[Paper Review] Mass and tidal parameter extraction from gravitational waves of binary neutron stars mergers using deep learning

Shriya Soma, H. Stöcker+1 more|arXiv (Cornell University)|Jun 30, 2023

Pulsars and Gravitational Waves ResearchPhysics and Astronomy82 references3 citations

TL;DR

This paper proposes a deep learning pipeline for classifying gravitational wave signals from binary neutron star (BNS) mergers and estimating their component masses and tidal deformability. Using a two-stage neural network—first classifying BNS vs. binary black hole signals, then regressing mass and tidal parameters—it achieves high-accuracy parameter estimation, demonstrating deep learning as an efficient alternative to conventional Bayesian inference for constraining neutron star equations of state.

ABSTRACT

Gravitational Waves (GWs) from coalescing binaries carry crucial information about their component sources, like mass, spin and tidal effects. This implies that the analysis of GW signals from binary neutron star mergers can offer unique opportunities to extract information about the tidal properties of NSs, thereby adding constraints to the NS equation of state. In this work, we use Deep Learning (DL) techniques to overcome the computational challenges confronted in conventional methods of matched-filtering and Bayesian analyses for signal-detection and parameter-estimation. We devise a DL approach to classify GW signals from binary black hole and binary neutron star mergers. We further employ DL to analyze simulated GWs from binary neutron star merger events for parameter estimation, in particular, the regression of mass and tidal deformability of the component objects. The results presented in this work demonstrate the promising potential of DL techniques in GW analysis, paving the way for further advancement in this rapidly evolving field. The proposed approach is an efficient alternative to explore the wealth of information contained within GW signals of binary neutron star mergers, which can further help constrain the NS EoS.

Motivation & Objective

• To overcome computational bottlenecks in conventional matched-filtering and Bayesian inference for gravitational wave parameter estimation.
• To develop a deep learning pipeline that classifies binary neutron star (BNS) merger signals from binary black hole (BBH) signals.
• To estimate component masses and tidal deformability (Λ̃) from simulated BNS gravitational waveforms using end-to-end regression.
• To enable efficient, scalable analysis of future BNS detections with advanced detectors like the Einstein Telescope and Cosmic Explorer.
• To support constraints on the neutron star equation of state using mass and tidal parameter estimates.

Proposed method

• A convolutional neural network (CNN) is trained to classify gravitational wave signals as originating from BNS or BBH mergers using time-domain waveforms.
• A separate regression network is trained to predict component masses and combined tidal deformability (Λ̃) from the same time-domain waveforms.
• Simulated waveforms are generated using the IMRPhenomPv2_NRTidalv2 approximant, covering a range of masses and tidal parameters.
• The training data includes noise-embedded waveforms to improve robustness to detector noise in real-world conditions.
• The pipeline combines classification and regression networks into a two-stage workflow for end-to-end analysis of BNS signals.
• The method is evaluated using synthetic signals with known parameters, enabling quantitative assessment of estimation accuracy.

Experimental results

Research questions

• RQ1 Can deep learning effectively classify gravitational wave signals from binary neutron star mergers versus binary black hole mergers?
• RQ2 To what extent can deep learning regress component masses and tidal deformability from noisy gravitational wave signals?
• RQ3 How does the performance of deep learning compare to conventional Bayesian inference in terms of accuracy and computational cost?
• RQ4 Can the proposed pipeline be extended to include additional parameters such as spin, inclination, and distance?
• RQ5 How do limitations in waveform modeling (e.g., 5PN tidal effects, frequency cutoff at 2048 Hz) affect the reliability of the estimated parameters?

Key findings

• The deep learning classifier achieves high accuracy in distinguishing BNS from BBH gravitational wave signals, even in the presence of detector noise.
• The regression network successfully estimates component masses and combined tidal deformability (Λ̃) with low mean absolute error, demonstrating robustness to noise and waveform model limitations.
• The method provides a computationally efficient alternative to traditional Bayesian parameter estimation, significantly reducing processing time for large-scale analyses.
• The estimated tidal deformability (Λ̃) values are consistent with known physical trends, supporting their use in constraining the neutron star equation of state.
• The pipeline is extensible and can be adapted to include additional parameters such as spin and inclination in future work.
• The study identifies limitations, including a frequency cutoff at 2048 Hz that excludes post-merger and ringdown information, and model dependence on the IMRPhenomPv2_NRTidalv2 waveform approximant.