What You Don't Know Can Hurt You: Use and Abuse of Astrophysical Models in Gravitational-wave Population Analyses

TL;DR

This paper examines the use and limitations of astrophysical models in gravitational-wave data analysis, highlighting biases and uncertainties in inferring black hole formation channels from current and future observations.

Contribution

It demonstrates the potential biases and uncertainties in population synthesis-based analyses of black hole formation channels using gravitational-wave data.

Findings

Unobserved channels are inconsistent with GWTC-3 data.

Uncertainty in branching ratios decreases with larger catalogs.

Biases can be as large as 50% when analyzing simulated universes.

Abstract

One of the goals of gravitational-wave astrophysics is to infer the number and properties of the formation channels of binary black holes (BBHs); to do so, one must be able to connect various models with the data. We explore benefits and potential issues with analyses using models informed by population synthesis. We consider 5 possible formation channels of BBHs, as in Zevin et al. (2021b). First, we confirm with the GWTC-3 catalog what Zevin et al. (2021b) found in the GWTC-2 catalog, i.e. that the data are not consistent with the totality of observed BBHs forming in any single channel. Next, using simulated detections, we show that the uncertainties in the estimation of the branching ratios can shrink by up to a factor of $\sim 1.7$ as the catalog size increases from $50$ to $250$, within the expected number of BBH detections in LIGO-Virgo-KAGRA's fourth observing run. Finally, we show that this type of analysis is prone to significant biases. By simulating universes where all sources originate from a single channel, we show that the influence of the Bayesian prior can make it challenging to conclude that one channel produces all signals. Furthermore, by simulating universes where all 5 channels contribute but only a subset of channels are used in the analysis, we show that biases in the branching ratios can be as large as $\sim 50\%$ with $250$ detections. This suggests that caution should be used when interpreting the results of analyses based on strongly modeled astrophysical sub-populations.