Classifying the nuclear equation of state in LVK interferometric noise through core-collapse supernova gravitational-wave signatures using convolutional neural networks

Abstract

This paper presents a convolutional neural network (CNN) approach to classifying the nuclear equation of state (EOS). As illustrative examples, we use five two-dimensional core-collapse supernova (CCSN) simulations that differ only in their EOS. We analyze estimates of the initial slope of the high-frequency feature (HFF) reconstructed in real interferometric data from the O3b LIGO-Virgo-KAGRA (LVK) observing run at Galactic source distances of 1, 5, and 10 kpc. The CNN classifier achieves an overall accuracy of 98.58% at 1 kpc and 52.43% at 5 kpc. At 10 kpc, its ability to distinguish among the EOS classes is effectively lost. The successful EOS classification at 1 kpc suggests that this approach may be scalable to next-generation observatories. The expected order-of-magnitude sensitivity improvements of Cosmic Explorer and the Einstein Telescope could enable comparable classification performance at approximately ten times the current distance. More detailed performance metrics, including the macro-averaged one-vs-rest (OvR) area under the curve (AUC), yield values of 0.97 and 0.98 at 1 kpc. These results indicate strong classification performance both across the complete set of EOS classes and for the individual classes.