Poster #050, Ground Motions

Verification of Iwan-type Plasticity in the Discontinuous-Mesh GPU-powered Wave Propagation Code AWP

Daniel Roten, Kim B. Olsen, Steven M. Day, & Yifeng Cui
Poster Image: 

Poster Presentation

2021 SCEC Annual Meeting, Poster #050, SCEC Contribution #11636 VIEW PDF
The hysteretic stress-strain relationship of the shallow crust and its effects on strong ground motions and fault zone deformation represents one of SCEC’s research priorities and is the focus of a technical activity group. Using funding from SCEC, we have improved the capability of the AWP finite difference code to capture the realistic stress-strain relationship of soils by tracking multiple yield surfaces. This Iwan-type nonlinearity was first implemented in the computationally simpler, CPU-based version of AWP and verified against the nonlinear wave propagation code NOAH using 1D and 2D site response benchmarks. The CPU version of AWP with Iwan nonlinearity was also used to predict re...alistic nonlinear effects during a M 7.8 earthquake on the southern San Andreas fault. However, the computational cost of the Iwan model, which is roughly proportional to the number of yield surfaces (typically 10 or higher to obtain sufficient accuracy), and the limited parallel efficiency of the CPU version reduced the maximum frequency of these simulations to values of 1 Hz or less. This limit is further reduced in regions of strong nonlinearity, where shear-modulus degradation decreases the shear-wave velocity and shortens the seismic wavelength. In order to increase the resolution and efficiency of wave propagation with realistic nonlinearity, we have implemented the Iwan model in the discontinuous-mesh, GPU-accelerated code version AWP-GPU-DM. The code supports multiple yield surfaces in the uppermost, highest-resolution mesh which is dimensioned to encompass shallow low-velocity sediments. At greater depth, the grid spacing is increased stepwise (by a factor of 3 at each DM interface) to represent higher-velocity, linear rocks. We verify the code’s accuracy by reproducing nonlinear site response in the time and frequency domains using the same 1D and 2D site response benchmarks used for the CPU version. A parallel efficiency of more than 85% was measured using 16,384 V100 GPUs running AWP-GPU-DM with 10 yield surfaces on the Summit supercomputer at OLCF. The code will be used to carry out a hero-run San Andreas simulation with Iwan-type nonlinearity for frequencies up to 5 Hz.