GENERAL RELATIVISTIC GRAVITATIONAL WAVES FROM GR EFFECTIVE POTENTIAL

Abstract

Core-collapse supernovae (CCSNe) are among the most powerful cosmic explosions that produce gravitational waves (GWs). The information about the explosion mechanism and stellar interior is present within the GWs. However, complete general relativity (GR) simulations of CCSNe are computationally demanding, so approximate techniques like the General Relativistic Effective Potential (GREP) approximation are required. In this thesis, machine learning and regression are formulated to act as a mapping function of efficient GREP simulations and accurate GR waveforms. We compare ridge regression and neural networks for projecting GREP-produced waveforms to their GR equivalents over four equations of state (SFHo, LS220, HSDD2, GShenFSU2.1). Our findings are that the most effective neural networks are symmetric in architecture (e.g. 80-40-40-80), achieving a waveform accuracy of RRMSE < 0.07 compared to full GR simulations in efficient computation. The trained models preserve key features like bounce timing and ringdown frequencies, enabling efficient generation of accurate waveforms for gravitational wave astronomy.