Deep learning waveform anomaly detector for numerical relativity catalogs

TL;DR

This paper presents a deep learning model designed to detect anomalies in numerical relativity waveform catalogs, improving the accuracy of gravitational wave templates for better astrophysical parameter estimation.

Contribution

The study introduces a novel deep learning approach for identifying waveform anomalies in numerical relativity simulations, covering various binary black hole configurations.

Findings

Identified seven types of waveform anomalies in the SXS catalog.

Analyzed 1341 simulations with diverse mass ratios and spins.

Enhanced understanding of waveform irregularities in numerical relativity data.

Abstract

Numerical Relativity has been of fundamental importance for studying compact binary coalescence dynamics, waveform modelling, and eventually for gravitational waves observations. As the sensitivity of the detector network improves, more precise template modelling will be necessary to guarantee a more accurate estimation of astrophysical parameters. To help improve the accuracy of numerical relativity catalogs, we developed a deep learning model capable of detecting anomalous waveforms. We analyzed 1341 binary black hole simulations from the SXS catalog with various mass-ratios and spins, considering waveform dominant and higher modes. In the set of waveform analyzed, we found and categorised seven types of anomalies appearing in the coalescence phases.