Decoding the Cosmic Orchestra: Reconstruction of Binary Black Hole Harmonics in LIGO Using Deep Learning

We present a deep-learning-based approach for accurate gravitational-wave (GW) signal reconstruction from coalescing binary black hole mergers detected by LIGO and Virgo interferometers. Traditional matched filtering methods, while optimal for stationary and Gaussian data, struggle with nonstationary noise transients and often overlook key physical effects such as precessing black hole spins and higher-order waveform harmonics, limiting the search for sources like intermediate-mass black hole binaries. Our model addresses these challenges by accurately reconstructing precessing binary black hole signals with higher-order modes, achieving high overlap with injected signals across diverse mass and spin configurations in real LIGO noise. Additionally, our framework, called AWaRe, demonstrates robustness in recovering GWs with features it has not been trained on, including higher black hole masses, eccentricity, and complex waveform modulations. Tested on real GW events, AWaRe achieves overlap between 85% and 98% with reconstructions from established methods such as Coherent WaveBurst and LALInference, and its uncertainty estimates closely match those from BayesWave. These results highlight the potential of deep learning in advancing GW signal analysis, enabling rapid, reliable signal reconstructions and opening new opportunities for astrophysical investigations and enhancements to current GW search methodologies.