A systematic study of gravitational wave population inference with density estimation.

Astrophysical model parameters derived from population analyses of gravitational wave (GW) events provide insights into astrophysics, including black hole formation channels, the neutron star equation of state, and cosmology. These properties can be inferred using hierarchical Bayesian statistics that account for selection effects. However, due to the complexities of gravitational wave waveform models, single event parameter estimation limits the number of posterior samples describing the GW data therefore limiting the validity of population analyses due to increased uncertainty in the evaluation of Monte Carlo integrals as catalog sizes increase. We conduct a systematic study of GW population inference by using density estimation methods to model the gravitational wave event posterior distributions. This approach allows for an arbitrary number of samples drawn from the density estimate, as well as evaluation of the posterior probability function at any point in the parameter space, thereby circumventing the accuracy constraints typically imposed by Monte Carlo integration and expanding our ability to infer complex astrophysics involving exact physical relations, such as the neutron star equation of state and cosmological inference. We quantify the uncertainty in analysis with flexible sampling and compare it to the standard population inference approach which directly uses samples from parameter estimation.