4–5 Mar 2024
European Gravitational Observatory
Europe/Rome timezone

Deep learning to detect gravitational waves from binary close encounters: fast parameter estimation using normalizing flows

4 Mar 2024, 16:20
20m
Auditorium (European Gravitational Observatory)

Auditorium

European Gravitational Observatory

Via E. Amaldi,5 56021 Cascina (PI) - Italy
Talk Session 3

Speaker

Federico De Santi

Description

Among astrophysical gravitational waves sources yet undetected, of great interest are the binary close encounters involving black holes and/or neutron stars. These systems are characterized by high orbital eccentricities and form via dynamical interactions in dense stellar environments, like globular clusters or Active Galactic Nuclei disks. Their detection could shed light on the different formation channels and could allow tests of General Relativity in the strong field regime as well as multimessenger observations.

The expected gravitational wave emission from these events differs from standard coalescences, being instead characterized by repeated short duration bursts emitted at each periastron passage of the two objects. The burst nature of the signal paired with the expected low signal-to-noise ratio makes them a challenging source either for detection or parameter estimation with traditional data analysis approaches. We present HYPERION (“HYP-er fast close EncounteR Inference from Observation with Normalizing-flows”): a novel data analysis pipeline based on probabilistic machine learning and normalizing flows to infer Bayesian posterior. We show that our method is very promising, since it can make detection and inference several orders of magnitude faster than traditional techniques while maintaining high accuracy on the parameters, thus showing how machine learning could help in studying these sources.

Primary author

Co-authors

massimiliano razzano Francesco Fidecaro (University of Pisa and INFN) Luca Muccillo Lucia Papalini (Pisa) Dr Barbara Patricelli (University of Pisa and INFN - Pisa)

Presentation materials