic
guards, and explicit bounds on both omitted tails. Scaled Kummer identities
DLMF 13.6.9–13.6.10 control exponential dependencies. The small-r bound uses
K0(mr)<=K0(r)<=log(2/r)+1 for the selected integer masses m>=1;
the large-r bound uses I0(sr)<=exp(sr) and
K0(mr)<=sqrt(pi/(2mr))*exp(-mr). It adds both omitted contributions to
the quadrature radius and records actual accuracy bits. This independent
check is at lower precision than the stored digits; its actual completed
coverage and minimum accuracy are recorded in check-results.json.
parameter_check.py is a separately written mpmath implementation with direct
atan2/discriminant handling and independent box/perimeter enumeration. It
checks every entry at 60 working places, and every selected t for mass systems
(1,1,1),(1,1,2),(1,2,2),(1,2,4),(2,2,3),(1,1,10),(2,3,7),(3,4,5)
at 135 working places. It shares the analytic y integration and tanh-sinh method, so it is a
secondary branch/implementation check, not an independent mathematical method.
The Bessel checker supplies the independent representation.
Known nonzero controls use trigamma for 3 L(2,chi_-3), pi^2/4 and 2 log 2,
at 170 working decimal places. A directly derived logarithmic formula
checks all six degenerate vacuum cases, making eight known-answer controls. Wrong factor four, wrong sign of t, masses
versus squared masses, false scale invariance, NaN and vacuous wide intervals
are deliberately tested. The correct degree -2 scaling is checked separately.
The final comparison reads the authoritative stored decimal intervals,
including their complete last-place uncertainty, not only midpoints.
The archived skill manifest is
../20261118T085652Z-build-skills/manifest.json, SHA-256 recorded in
skill-validation.json; both complete bundle hashes match:
d6e85d49873fac5bc748d068b0288cf242b0f9aebf3a9a0fff70295019227420.
Full sources and hashes: source-hashes.json. Computation software versions,
JSONL checkpoints, attached generator hashes, deterministic checks and actual
stored-value comparisons are retained beside this report. This builder does
not issue the independent scientific or argument-selection acceptance.