The International Conference for High Performance Computing, Networking, Storage and Analysis
Scalable, Adaptive Methods for Forward and Inverse Modeling of Continental-Scale Ice Sheet Flow.
Student: Tobin Isaac (University of Texas at Austin)
Advisor: Omar Ghattas (University of Texas at Austin)
Abstract: Projecting sea-level rise is made difficult by the complexity of accurately
modeling ice sheet dynamics for the polar ice sheets and the uncertainty in
key, unobservable parameter fields; my research addresses the inference of the
basal friction field beneath the Antarctic ice sheet. I develop scalable
algorithms and numerical methods that make tractable the calculation of a
friction field with quantified uncertainties. These contributions fall in the
categories of adaptive mesh refinement (AMR), efficient solvers for nonlinear
PDEs, and Bayesian statistical inversion, all with an emphasis on scalability
and high performance computing. I have developed algorithms for octree-based
AMR that have scaled well to 458K processes on ~30K BG/Q nodes. I have
developed a solver for high-order discretizations of the nonlinear Stokes
equations of ice sheet dynamics that scales to ~600M dofs. I am developing
Hessian-approximation techniques for Bayesian inference for problems whose
parameter-to-observable map requires solving systems of PDEs.
