Realization of State-Space Models for Wave Propagation Simulations

• Main Authors: Ketcham, Stephen A, Phan, Minh Q, Darling, Richard S, McKenna, Mihan H
• Format: Report
• Language: English
• Subjects: ALGORITHMS; HIGH PERFORMANCE COMPUTING; Radiofrequency Wave Propagation; SCATTERING; SEISMOLOGY; TIME SERIES ANALYSIS; WAVE PROPAGATION

Fully three-dimensional numerical solutions can quantify exterior seismic or acoustic propagation throughout complex geologic or atmospheric domains. Results from impulsive sources typically reveal propagating waves plus reverberations typical of multi-path scattering and wave-guide behavior, with decay toward quiescent motions as the dominant wave energy moves out of the domain. Because such computations are expensive and yield large data sets, it is advantageous to make the data reusable and reducible for both direct and reciprocal simulations. Our objective is efficient time-domain simulation of the wave-field response to sources with arbitrary time series. For this purpose we developed a practical and robust technique for superstable model identification. A superstable model has the form of a state-space model, but the output matrix contains the system dynamics. It simulates propagation with the fidelity of the pulse response calculated for the numerical system. Our development of the superstable technique was motivated by our initial application of the Eigensystem Realization Algorithm to wave-field systems from high-performance-computing analyses, where we recognized exterior propagation features allowing superstable model assignment. Most importantly the pulse response and its decay over the domain are captured in a finite duration, and decay to zero beyond a finite number of time steps implies a system with zero eigenvalues. We demonstrate propagation system identification with pulse response data derived from supercomputer analysis, and conclude that, using superstable-identified systems, we are able to create reusable and reducible propagation-system models that accurately simulate the wave field using a fraction of the original computational resources. The original document contains color images.