Ultra-Scalable Algorithms for Large-Scale Uncertainty Quantification in Inverse Wave Propagation
Technical Report,30 Sep 2012,30 Nov 2015
Institute for Computational Engineering and Sciences Austin United States
Pagination or Media Count:
The overall aim of this project was to develop scalable algorithms for the inverse problem of inferring, with associated uncertainty, the heterogeneity of a medium or shape of a scatterer from reflectedtransmitted waves acoustic, elastic, electromagnetic at very large scale. The resulting Bayesian wave inverse propagation problem has been intractable using contemporary algorithms. Research was conducted under three complementary subprojects. The first subproject led by O. Ghattas focused on scalable algorithms for large-scale Bayesian inverse problems governed by time domain wave propagation. The second subproject led by G. Biros focused on fast algorithms for inverse scattering and uncertainty quantification based on volume integral equation formulations for the inverse medium problem. The third subproject ledby L. Demkowicz and J. Gopalakrishnan focused on new, highly efficient discretizations for wave propagation in the form of the discontinuous Petrov Galerkin DPG method and associated solvers. Results and conclusions in each sub-project area are discussed in separate sections of the report.