Abstract: In the first part of our presentation, we introduce a groundbreaking advancement in computational fluid dynamics: the GP-MOOD method, a cutting-edge shock-capturing high-order algorithm. This innovative approach leverages Gaussian Processes (GP) to simulate compressible, shock-dominant fluid flows. Specifically, it incorporates the Multidimensional Optimal Order Detection (MOOD) strategy to manage solutions at shocks and discontinuities effectively. MOOD not only enhances numerical stability but also ensures precision through an adaptive shock-capturing framework. Expanding on this foundation, we delve into our pioneering integration of machine learning within the GP-MOOD framework. By employing a neural network (NN) to discern shock patterns, our NN-GP-MOOD approach achieves an unparalleled equilibrium between high-order accuracy and stability, dynamically adjusting to local flow conditions. This sophisticated strategy optimizes numerical accuracy and stability while maintaining the overall predictive capabilities in simulating shock-dominant flows. If time allows, we will provide a brief overview of our ongoing real-world application—an ambitious project focused on developing computer simulation models that aim to explore safety mechanisms within the Advanced Photon Source Upgrade (APS-U) at the Argonne National Laboratory (ANL) in Chicago. We showcase the practical implications and versatility of our methodology in tackling complex engineering challenges.

Short Bio: Dongwook Lee is Professor and Graduate Director of the Applied Mathematics Department at UC Santa Cruz. Dongwook’s research interests emphasize developing numerical schemes of high-order shock-capturing methods for computational magnetohydrodynamics (MHD) and gas dynamics on large-scale computing architectures. Dongwook applies these numerical methods to investigate nonlinear flow problems in astrophysics and high-energy-density physics. Before joining UC Santa Cruz, he worked as a senior applied mathematician at the Flash Center for Computational Science at the University of Chicago. Dongwook was the primary code architect of the unsplit hydrodynamics and magnetohydrodynamics solvers in the FLASH code from 2006 to 2014. Dongwook received his Ph.D. in Applied Mathematics and Scientific Computation from the University of Maryland at College Park in 2006.
 

Location:

Location: Building 401, Conf Room B2100
Teams Virtual Link: ASD Seminar Prof Lee