Uncovering the Origins of Binary Black Holes with Normalising Flows

The growing number of gravitational-wave detections from compact-objects binaries enables more precise measurements of the population properties. The observed population is most likely drawn from multiple formation channels. Population-synthesis simulations allow detailed modelling of each of these channels. Comparing these models with the observations allows us to constrain the uncertain physics of compact-object binary formation and evolution. However, the most detailed population-synthesis codes are highly computationally expensive, taking on the of order months to run for a single set of input parameters.

A solution to this computational challenge is the use of machine learning to emulate the predictions from population synthesis. We demonstrate the use of normalising flows to solve this problem, interpolating between the populations predicted for different simulation inputs, and using the emulated distributions for hierarchical inference. Using population synthesis from leading formation channels as training data, we apply our approach to the current catalog of gravitational-wave observations to infer the branching ratios of different channels and details of binary stellar evolution.