IMRPhenomT-family dominant merger qualification receipt — 2026-09-29

Scope and inputs

The versioned paper-derived scalar source reconstructs the updated THM Eqs. (28)–(32) dominant merger frequency, slope, phase difference, and signed amplitude from supplied cutoff, interior, peak, and QNM inputs. It imports neither native backend nor fitted target functions. This is a component equation check across T/THM and the shared aligned carrier used by TP/TPHM, not a whole-waveform qualification.

Source/test SHA-256, identical in local worktree and len snapshot:

File	SHA-256
src/diffgw_reference/phenomt_merger.py	36d2518cdf45865faf87935b093ef9c891f49df8da620abe982a8e746a575117
tests/test_phenomt_merger_math_reference.py	a16faa1913849882c7e9380edff823bf7f8828a5e9a78d7657321f41fa62e89c

Execution

Host len, directory /home/xangma/tmp/diffgw-phenomt-qualification-20260929, Python 3.12.3, Torch 2.14.0+cpu, JAX 0.11.2 in FP64, mpmath 1.3.0, SymPy 1.14.0, LALSimulation 6.2.1. The imported _lalsimulation extension's SHA-256 is e816879d23e652e8b3928496d6fd261de0832b5c3781c50e33a458042495b42e. The frozen tests compare equal masses, unequal aligned spins, and a 10:1 mass-ratio case for each backend.

Check	Command suffix after PYTHONPATH=src:. python -m pytest -q	Observed result
New merger plus existing inspiral/ringdown scalar gates	tests/test_phenomt_merger_math_reference.py tests/test_phenomt_ringdown_frequency_math_reference.py tests/test_phenomt_ringdown_amplitude_math_reference.py tests/test_phenomt_inspiral_collocation_math_reference.py	29 passed in 67.66 s; phenomt_scalar_combined.log
Aligned THM and NNLO TP/TPHM against pinned binary	tests/test_phenomt_family_lal.py::test_lal_binary_reference	9 passed in 11.87 s; phenomt_aligned_nnlo_binary.log
q=10 MSA/PV300 strict binary gate	-rx tests/test_phenomt_family_lal.py::test_high_mass_ratio_precession_unresolved_gate	4 strict expected failures in 31.64 s; phenomt_q10_gate.log

The merger tests require coefficient, frequency, slope, and amplitude agreement at rtol=2e-10, atol=2e-11, and phase differences at rtol=2e-10, atol=2e-9. They compare off-collocation times as well as boundary values. Scalar self-checks use a SymPy expression, 75-digit collocation, and symmetric derivative probes. Both Torch and JAX pass all three cases.

Interpretation and remaining gates

The passing result checks the published interpolation algebra and the native join choices when fitted values, PN cutoff data, and QNM rates are supplied. The THM fit surfaces and raw NR/EOB targets are not recreated by the paper; the pinned MIT phenomxpy asset remains their identified implementation source. Mode-specific HM merger, signed zero crossings, remnant/QNM lineage, precessing ringdown input splice, and complete waveform accuracy remain separate gates.

All four q=10 MSA/numerical TP/TPHM gates still fail their retained binary thresholds. The phase and rotation blocker localizes the discrepancy but does not establish a paper-backed production correction. This receipt changes no precession equation, option, or threshold. The prior measured q=10 TP/TPHM relative L2 values are 0.0202333/0.0225165 for MSA (limit 0.01) and 0.308418/0.307103 for numerical (limit 0.03). The published-equation MSA scalar matches native before the binary's pre-MECO rotation diverges; a binary-compatibility spin-contraction choice reduces that residual but lacks paper provenance. The independently integrated numerical trajectory matches native within 4.78e-9 rad while the public binary's reconstructed rotation differs by up to 0.39276 rad. These observations locate the remaining work in option/convention and pre-merger orientation qualification, not in the newly checked merger interpolation.