Machine learning based source properties inference for fourth observing run of LIGO-Virgo-KAGRA

Prompt identification of gravitational wave (GW) sources capable of producing electromagnetic emissions is essential to extracting the maximum amount of physical information, as it allows astronomers to launch observing campaigns within minutes of the GW triggers. Timely sky localization of electromagnetically bright events, such as GW170817, makes possible the identification of the progenitor's host galaxy, the study of relativistic jet and non-relativistic ejecta formation, and detailed analyses of the post-merger phase. Astronomers are also interested in following up GW sources whose component masses lie in the "lower mass-gap" region between neutron stars and black holes (3-5 solar masses). In this work, we present a machine learning-based suite of algorithms for the inference of GW source properties that provide (1) the probability that at least one of the compact objects in the source binary progenitor has a mass consistent with the mass of a neutron star, (2) the probability that the merger ejected a nonzero mass outside the final remnant compact object, and (3) the probability that either of the component masses lie in the lower mass-gap region with high confidence and within seconds of a GW trigger. The algorithms provide these quantities within seconds and with an accuracy above 90%.