Pre-Merger Detection and Characterization of Inspiraling Binary Neutron Stars Derived from Neural Posterior Estimation

TL;DR

This paper introduces a machine learning framework that detects and characterizes inspiraling binary neutron star signals in gravitational wave data, providing increasingly precise estimates as the merger approaches, aiding early multi-messenger observations.

Contribution

It presents a novel neural network-based approach combining residual networks and normalizing flows for early detection and parameter estimation of neutron star mergers.

Findings

Effective detection of inspiraling neutron star signals.

Improved parameter estimation closer to merger time.

Enhanced sky localization accuracy over time.

Abstract

As the sensitivity of the international gravitational wave detector network increases, observing binary neutron star signals will become more common. Moreover, since these signals will be louder, the chances of detecting them before their mergers increase. However, this requires an efficient framework. In this work, we present a machine-learning-based framework capable of detecting and analyzing binary neutron star mergers during their inspiral. Using a residual network to summarize the strain data, we use its output as input to a classifier giving the probability of having a signal in the data, and to a normalizing-flow network to perform neural posterior estimation. We train a network for several maximum frequencies reached by the signal to improve the estimate over time. Our framework shows good results both for detection and characterization, with improved parameter estimation as we get closer to the merger time. Thus, we can effectively evolve the precision of the sky location as the merger draws closer. Such a setup would be important for future multi-messenger searches where one would like to have the most precise information possible, as early as possible.