A Machine Learning Based Source Property Inference for Compact Binary Mergers

TL;DR

This paper introduces a machine learning approach to quickly infer the presence of neutron stars and remnant matter in binary mergers from gravitational wave data, aiding rapid follow-up observations.

Contribution

It presents a novel supervised machine learning method to assess merger properties in real time, improving the identification of potential electromagnetic counterparts.

Findings

Effective real-time inference of neutron star presence

Enhanced detection of remnant matter post-merger

Mitigation of statistical and systematic errors in parameter estimation

Abstract

The detection of the binary neutron star (BNS) merger, GW170817, was the first success story of multi-messenger observations of compact binary mergers. The inferred merger rate along with the increased sensitivity of the ground-based gravitational-wave (GW) network in the present LIGO/Virgo, and future LIGO/Virgo/KAGRA observing runs, strongly hints at detection of binaries which could potentially have an electromagnetic (EM) counterpart. A rapid assessment of properties that could lead to a counterpart is essential to aid time-sensitive follow-up operations, especially robotic telescopes. At minimum, the possibility of counterparts require a neutron star (NS). Also, the tidal disruption physics is important to determine the remnant matter post merger, the dynamics of which could result in the counterparts. The main challenge, however, is that the binary system parameters such as masses and spins estimated from the real time, GW template-based searches are often dominated by statistical and systematic errors. Here, we present an approach that uses supervised machine-learning to mitigate such selection effects to report possibility of counterparts based on presence of a NS component, and presence of remnant matter post merger in real time.