next up previous
Next: About this document ...

Tristan Konolige
Improved Adaptive Smoothed Aggregation

College of Engineering and Applied Science
University of Colorado Boulder
1111 Engineering Drive
ECOT 717
430 UCB
Boulder
CO 80309-0430 USA
tristan.konolige@colorado.edu
Ben Southworth
Delyan Kalchev
Stephen McCormick
Thomas Manteuffel

Smoothed Aggregation (SA) multigrid has proved to be a useful AMG method for solving discretizations of elliptic problems. However, SA requires knowledge of slow to converge near-nullspace vectors in order to achieve good performance. Previous work in adaptively finding near-nullspace vectors relied on heuristics and had trouble when near-nullspace vectors were too similar. Our improved version of Adaptive Smoothed Aggregation (aSA) uses a more principled approach to find near-nullspace vectors. By leveraging the Weak Approximation Property and global and local orthogonalization processes, we ensure that our candidates better match the near-nullspace vectors. We extend this method to find candidates on all multigrid levels. Our technique can be used as a tool to find near-nullspace vectors for developing new multigrid methods or as a stand alone solver for difficult problems.





root 2016-02-22