IMRPhenomT inspiral fit-input transcription, 2026-09-27

T-PN-FIT-INPUT-v1 checks the aligned (2,2) TaylorT3 time shift and five inspiral frequency collocation targets against the pinned MIT phenomxpy author source. It is a finite-grid transcription and assembly check, not independent reconstruction of the numerical calibration or complete waveform qualification. The component specification freezes the grid and thresholds.

Source and results

The author asset is phenomxpy/phenomt/fits.py at revision 4121fafa0484f3db2a7352f4eea697faaaf29f65, SHA-256 2dfcca659cd50d927b9e164867913a586e9b1372f58ffde424f8bfbd74b56e21. The source-derived fixture has SHA-256 729963d920439c7741c631fbc9982704ec1a785007d7b15911a146cd1b7936c3. On len, the source verifier reproduced all six outputs in five frozen cases within 2e-13 relative or absolute error. The backend gate passed 37 tests, no skips: 15 fit-input cases (Python, Torch, JAX), six collocation cases (Torch and JAX), and 16 TaylorT3 checks. The second gate ran without JAX_ENABLE_X64 set; the maintained tests explicitly enabled JAX float64.

The source-backed fit results agreed with the shared native fit functions at rtol=2e-11, atol=5e-10 for t0 and rtol=2e-11, atol=2e-12 for the five frequencies. Scalar pseudo-PN coefficients assembled from source-derived fit targets and independent TaylorT3 coefficients agreed with both native carriers at rtol=2e-8, atol=1e-8, including on-node and off-node frequency checks. No tolerances were changed after the runs.

Reproduction

Host: len, CPU. Scratch cwd: /home/xangma/tmp/diffgw-phenomt-fit-inputs-20260927. The pinned author asset must be supplied separately at author-fits.py with the exact hash above. Python is /home/xangma/.cache/diffgw-taylorf2-hm-venv/bin/python (3.12.3); package versions: NumPy 2.5.3, PyTorch 2.14.0+cpu, JAX 0.11.2, mpmath 1.3.0, pytest 9.1.1.

PYTHONPATH=src /home/xangma/.cache/diffgw-taylorf2-hm-venv/bin/python scripts/verify_phenomt_author_inspiral_fits.py --author-fits author-fits.py
JAX_PLATFORMS=cpu PYTHONPATH=src:tests /home/xangma/.cache/diffgw-taylorf2-hm-venv/bin/python -m pytest -q tests/test_phenomt_inspiral_fit_inputs_math_reference.py tests/test_phenomt_inspiral_collocation_math_reference.py tests/test_phenomt_taylor_t3_math_reference.py --reference-device=cpu -ra

Outputs: verified 5 cases and six fit outputs per case; 37 passed in 10.13s. The second command's log is gate-no-env-x64.log in the scratch cwd. The published THM paper defines the collocation construction but does not print these fit coefficients or the full NR/EOB target tables; hence physical calibration provenance remains conditional in the research ledger.