UC Santa Cruz

Chaotic fluid motion, known at the small scales as turbulence, can significantly alter the large-scale evolution of astrophysical events. For example, the growth and impact of convection to produce a successful core-collapse supernova (CCSN) depend upon the evolution of turbulence. The ideal way to investigate it in such an environment would be with high-resolution direct numerical simulations (DNS) that resolve the turbulent energy cascade down to some small scale where the energy could be safely assumed to dissipate as heat. Unfortunately, given the high Reynolds number and a large range of spatial scales, this is well beyond the current state-of-the-art computational 3D CCSN models. Since turbulence cannot be properly simulated in 1D or 2D, a subgrid-scale model (SGS) is needed to capture the unresolved 3D physics. Simple analytical SGS models often lack accuracy, though, and complex ones are difficult to tune, resulting in limited generalizability based on initial conditions. Given the recent successes in turbulence SGS modeling with Machine Learning (ML) in adjacent fields, this thesis develops an ML algorithm to analyze current simulations and study the features of turbulence in CCSN. To demonstrate its efficacy, we test our ML approach on modeling dynamic 3D HD & MHD turbulence, integrating the former into a 1D code to study the role of one specific feature of turbulence (the effective turbulent pressure) on the fate of CCSN simulations. Furthermore, our ML tools can be used to study the broader effects of turbulence, explore other ML architectures, and be integrated in the outside 1- and multi-dimensional CCSN codes. The ML framework (Sapsan) and its implementation into a 1-dimensional code (COLLAPSO1D) are open-sourced and available for public use.