Narrow Population Inference Enhanced by Analytical Likelihood Models

Abstract

The growing catalog of gravitational-wave events has revealed substantial diversity in the properties of compact-binary mergers. However, commonly used population-inference methods based on discrete posterior samples can struggle to constrain narrow population features, resulting in biased or unstable estimates of population hyperparameters. We first demonstrate this limitation using a toy population model by comparing parameter recovery with a continuous likelihood model against discrete approximations constructed from $10^3$, $10^4$, and $10^5$ samples. We then perform the same comparison using synthetic eccentric and multisource populations introduced in previous studies. Although the continuous and discrete approaches yield broadly consistent results, the continuous approximation more accurately recovers the parameters of narrow simulated populations. In particular, while both methods produce similar mass distributions, appreciable differences arise for narrowly distributed parameters such as spin and eccentricity. Our results indicate that the continuous approach provides more reliable inference for spin and eccentricity, whose narrow population distributions can be inadequately represented by finite sample sets. Continuous likelihood models therefore offer a valuable tool for improving population inference and extracting more robust information about the formation and evolution of compact-binary systems.