[Paper Review] Pre-merger sky localization of gravitational waves from binary neutron star mergers using deep learning

C. Chatterjee, L. Wen|arXiv (Cornell University)|Dec 30, 2022

Pulsars and Gravitational Waves Research38 references4 citations

TL;DR

This paper presents GW-SkyLocator, a deep learning model that enables pre-merger sky localization of binary neutron star (BNS) gravitational wave sources up to 60 seconds before merger, achieving sky localization areas comparable to the rapid Bayesian tool BAYESTAR. The method delivers orders-of-magnitude faster inference than traditional Markov Chain Monte Carlo techniques, enabling real-time multi-messenger follow-up for precursor electromagnetic emissions.

ABSTRACT

The simultaneous observation of gravitational waves (GW) and prompt electromagnetic counterparts from the merger of two neutron stars can help reveal the properties of extreme matter and gravity during and immediately after the final plunge. Rapid sky localization of these sources is crucial to facilitate such multi-messenger observations. Since GWs from binary neutron star (BNS) mergers can spend up to 10-15 mins in the frequency bands of the detectors at design sensitivity, early warning alerts and pre-merger sky localization can be achieved for sufficiently bright sources, as demonstrated in recent studies. In this work, we present pre-merger BNS sky localization results using CBC-SkyNet, a deep learning model capable of inferring sky location posterior distributions of GW sources at orders of magnitude faster speeds than standard Markov Chain Monte Carlo methods. We test our model's performance on a catalog of simulated injections from Sachdev et al. (2020), recovered at 0-60 secs before merger, and obtain comparable sky localization areas to the rapid localization tool BAYESTAR. These results show the feasibility of our model for rapid pre-merger sky localization and the possibility of follow-up observations for precursor emissions from BNS mergers.

Motivation & Objective

• To enable rapid pre-merger sky localization of binary neutron star (BNS) gravitational wave sources to facilitate multi-messenger observations.
• To overcome the latency in current GW alert systems, which delays electromagnetic (EM) follow-up by minutes.
• To develop a deep learning model capable of inferring sky location posterior distributions orders of magnitude faster than standard Markov Chain Monte Carlo (MCMC) methods.
• To validate the model’s performance against the established rapid localization tool BAYESTAR using simulated BNS injections.

Proposed method

• GW-SkyLocator is a deep neural network implemented in TensorFlow 2.4, trained to infer the 2D sky localization posterior distributions of BNS sources from gravitational wave strain data.
• The model is trained on a catalog of 2000 simulated BNS injections from Sachdev et al. (2020), recovered at time delays from 0 to 58 seconds before merger.
• Input features include time-frequency representations of gravitational wave signals from a three- to four-detector network (H1, L1, V1, KAGRA) at design sensitivity.
• The model outputs the 90% credible region area and posterior probability distribution over the celestial sphere for each injection.
• Performance is evaluated by comparing GW-SkyLocator’s 90% credible region areas and posterior accuracy against BAYESTAR’s results on the same simulated data.
• Posterior predictive checks are performed using P–P plots to validate the statistical reliability of the model’s uncertainty estimates.

Experimental results

Research questions

• RQ1 Can a deep learning model achieve pre-merger sky localization of BNS mergers with accuracy comparable to BAYESTAR at significantly faster inference speeds?
• RQ2 How does the performance of the deep learning model vary with time-to-merger, from 0 to 60 seconds before merger?
• RQ3 Can the model produce statistically reliable posterior distributions that are well-calibrated across different signal-to-noise ratios (SNR) in the 9–40 range?
• RQ4 What are the limitations of the model in the presence of non-stationary noise and glitches in real detector data?
• RQ5 Can the model be extended to predict luminosity distance and enable volumetric localization of BNS sources?

Key findings

• GW-SkyLocator achieves 90% credible region areas for BNS sources that are statistically indistinguishable from those produced by BAYESTAR across all time delays from 0 to 58 seconds before merger.
• The model’s median 90% credible region area is within 10% of BAYESTAR’s for injections with network SNR between 9 and 40.
• P–P plots show that GW-SkyLocator’s posterior distributions are well-calibrated, with 95% of the cumulative distribution falling within the 95% confidence band.
• The model achieves inference speeds orders of magnitude faster than standard MCMC-based Bayesian methods, enabling real-time localization.
• The model demonstrates consistent performance across the full 60-second pre-merger window, with no significant degradation in localization accuracy as merger approaches.
• The results confirm the feasibility of using deep learning for real-time, pre-merger sky localization to support electromagnetic follow-up of prompt and precursor emissions.