Unveiling gravitational waves from core-collapse supernovae with MUSE

TL;DR

This work tackles the challenge of detecting gravitational waves from core-collapse supernovae, whose signals are weak and stochastic, by developing MUSE, a CNN-based pipeline trained on phenomenological waveforms to enable model-independent searches in Einstein Telescope data. MUSE leverages time-frequency spectrograms as input, a Mini Inception-Resnet classifier, and curriculum learning to robustly distinguish CCSN signals from noise. Evaluations on a representative set of 3D CCSN simulations show the 2L ET configuration with a 45° inclination yields the best performance, with detectability of Kuroda2016-like signals above $90\%$ efficiency at 50 kpc; results vary with waveform loudness and complexity. The study also outlines plans to apply MUSE to real O3 data and to broaden the waveform catalog, including magneto-rotational CCSN signals, to extend sensitivity for future gravitational-wave astronomy.

Abstract

The core collapse of a massive star at the end of its life can give rise to one of the most powerful phenomena in the Universe. Because of violent mass motions that take place during the explosion, core-collapse supernovae have been considered a potential source of detectable gravitational waveforms for decades. However, their intrinsic stochasticity makes ineffective the use of modelled techniques such as matched filtering, forcing us to develop model independent technique to unveil their nature. In this work we present MUSE pipeline, which is based on a classification procedure of the time-frequency images using a Convolutional Neural Network. The network is trained on phenomenological waveforms that are built to mimic the main common features observed in numerical simulation. The method is finally tested on a representative 3D simulation catalog in the context of Einstein Telescope, a third generation GW telescope. Among the three detector geometries considered here, the 2L with a relative inclination of $45^\circ$ is the one achieving the best results, thus being able to detect a Kuroda2016-like waveform with an efficiency above $90\%$ at 50 kpc.

Figures (6)

• Figure 1: (Left) Time series of $h_{+}$ and $h_{\times}$ representations of waveforms. (Right) GW waveforms are embedded in design-sensitivity noise of the Einstein Telescope (ET), assuming a triangular configuration. We generate time-frequency spectrograms for each ET channel (ET1, ET2, ET3), using their respective RGB colors (red, green and blue) that will be the input to the neural network.
• Figure 2: (top) three interferometer configurations among the ones discussed in Branchesi:2023mws for Einstein Telescope (ET): triangle configuration and 2L with parallel arms and relative inclination of 45$^\circ$. (bottom) ET sensitivity curve considered in Branchesi:2023mws for a 10 km arms interferometer compared with O3 sensitivities.
• Figure 3: Receiver Operating Characteristic (ROC) curves (efficiency $\eta$ vs against false alarm rate $FAR$) for three interferometer configurations considered in this work. The dashed grey line stands the random classifier, which makes predictions purely by chance, without learning from the data.
• Figure 4: $\theta$ distribution for test dataset of ET 2L with a relative inclination of 45$^\circ$. The vertical dashed line stands for the threshold value that allow to reach a FAR of 5%. Given the counts of the $i$th bin $c_i$ and its width $b_i$, we define the probability density as $c_i/(\sum_i^N c_i \times b_i)$, where $N$ is the total number of bins of the histogram.
• Figure 5: Detection efficiency $\eta$ (Eq. \ref{['eff']}) in function of source distance for five numerical simulation samples sharing the same progenitor mass of $15M_\odot$: (1) Andresen et al. 2019 Andresen2019; (2) Kuroda et al. 2016 Kuroda:2016; (3) Kuroda et al. 2017 Kuroda2017; (4) Mezzacappa 2020 Mezzacappa2020; (5) Yakunin et al. 2017 Yakunin2017. The curves are computed considering ET 2L with a relative inclination of $45^\circ$. The dashed line stands for the distance of the Large Magellanic Cloud (LMC).