Searching for binary black hole sub-populations in gravitational wave data using binned Gaussian processes

TL;DR

This paper introduces a non-parametric, binned Gaussian process model to identify and analyze multiple subpopulations of binary black holes in gravitational wave data, revealing distinct formation channels and their characteristics.

Contribution

The study presents a novel non-parametric approach using binned Gaussian processes to detect and characterize multiple binary black hole subpopulations in gravitational wave data.

Findings

Identified two distinct subpopulations with different mass and spin features.

Found the high-spin subpopulation lacks a feature at 30-40 solar masses, aligning with stellar binary evolution.

Observed the low-spin subpopulation peaks around 30 solar masses, consistent with dynamical formation in clusters.

Abstract

Astrophysically motivated population models for binary black hole observables are often insufficient to capture the imprints of multiple formation channels. This is mainly due to the strongly parametrized nature of such investigations. Using a non-parametric model for the joint population-level distributions of binary black hole component masses and effective inspiral spins, we find hints of multiple subpopulations in the third gravitational wave transient catalog. The higher (more positive) spin subpopulation is found to have a mass-spectrum without any feature at the $30-40M_{\odot}$, which is consistent with the predictions of isolated stellar binary evolution, simulations for which place the pile up due to pulsational pair-instability supernovae near $50M_{\odot}$ or higher. The other sub-population with effective spins closer to zero shows a feature at $30-40M_{\odot}$ and is consistent with binary black holes formed dynamically in globular clusters, which are expected to peak around $30M_{\odot}$. We also compute merger rates for these two subpopulations and find that they are consistent with the theoretical predictions of the corresponding formation channels. We validate our results by checking their robustness against variations of several model configurations and by analyzing large simulated catalogs with the same model.