The role of multi-method linear solvers in PDE-based simulations

Sanjukta Bhowmick, L. McInnes, B. Norris, P. Raghavan

Research output: Contribution to journalArticle

14 Citations (Scopus)

Abstract

The solution of large-scale, nonlinear PDE-based simulations typically depends on the performance of sparse linear solvers, which may be invoked at each nonlinear iteration. We present a framework for using multi-method solvers in such simulations to potentially improve the execution time and reliability of linear system solution. We consider composite solvers, which provide reliable linear solution by using a sequence of preconditioned iterative methods on a given system until convergence is achieved. We also consider adaptive solvers, where the solution method is selected dynamically to match the attributes of linear systems as they change during the course of the nonlinear iterations. We demonstrate how such multi-method composite and adaptive solvers can be developed using advanced software architectures such as PETSc, and we report on their performance in a computational fluid dynamics application.

Original languageEnglish (US)
Pages (from-to)828-839
Number of pages12
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume2667
StatePublished - Dec 1 2003

Fingerprint

Linear systems
Composite materials
Software architecture
Iterative methods
Linear Systems
Composite
Preconditioned Iterative Methods
Iteration
Computational fluid dynamics
Simulation
Nonlinear PDE
Software Architecture
Computational Fluid Dynamics
Execution Time
Hydrodynamics
Attribute
Software
Demonstrate
Framework

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

@article{f7b437ae350b4e47a456b8a5ab0654ea,
title = "The role of multi-method linear solvers in PDE-based simulations",
abstract = "The solution of large-scale, nonlinear PDE-based simulations typically depends on the performance of sparse linear solvers, which may be invoked at each nonlinear iteration. We present a framework for using multi-method solvers in such simulations to potentially improve the execution time and reliability of linear system solution. We consider composite solvers, which provide reliable linear solution by using a sequence of preconditioned iterative methods on a given system until convergence is achieved. We also consider adaptive solvers, where the solution method is selected dynamically to match the attributes of linear systems as they change during the course of the nonlinear iterations. We demonstrate how such multi-method composite and adaptive solvers can be developed using advanced software architectures such as PETSc, and we report on their performance in a computational fluid dynamics application.",
author = "Sanjukta Bhowmick and L. McInnes and B. Norris and P. Raghavan",
year = "2003",
month = "12",
day = "1",
language = "English (US)",
volume = "2667",
pages = "828--839",
journal = "Lecture Notes in Computer Science",
issn = "0302-9743",
publisher = "Springer Verlag",

}

TY - JOUR

T1 - The role of multi-method linear solvers in PDE-based simulations

AU - Bhowmick, Sanjukta

AU - McInnes, L.

AU - Norris, B.

AU - Raghavan, P.

PY - 2003/12/1

Y1 - 2003/12/1

N2 - The solution of large-scale, nonlinear PDE-based simulations typically depends on the performance of sparse linear solvers, which may be invoked at each nonlinear iteration. We present a framework for using multi-method solvers in such simulations to potentially improve the execution time and reliability of linear system solution. We consider composite solvers, which provide reliable linear solution by using a sequence of preconditioned iterative methods on a given system until convergence is achieved. We also consider adaptive solvers, where the solution method is selected dynamically to match the attributes of linear systems as they change during the course of the nonlinear iterations. We demonstrate how such multi-method composite and adaptive solvers can be developed using advanced software architectures such as PETSc, and we report on their performance in a computational fluid dynamics application.

AB - The solution of large-scale, nonlinear PDE-based simulations typically depends on the performance of sparse linear solvers, which may be invoked at each nonlinear iteration. We present a framework for using multi-method solvers in such simulations to potentially improve the execution time and reliability of linear system solution. We consider composite solvers, which provide reliable linear solution by using a sequence of preconditioned iterative methods on a given system until convergence is achieved. We also consider adaptive solvers, where the solution method is selected dynamically to match the attributes of linear systems as they change during the course of the nonlinear iterations. We demonstrate how such multi-method composite and adaptive solvers can be developed using advanced software architectures such as PETSc, and we report on their performance in a computational fluid dynamics application.

UR - http://www.scopus.com/inward/record.url?scp=35248819657&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=35248819657&partnerID=8YFLogxK

M3 - Article

VL - 2667

SP - 828

EP - 839

JO - Lecture Notes in Computer Science

JF - Lecture Notes in Computer Science

SN - 0302-9743

ER -