Writing gwconfig.yaml

The prior configuration file is a YAML document with four top-level sections: variable_params, priors, constraints, and waveform_transforms. Together they fully specify the parameter space from which every training waveform is drawn.

Complete reference example

variable_params:
  - mass1
  - mass2
  - ra
  - dec
  - inclination
  - coa_phase
  - polarization
  - chirp_distance
  - spin1_a
  - spin1_azimuthal
  - spin1_polar
  - spin2_a
  - spin2_azimuthal
  - spin2_polar
  - injection_time
  - tc

priors:

  injection_time:
    name: uniform
    min: 1238166018   # GPS start of O3a
    max: 1253977218   # GPS end of O3a

  tc:
    name: uniform
    min: 11.0         # earliest coalescence time (s) within the segment
    max: 11.2

  mass1:
    name: uniform
    min: 7.0
    max: 50.0

  mass2:
    name: uniform
    min: 7.0
    max: 50.0

  spin1_a:
    name: uniform
    min: 0.0
    max: 0.99

  spin1_orientation:
    name: uniform_solidangle
    polar-angle: spin1_polar
    azimuthal-angle: spin1_azimuthal

  spin2_a:
    name: uniform
    min: 0.0
    max: 0.99

  spin2_orientation:
    name: uniform_solidangle
    polar-angle: spin2_polar
    azimuthal-angle: spin2_azimuthal

  sky:
    name: uniform_sky
    ra: ra
    dec: dec

  inclination:
    name: sin_angle

  coa_phase:
    name: uniform_angle

  polarization:
    name: uniform_angle

  chirp_distance:
    name: uniform_radius
    min: 130.0
    max: 350.0

constraints:
  - name: mass_order

waveform_transforms:

  spin1_cartesian:
    name: spherical_to_cartesian
    x: spin1x
    y: spin1y
    z: spin1z
    radial: spin1_a
    polar: spin1_polar
    azimuthal: spin1_azimuthal

  spin2_cartesian:
    name: spherical_to_cartesian
    x: spin2x
    y: spin2y
    z: spin2z
    radial: spin2_a
    polar: spin2_polar
    azimuthal: spin2_azimuthal

  mass_params:
    name: mass1_mass2_to_mchirp_q

  distance:
    name: chirp_distance_to_distance

Section reference

variable_params

A flat list of every raw parameter name that is sampled from a prior. The order here does not affect sampling; the final output tensor uses alphabetically sorted names. Parameters produced only by transforms (spin1x, mchirp, distance, …) should not appear here — they are added automatically.

priors

Each entry maps a prior name to a distribution definition. The name field selects the distribution class (see Available Distributions). All other fields are passed as keyword arguments to the class constructor.

Field	Effect
name	Distribution type. One of uniform, uniform_angle, sin_angle, uniform_sky, uniform_solidangle, uniform_radius.
min / max	Bounds for uniform and uniform_radius.
polar-angle / azimuthal-angle	Output parameter names for uniform_solidangle joint sampler.
ra / dec	Output parameter names for uniform_sky joint sampler.

Changing the mass range

To restrict the search to neutron-star–black-hole (NSBH) binaries:

mass1:
  name: uniform
  min: 5.0
  max: 20.0

mass2:
  name: uniform
  min: 1.0
  max: 3.0

To search only equal-mass systems near the pair-instability gap:

mass1:
  name: uniform
  min: 40.0
  max: 80.0

mass2:
  name: uniform
  min: 40.0
  max: 80.0

Changing the distance range

Narrowing the chirp distance range increases the fraction of high-SNR events in the training set. Widening it extends coverage toward the detection horizon but produces more threshold-SNR examples. The uniform_radius distribution automatically assigns the correct \(r^2\) weight.

chirp_distance:
  name: uniform_radius
  min: 100.0   # Mpc  — closer / louder
  max: 600.0   # Mpc  — further / quieter

constraints

Named Boolean filters applied via rejection sampling after the base parameters are drawn. Currently one named constraint is available:

Name	Effect
mass_order	Enforces \(m_1 \geq m_2\) by swapping components when the sampled masses violate the ordering. This is a deterministic projection, not true rejection, so it does not waste samples.

To remove the mass-ordering constraint (useful when the waveform model already handles it):

constraints: []

waveform_transforms

Deterministic reparametrisations applied to the sampled base parameters before the tensor is returned. Three transforms are available:

name	Description
spherical_to_cartesian	Converts spherical spin components (a, polar, azimuthal) to Cartesian (x, y, z) via \(x = a \sin\theta \cos\phi\), \(y = a \sin\theta \sin\phi\), \(z = a \cos\theta\). Required by IMRPhenomPv2.
mass1_mass2_to_mchirp_q	Computes chirp mass \(\mathcal{M}_c\) and mass ratio \(q = m_1/m_2\) from the component masses. Added to the output tensor as mchirp and q.
chirp_distance_to_distance	Converts chirp distance to luminosity distance using the chirp mass: \(d_L = d_\text{chirp} \cdot \left(\mathcal{M}_c / 1.2\,M_\odot\right)^{5/6}\). Added as distance.

Transforms that are not needed can simply be omitted from the waveform_transforms block.