A Fast Solver for a Nonsmooth Optimization problem in a phase separation model
14:00 - 16:00
Dr. Pawan Kumar (FU Berlin)
Großer Seminarraum "Prigogine"
A Fast Solver for a Nonsmooth Optimization problem in a phase separation model Short CVThe
speaker has spent roughly 8.5 years in Europe working on research
projects in various institutes. Currently, he is doing postdoc in the
department of mathematics and computer science at FU, Berlin. Before
that, he had two postdoctoral positions at Fraunhofer ITWM and KU
Leuven. He has his PhD from INRIA, Saclay and a master degree in
mathematics and computing from IIT, Guwahati. His research interests
include various aspects of PDE solvers such as developing new iterative
solvers, their convergence analysis, and their parallel implementation.Abstract Phase separation phenomena are now typically modelled by Cahn-Hilliard equations. To model the observed deep quench event, a nonsmooth obstacle potential have been proposed. Recently, a Newton-Schur method was proposed that may be viewed as Uzawa iteration with a flavour of active set strategy. Solution of obstacle problem identifies the active sets, then the problem reduces to a linear problem on nonactive set. This corresponds to solving a saddle point problem but on truncated domains.In this talk, the speaker will discuss efficient preconditioning for solving linear saddle point problems on truncated domains. In particular, some optimal existing saddle point solvers will be adapted on truncated domains. Spectrum analysis and numerical experiments will show the robustness of these solvers for various evolutions.
 P. Kumar, Fast Preconditioned Solver for Truncated Saddle Point Problem in Nonsmooth Cahn-Hilliard Model, Studies in Computational Intelligence, Book Chapter, to appear, 2016
 P. Kumar, Preconditioners based on approximation of nonstandard norms for phase separation applications, ICNAAM 2015, Greece, AIP proceedings, to appear, 2016