ooreilly's picture

Ossian O'Reilly

Southern California Earthquake Center
University of Southern California
Postdoctoral Scholar

Expertise: Scientific Computing, Numerical Methods, Numerical Analysis, High-performance computing 
About Me Publications
Ossian O'Reilly contributes to the SCEC Cybershake and High F projects, developing software to enable large-scale ground motion simulations with topography. He leads the numerical methods development to derive and construct a method that is stable, accurate, and useful in practice (that is, capable of running at extremely large-scale for varied model geometries, scales well, and is not prohibitively computationally expensive). The work is rigorously verified using manufactured solutions against benchmarks from other codes to ensure accuracy, stability. These benchmark studies are conducted with collaborators at UCSD and SDSU. He is also responsible for implementing the numerical scheme in the in the most recent version of the wave propagation code AWP-ODC, developed and maintained by UCSD researchers. 

To accomplish this task, O'Reilly applies skills he acquired while developing high order numerical methods for wave propagation throughout his PhD (for example, simulation of waves in solids interacting with faults and fluid-filled fractures). In these past works, complex geometry enters in the form of curved or branching faults that are have been meshed using both unstructured grids and combination structured grids and by curvilinear grid mappings. Prototype codes for exploring and verifying the numerical methods for these projects, and other ones, have mainly been developed in MATLAB/Octave, or Python for rapid prototyping. Production codes have been developed in either C, C++, or Fortran and been parallelized using MPI and GPUs.  

O'Reilly contributed to early developments of the SCEC special project “Mining Seismic Wavefields” spearheaded by Greg Beroza (Stanford). He evaluated algorithms used in "FAST" (Fingerprinting And Similarity Thresholding, https://github.com/stanford-futuredata/FAST) and laid the foundation for a C++ prototype implementation.

He is also involved with automating mesh generation for the SCEC Community Fault Model (https://www.scec.org/research/cfm). The approach under consideration is to use spline fitting techniques to construct BSpline surfaces, which will then be meshed by industry standard mesh generators.