Binary Neutron Star Merger Search Pipeline Powered by Deep Learning

Abstract

Gravitational waves are now routinely detected from compact binary mergers, with binary neutron star mergers being of note for multi-messenger astronomy as they have been observed to produce electromagnetic counterparts. Novel search pipelines for these mergers could increase the combined search sensitivity, and could improve the ability to detect real gravitational wave signals in the presence of glitches and non-stationary detector noise. Deep learning has found success in other areas of gravitational wave data analysis, but a sensitive deep learning-based search for binary neutron star mergers has proven elusive due to their long signal length. In this work, we present a deep learning pipeline for detecting binary neutron star mergers. By training a convolutional neural network to detect binary neutron star mergers in the signal-to-noise ratio time series, we concentrate signal power into a shorter and more consistent timescale than strain-based methods, while also being able to train our network to be robust against glitches. We compare our pipeline's sensitivity to the three offline detection pipelines using injections in real gravitational wave data, and find that our pipeline has a comparable sensitivity to the current pipelines below the 1 per 2 months detection threshold. Furthermore, we find that our pipeline can increase the total number of binary neutron star detections by 12% at a false alarm rate of 1 per 2 months. The pipeline is also able to successfully detect the two binary neutron star mergers detected so far by the LIGO-Virgo-KAGRA collaboration, GW170817 and GW190425, despite the loud glitch present in GW170817.