Arianna Scarpa: "Connecting Astrophysics and GW observations with machine learning"

Wednesday 7 Jan 2026, 12:30 → 14:00 Europe/Rome

Sala Lauree

The latest Gravitational Wave Transient Catalogs (GWTCs) provide us with an unprecedented opportunity to understand how the universe is expanding and compact objects are formed. This opportunity comes with the burden of having to model the population distribution of compact binaries, in particular binary black holes and connecting these "phenomenological" models with astrophysical processes. To overcome the modelling difficulty, I introduce a non-parametric framework based on the machine learning technique of Normalizing Flows to model the joint mass-redshift distribution of binary black holes. The method replaces standard phenomenological prescriptions with Normalizing Flows capable of learning complex and multimodal distributions, allowing the model to capture mass-redshift correlations directly from astrophysical synthesis catalogs or from mock datasets. Normalizing Flows trained on redshift evolving synthetic populations accurately reconstruct the underlying features and remove the systematic bias on the Hubble constant, observed when using traditional parametric models. The framework can be extended to multiple formation channels through multiplicative factors, allowing joint inference of cosmology and absolute channel abundances. This methodology allows to quantify the agreement between astrophysical synthesis catalogs and gravitational wave observations: by fitting both simulated datasets and real events from the GWTC catalogs, the model provides a direct way to test the agreement between population synthesis predictions and current GW data.