Parameters and conventions
Use these conventions when setting up a waveform or comparing providers. Model-specific options and supported domains are defined by the waveform catalog, the T-family guide and the API reference. Defaults should be matched explicitly in cross-package comparisons.
Physical parameters
| Quantity | Convention used in the documented CBC examples | Checks when changing models or providers |
|---|---|---|
| Component masses | mass1, mass2 in solar masses, with mass1 >= mass2 |
Keep each spin and tidal parameter attached to its body if reordering masses. |
| Distance | distance in Mpc |
Match the distance and polarization projection before comparing amplitudes. |
| Spins | Dimensionless aligned chi1, chi2, or Cartesian spin1x through spin2z for supported precessing models |
Match the reference frequency, source frame and precession prescription. A matching vector alone does not establish matching conventions. |
| Tidal deformabilities | Dimensionless lambda1, lambda2 |
Use a model that supports the required tidal prescription. |
| Frequencies | Hz; uniform (f_lower, f_final, delta_f) or sample_frequencies for supported FD paths |
Match the actual bins, endpoint and cutoff policies. |
| Time | Seconds for sampled TD waveforms | Match the time origin, duration and sampling. Native T-family sampling differs from LAL's automatic epoch and length. |
| Angles | Radians | Match inclination and distinguish orbital reference phase from any waveform phase alignment used in validation. |
See generation examples for supported arguments. For eccentric models, record the eccentricity and its reference-frequency convention when comparing providers; the catalog alone does not establish a shared eccentricity convention.
Reference phase, frames and modes
Phase and reference-frequency behavior must be checked per model. For the
T family, phic is orbital phase at
f_ref; an omitted or zero f_ref uses f_lower in the sampling wrapper.
Its peak is located at tc on the chosen time grid. These rules must not be
assumed for every other approximant.
A mode label also needs a frame and projection convention. The T-family guide specifies aligned or coprecessing mode pairs and how selecting one harmonic includes both signs. The generation guide shows FD mode selection. Use the model's supported mode set rather than assuming all models with “HM” in their names contain the same harmonics.
For LAL comparisons, consult known differences for reference-phase, carrier, precession and calibration choices. Record the full options alongside the approximant name.
Outputs and batching
Waveform generation returns plus and cross polarizations. In the documented
FD examples, a single waveform has shape (N,) and a batch has shape (B, N).
Native T-family batches use a shared time grid; return_times=True also returns
that grid. See generation and
execution for examples and transformation restrictions.
Domains and invalid inputs
A calibration domain, an implemented input domain and a tested parameter set are different things. The validation records establish only the configurations they describe. Missing domain-wide evidence should be treated as unestablished.
Invalid-input behavior is also model-specific. For example, unsupported T-family MSA points raise errors in eager evaluation and produce NaNs for affected rows under JAX tracing. Its singular equal-mass precessing MSA limit is unsupported. The T-family restrictions explain alternative prescriptions and other rejected options. Do not assume that every model or transformed execution path rejects inputs in the same way.