Preconditioning the Conjugate Gradient Method for a Hybridized Summation-By-Parts Finite-Difference Discretization of Poisson's Equation with Application to Seismic Simulations
Abstract
We consider the solution by the conjugate gradient (CG) method of the linear systems arising from a hybridized summation-by-parts finite-difference discretization of a Poisson problem that occurs in the simulation of seismic faults. We study the efficacy of three different preconditioning schemes-- Jacobi, incomplete Cholesky, and algebraic multigrid-- at reducing the number of CG iterations required for convergence, finding that while all three preconditioners are effective, the latter two are superior. Code for all of our studies, written in the Julia programming language, is provided.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2022
- Accession Number
- AD1184946
Entities
People
- Timothy P. James
Organizations
- Naval Postgraduate School