An Implementation to Identify the Properties of Multiple Population of Gravitational Wave Sources

The rapidly increasing sensitivity of gravitational wave detectors is enabling the detection of a growing number of compact binary mergers. These events are crucial for understanding the population properties of compact binaries. However, many previous studies rely on computationally expensive inference frameworks, limiting their scalability. In this work, we present GWKokab, a JAX-based framework that enables modular model building with independent rate for each subpopulation such as BBH, BNS, and NSBH binaries. It provides accelerated inference using the normalizing flow based sampler called flowMC and is also compatible with NumPyro samplers. To validate our framework, we generated two synthetic populations, one comprising spinning eccentric binaries and the other circular binaries using a multi-source model. We then recovered their injected parameters at significantly reduced computational cost and demonstrated that eccentricity distribution can be recovered even in spinning eccentric populations. We also reproduced results from two prior studies: one on non-spinning eccentric populations, and another on the BBH mass distribution using the third Gravitational Wave Transient Catalog (GWTC-4). We anticipate that GWKokab will not only reduce computational costs but also enable more detailed subpopulation analyses such as their mass, spin, eccentricity, and redshift distributions in gravitational wave events, offering deeper insights into compact binary formation and evolution.

💡 Research Summary

The paper introduces GWKokab, a new JAX‑based framework designed to perform fast and flexible population inference for gravitational‑wave (GW) sources. As GW detectors become increasingly sensitive, the number of observed compact‑binary mergers (binary black holes, binary neutron stars, and neutron‑star‑black‑hole binaries) is rapidly growing, demanding scalable statistical tools. Existing frameworks such as POPMODELS, GWPopulation, and GWInfer can model sub‑populations but suffer from high computational cost, limited support for independent redshift or eccentricity distributions, and difficulty handling mixture models.

GWKokab addresses these limitations through three main innovations. First, it adopts a modular architecture where each sub‑population i is described by its own intrinsic rate density R_i(λ) and a normalized parameter distribution p_i(λ|Λ_i). The total source‑frame merger rate density is then a simple sum ρ(λ|Λ)=∑_i R_i(λ)p_i(λ|Λ_i). Redshift evolution is parametrized by a power‑law (1+z)^κ_i, allowing independent evolution for each sub‑population. Second, the framework implements hierarchical Bayesian inference: individual event posteriors ℓ_j(λ) (obtained from standard PE pipelines) are combined with the population model via an inhomogeneous Poisson likelihood L(Λ)∝exp