NSBH remnant guard source boundary
Scope and status
nsbh.remnant.guards/v0 is blocked, not an executable paper-alignment
gate. This audit separates the published remnant fit and physical bounds from
the extra clipping/fallback rules in the native Torch/JAX providers. The
printed-precision interval gate remains
valid on its five guard-inactive cases; this page does not change its domain,
tolerance, or status. No whole-waveform claim follows from this audit.
Inputs, conventions, and options
Write \(q=m_{\rm BH}/m_{\rm NS}\), \(\nu=q/(1+q)^2\), \(a=a_{\rm BH}\), and \(\Lambda=\Lambda_{\rm NS}\). Set the NS spin to zero and \(m_{\rm NS}=1.4\,M_\odot\). The diagnostic inputs are \((q,a,\Lambda)=(3,0,100)\) and \((5,-0.5,100)\). Both lie within the reported individual parameter ranges \(\nu\in[0.109,0.222]\), \(a\in[-0.5,0.75]\), and \(\Lambda\in[100,2500]\); this does not assert that either tuple is an NR calibration datum.
Mathematics and components
Zappa et al. (2019), main-text Eqs. (1)–(2) and Supplementary Eqs. (4)–(6), Table 1, §I define \(X_\bullet=M_\bullet/M\) and \(a_\bullet\), then multiply the corresponding Jiménez-Forteza BBH quantity by a rational tidal factor \(G_F\). The paper describes \(X_\bullet>1\) and \(a_\bullet<-1\) as unphysical extrapolation outcomes and says it forces agreement with BBH, while also writing “\(X_\bullet=1\) and \(|a_\bullet|\leq1\).” Those statements do not give a unique pointwise prescription: setting the total mass ratio to one and setting the tidal factor to one differ whenever \(X_{\rm BBH}<1\). The text does not state a numerical switch at \(\nu=0.188\), a factor-above-one clip, or equality/tie conventions.
The independent exact-rational scalar reference propagates half-last-printed-digit coefficient intervals through the published equations. The following bounds are rounded outward for display; the underlying interval evaluation uses exact fractions:
| \((q,a,\Lambda)\) | \(G_M\) | \(G_a\) | Unclipped \(X_\bullet\) |
|---|---|---|---|
| \((3,0,100)\) | \((1.000052,1.002250)\) | \((1.013955,1.017116)\) | \((0.971295,0.973441)\) |
| \((5,-0.5,100)\) | \((1.005855,1.007315)\) | \((1.012851,1.015142)\) | \((0.992122,0.993630)\) |
Both tidal factors are unambiguously above one in these examples, yet the total mass ratio remains below one. The source also explicitly describes the small-\(\Lambda\) increase of nonspinning NSBH mass and spin relative to BBH in its main text. Thus \(G>1\) alone is not the paper's stated unphysical condition. The Jiménez-Forteza et al. (2017) unity spin cap is stated for its precessing augmented fit (Eq. (20)); it does not determine a guard for the aligned-spin BBH baseline used here.
Implementation mapping
Each native zappa_remnant_properties in
src/torchwave/approximants/imrphenom_nsbh.py and
src/jaxwave/approximants/imrphenom_nsbh.py independently sets a mass or
spin factor above one to one. Both also set both factors to one when
(a < 0 and nu < 0.188) or a < -0.5. For \((3,0,100)\), factor clipping is
active although the unclipped total quantities remain physical. For
\((5,-0.5,100)\), factor clipping and the nu < 0.188 fallback are active;
\(a=-0.5\) does not satisfy the separate strict a < -0.5 test. These are
descriptions of native behavior, not source-derived rules. This audit does
not change either provider.
Evidence and limitations
The cited paper supplies fit coefficients and broad physical constraints but not the predicates or output substitutions needed for a unique scalar guard reference. In particular, the paper's “agree with BBH” and “\(X_\bullet=1\)” formulations are unequal, and it does not identify the native \(\nu=0.188\) threshold. A gate that compared the native guarded outputs to a scalar transcription of those native rules would be circular; a gate that imposed one interpretation of the paper would invent a rule. The exact printed-coefficient intervals above diagnose this gap without inferring missing coefficient digits or using LAL source/binary output. Resolving the guard policy requires an explicit author prescription or a new project policy labeled separately from paper qualification. NR accuracy, negative-spin physical bounds over a wider domain, derivatives, and whole-waveform assembly remain unqualified.
Reproduce the checks
From the repository root, run
PYTHONPATH=src python3 -m pytest -q tests/test_nsbh_remnant_guard_source_boundary.py
for the source-only diagnostic. The test uses bbh_remnant_interval and
zappa_factor_interval with the two tuples above and asserts that both
printed-coefficient factors exceed one while the full remnant values remain
physical; it makes no native guarded-output comparison. The existing
guard-inactive gate is
PYTHONPATH=src python3 -m pytest -q tests/test_nsbh_remnant_rounding_math_reference.py
where Torch/JAX are installed. Check the six-section catalog with
PYTHONPATH=src python3 scripts/generate_waveform_library_table.py --check.