Multigrid preconditioning on Xeon Phi

Matthias Bolten, University of Kassel Institute of Mathematics
Matthias K. Gobbert, Department of Mathematics and Statistics, UMBC

In many applications, including simulations in engineering or life sciences, the solution of partial differential equations is needed. Often this requires the solution of a linear system, e.g., when a steady state solution is demanded or an implicit method is used. Multigrid methods are known to be optimal in many cases.

On parallel computers and modern computer architectures the use of structure is important in order to use the available processing power as efficient as possible. This can be achieved by using structured discretizations, that are also favorable for geometric multigrid methods.

In this project we want to evaluate an existing parallel multigrid solver on the computing ressources at UMBC and assess the applicability in numerical simulation of calcium waves in the heart cell.