On reducing communication and improving adaptivity in nonlinear multigrid methods
Jed Brown
Department of Computer Science, University of Colorado Boulder
Date and time:
Friday, October 3, 2014 - 3:00pm
Location:
ECCR 245
Abstract:
FAS multigrid exposes structure that can be exploited to reduce communication and memory motion while retaining optimal convergence as well as to adaptively solve problems with localized nonlinearities in heterogeneous media. Inspired by an oft-overlooked observation made by Achi Brandt more than 3 decades ago, we present an updated interpretation of so-called tau corrections representing the influence of fine scales on coarser scales, new computational results, and a prospectus for modern applications and computing architectures.