GPU-Powered Simulations of Seismic Waves in Nonlinear Media

Three-dimensional wave propagation simulations help to predict source directivity and basin amplification effects that often lead to strong shaking and structural damage during large earthquakes. With the deployment of GPU-based supercomputers, the resolution of such simulations has been significantly improved, allowing seismologists to resolve the wavefield at frequencies relevant for common buildings (0 - 10 Hz).  During this presentation I will talk about the challenge of accounting for nonlinear material behavior in such high-frequency simulations, and discuss the implementation of Drucker-Prager plasticity in our highly optimized finite difference AWP-ODC GPU code.  I will also show results of recent simulations on NCSA's Blue Waters and ORNL's Titan, which show that nonlinearity near the fault and in shallow sediments could drastically reduce the level of ground shaking that must be expected during future M 7.8 earthquakes on the southern San Andreas fault.

Dr. Daniel Roten is a Computational Research Seismologist in the Department of Geological Sciences at San Diego State University. His main research interest concerns the prediction and understanding of strong ground motion during earthquakes, especially with respect to nonlinear effects near the source and on soft sediments.  He also has an affiliated research appointment with the High-Performance GeoComputing Lab of the San Diego Supercomputer Center, where he is part of the team that won the 2015 NVIDIA global impact award and which was an ACM Gordon Bell finalist in 2010.  He received his PhD in seismology from ETH Zurich in 2007.