SCEC Award Number 18211 View PDF
Proposal Category Collaborative Proposal (Integration and Theory)
Proposal Title Nonlinear Earthquake Simulations Through Robust and Accurate A Posteriori Sub-Cell Limiting
Investigator(s)
Name Organization
Alexander Breuer University of California, San Diego Yifeng Cui University of California, San Diego
Other Participants Graduate Student
SCEC Priorities 4a, 4c, 4b SCEC Groups CS
Report Due Date 03/15/2019 Date Report Submitted 05/14/2019
Project Abstract
 This project extended the Extreme Scale Discontinuous Galerkin Environment (EDGE) with a robust and accurate a posteriori finite volume sub-cell limiter. Sub-cell limiting is the new fundamental building block of the software to tackle demanding hyperbolic problems with strong non-linearities, especially in the context of emerging challenges in earthquake science. Further, over the duration of this project, we optimized and analyzed EDGE on commercial cloud solutions for high performance computing. A respective seismic simulation on 768 AWS c5.18xlarge instances, comprising 27,648 cores, sustained a performance of 1.09 FP32-PFLOPS [2]. This is, to the best of our knowledge, the first work of its kind at such a large scale. All conducted work of this project is publicly available through EDGE’s online resources. This includes all pre-processing scripts, the solver itself, and data, e.g., the layout of sub-grids or pre-computed operators.

The project covered four major steps, which are detailed throughout the project report. First, we derived sub-cell grids and operators in pre-processing through symbolic computations. Second, we included these pre-computed data structures in the main solver, running at high performance. Third, we enabled the a posteriori sub-cell limiting in the main solver, which requires the evaluation of admissibility criteria, projections of the DG-solution to the sub-cell-solution for inadmissible elements, time stepping of the sub-cell solution, and potential projections of the sub-cell solution to the DG-solution. Fourth, we tested the implemented procedure, when simulating seismic wave propagation.
Intellectual Merit As formalized in the 2018 Science Plan, studying nonlinear effects on all levels is a priority of SCEC5. A corresponding research priorities is, for example, P2.c, or the entire basic research question Q4. Enabling DG methods for nonlinear earthquake simulations prepares SCEC’s numerical toolbox for these challenges. Here, we see large potential of our proposed limiter in the context of nonlinear dynamic rupture simulations and nonlinear material behavior and response in vicinity of the fault and free-surface.
Broader Impacts This project supported a student intern of the Summer Undergraduate Research Experience (SURE).
Intended learning outcome is the ability to design a small and reliable software system in team- and project-based setting. The project will apply Scientific Computing and High Performance Computing (HPC) in the context of earthquake science.
The student intern studied the applicability of neural networks for seismic source inversion.
The training phase used artificial synthetics, generated from SGT solutions of the forward solver.
Exemplary Figure Fig. 2