A machine-learning classifier for the postmerger remnant of binary neutron stars

Abstract

Knowing the kind of remnant produced after the merger of a binary neutron star system, e.g., if a black hole forms or not, would not only shed light on the equation of state describing the extremely dense matter inside neutron stars, but also help understand the physical processes involved in the postmerger phase. Moreover, in the event of a gravitational-wave detection, predicting the presence of a neutron star remnant is crucial in order to advise potential electromagnetic follow-up campaigns. In this work, we use Gradient Boosted Decision Trees and publicly available data from numerical-relativity simulations to construct a classifier that predicts the outcome of binary neutron star mergers, based on the binary's parameters inferred from gravitational-wave inspiral signals: total mass, mass-weighted tidal deformability, mass ratio, and effective inspiral spin. Employing parameters that can be estimated from the inspiral part of the signal only allows us to predict the remnant independently on the detection of a postmerger gravitational-wave signal. We build three different classifiers to distinguish between various potential scenarios, we estimate their accuracy and the confidence of their predictions. Finally, we apply the developed classifiers to real events data, finding that GW170817 most likely led to the formation of a hypermassive neutron star, while GW190425 to a prompt collapse to a black hole.