IMRPhenomHM SXS:BBH:2158: prospective aligned-spin holdout
Freeze this protocol and its runner before retrieving or decoding 2158 strain. The source selection used only the official SXS v3.0.0 catalog metadata and file manifest. Record a failed or infeasible comparison without changing the case, conditioning, optimizer, or gates. This is one aligned-spin holdout; it does not replace or change any earlier SXS receipt.
Research target and case selection
London et al. (2018) defines the aligned-spin IMRPhenomHM model and compares it with full numerical-relativity signals. Its reported high matches are population results, not a guarantee for this particular simulation or each viewing angle. The model's omitted spheroidal-mode mixing and its finite six-mode content remain possible physical sources of mismatch. The SXS Catalog 3 paper documents the v3 strain product. The physical comparison, numerical conventions, and operational thresholds below follow the previously frozen 3144 protocol, except for the explicit case, resolutions, and spin inputs here.
Use the official
sxscatalogdata v3.0.0 simulations.json,
SHA-256 132d38520b3562c8f254b3c9aaf8efd4678428f94dbbabfef4436bc50c1c87c6.
Among nondeprecated SXS:BBH: entries, require run date at least
2018-01-18, 3 <= reference_mass_ratio <= 8, both reference spin z
components in [0.2,0.7], both transverse spin magnitudes below 0.01,
reference eccentricity below 0.002, at least 20 post-reference orbits, and
at least two Lev*:Strain_N2.h5 files. Sort by
|q-4| + |chi1z-0.5| + |chi2z-0.5|, then numeric SXS ID. The five eligible
IDs in that order are 2158, 2163, 2162, 2155, 2502. This metadata-only
rule selects SXS:BBH:2158v3.0. Selection does not establish complete
independence from model development.
The official case DOI resolves
to CaltechDATA record 3gb3n-e5t37.
The catalog gives reference q=3.0001323450539252, spin vectors
(-4.0913123854969844e-8, 4.2441616226361754e-8, 0.49997718539303)
and (-1.4945164577544106e-8, -1.639491623205979e-8,
0.5000516952582452), eccentricity 8.87e-5, reference time 1384 M,
22.603589434918916 post-reference orbits, initial orbital frequency
0.015690006867/M, and four available N2 strain resolutions. Use Lev4
and Lev3, the two highest. At 100 M_sun, the initial approximate (2,2)
frequency is below the 30 Hz metric floor.
Retrieve only these six files from the catalog's
/api/records/3gb3n-e5t37/files/<URL-encoded-key>/content links after
the protocol and runner are committed. Verify byte size and published MD5
before decoding; record local SHA-256. Keep raw and derived strain outside Git.
| Catalog key | Bytes | Published MD5 |
|---|---|---|
Lev4:Strain_N2.h5 |
2990318 | 2c2ccd0a8505ae5d223283dbe39734df |
Lev4:Strain_N2.json |
2232 | 3b6e8c5656052f726771a95c0a9b4408 |
Lev4:metadata.json |
3347 | 6f75989b5a2bb444a4ea612636994da2 |
Lev3:Strain_N2.h5 |
3280093 | 42bb7d97a1531a76d13b3cef0aa82d5c |
Lev3:Strain_N2.json |
2231 | 6f1c761e840c02ba4723e37048e35174 |
Lev3:metadata.json |
3345 | 955f6b41c590f543656f88f34bdc291d |
Frozen signals, comparison, and gates
Apply every conditioning and scoring choice in the 3144 protocol with Lev4
as the reference and Lev3 as the lower resolution: full available inertial
NR 2 <= ell <= 5 including m=0; the public sxs loader; 100 M_sun,
500 Mpc, f_ref=30 Hz; common post-reference interval, cubic complex-mode
interpolation to 8192 Hz, half-cosine 500 M rise and 100 M fall; two
fixed (inclination,source psi) views (0.7,0) and (1.2,0.4); direct
30–1024 Hz Fourier sum; PyCBC aLIGOZeroDetHighPower PSD; one common
Lev3-to-Lev4 orbital-phase/time alignment using the 2048-point phase grid
and eight-peak refinement; independent Torch/JAX template time, orbital
phase, and polarization optimization with four seeds (17,23,41,67).
Use exact catalog reference q for masses and model aligned spins
chi1=0.49997718539303, chi2=0.5000516952582452. The tiny transverse
catalog spins are not fitted. Native phic=tc=phi0=0; do not fit intrinsic
parameters, inclination, distance, any mode-specific factor, or overall
amplitude.
Both Lev3/Lev4 views must have match >=0.995 and weighted norm ratio in
[0.95,1.05] under the same global alignment. Each Torch and JAX
model/Lev4 view must have optimized match >=0.980, weighted norm ratio
in [0.85,1.15], and four-seed match spread <=1e-5. Report combined
resolution metrics, all optimizers, omitted-mode diagnostics, source and
package hashes, and direct/FFT consistency checks. A resolution failure
prevents physical qualification regardless of model scores. A model failure
may reflect a physical model limit and does not by itself prove an
implementation defect. One pass establishes only this case, band, mass,
PSD, views, and finite-resolution comparison; retain earlier red gates.