Block ILU smoothers for p-multigrid methods in Isogeometric Analysis

Master thesis (2022)

Authors

M. Looije Electrical Engineering, Mathematics and Computer Science

Contributors

Cornelis Vuik Numerical Analysis - EEMCS (supervisor 1)
R.P.W.M. Tielen Numerical Analysis - EEMCS (supervisor 1)
H.X. Lin Mathematical Physics - EEMCS (supervisor 2)
Faculty
Electrical Engineering, Mathematics and Computer Science
Copyright: © 2022 Mark Looije
Isogeometric analysis P-multigrid methods ILU factorizations
More Info
expand_more

Published Date

18-03-2022

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Electrical Engineering, Mathematics and Computer Science

Abstract

Isogeometric Analysis (IgA) is an extension of the more well known Finite Element Method (FEM). It allows for more accurate descriptions of boundary value problems on irregular domains. However, many of the traditional iterative solution strategies that are known to work well in FEM do not show the same behavior in IgA, especially for increasing order of basis functions p.

A method shown to have fast convergence in this situation is a p-multigrid method with a smoother based on a Block ILUT factorization. Most of the blocks of this factorization are efficiently calculated. The same holds for the smoothing steps.

It is therefore our objective to make changes to the Block ILUT smoother. Inspiration is taken from methods where ILU factorizations are constructed using a fixed-point iteration. We combine these with the existing Block ILUT smoother.
This ultimately leads to two new proposed methods we will call Block Fixed-point ILU and Block ParILUT.

The existing methods as well as the newly suggested methods are tested and compared on computational costs of the factorization, the number of nonzero entries in this factorization and the number of multigrid iterations needed
to reach convergence, if these factorizations are to be used as smoother. The benchmark used for these tests is a convection diffusion reaction (CDR) equation on a multipatch geometry with 4, 16 or 64 patches.