gravpop

Gravitational-wave population inference for narrow population features with minimal edge bias.

This library allows one to perform a gravitational wave population analysis, (Hussain et al., Thrane et al.) that allows exploration of population features even in narrow regions near the edges of a bounded domain.

It is similar to gwpopulation (with model implementations as close as possible) explicitly has a numpyro backend and can implement the TGMM population analysis method to probe narrow features.

Feel free to jump to the tutorial here

The approach splits parameter space into two sectors:

• An analytic sector (\(\theta^a\)), where the population model is represented as a weighted sum of multivariate truncated normal distributions, allowing for an analytical computation of the population likelihood.

• A sampled sector (\(\theta^s\)), which accommodates more general population models and utilizes Monte Carlo estimates of the population likelihood.

This technique represents posterior samples using a truncated Gaussian mixture model (TGMM), where the population likelihood, \(p(x)\), is expressed as a sum of truncated multivariate Gaussian components:

\[ p(x) = \sum_k w_k \, \phi_{[a,b]}(x \mid \mu_k, \Sigma_k). \]

This form of the posterior allows analytic evaluation in the analytic sector, and falls back to using Monte-Carlo based estimation in the sampled sector.

For implementing the Truncated Gaussian Mixture Model fit, see truncatedgaussianmixtures, a package designed to fit data to mixtures of truncated Gaussians.

Contents:

• Gravpop tutorial
• Data
• Models
• Population Likelihood
• Sampling

Indices and tables

• Index

• Module Index

• Search Page