External Attention Transformer: A Robust AI Model for Identifying Initial Eccentricity Signatures in Binary Black Hole Events in Simulated Advanced LIGO Data

Abstract

Initial orbital eccentricities of gravitational wave (GW) events associated with merging binary black holes (BBHs) should provide clues to their formation scenarios, mainly because various BBH formation channels predict distinct eccentricity distributions. However, searching for inspiral GWs from eccentric BBHs is computationally challenging due to sophisticated approaches to model such GW events. This ensures that Bayesian parameter estimation methods to characterize such events are computationally daunting. These considerations influenced us to propose a novel approach to identify and characterize eccentric BBH events in the LIGO-Virgo-KAGRA (LVK) collaboration data sets that leverages external attention transformer models. Employing simulated data that mimic LIGO O4 run, eccentric inspiral events modeled by an effective-one-body numerical-relativity waveform family, we show the effectiveness of our approach. By integrating this transformer-based framework with a convolutional neural network (CNN) architecture, we provide efficient way to identify eccentric BBH GW events and accurately characterize their source properties.