Direct sparse factorization of blocked saddle point matrices

Authors:
Claude Lacoursière;Mattias Linde;Olof Sabelström
Affiliations:
HPC2N/UMIT, Umeå University, Sweden;Algoryx Simulation AB, Umeå, Sweden;Department of Computing Science, Umeå University, Sweden
Venue:
PARA'10 Proceedings of the 10th international conference on Applied Parallel and Scientific Computing - Volume 2
Year:
2010

Citing 10
Cited 0

Introduction to algorithms

Introduction to algorithms
Matrix computations (3rd ed.)

Matrix computations (3rd ed.)
Highly Scalable Parallel Algorithms for Sparse Matrix Factorization

IEEE Transactions on Parallel and Distributed Systems
A Fast and High Quality Multilevel Scheme for Partitioning Irregular Graphs

SIAM Journal on Scientific Computing
Algorithm 837: AMD, an approximate minimum degree ordering algorithm

ACM Transactions on Mathematical Software (TOMS)
A numerical evaluation of sparse direct solvers for the solution of large sparse symmetric linear systems of equations

ACM Transactions on Mathematical Software (TOMS)
Matching-based preprocessing algorithms to the solution of saddle-point problems in large-scale nonconvex interior-point optimization

Computational Optimization and Applications
Towards Stable Mixed Pivoting Strategies for the Sequential and Parallel Solution of Sparse Symmetric Indefinite Systems

SIAM Journal on Matrix Analysis and Applications
Hybrid, Multiresolution Wires with Massless Frictional Contacts

IEEE Transactions on Visualization and Computer Graphics
Constraint Fluids

IEEE Transactions on Visualization and Computer Graphics

Quantified Score

Hi-index	0.00

Visualization

Abstract

We present a parallel algorithm for the direct factorization of sparse saddle-point matrices of moderate size coming from real-time multibody dynamics simulations. We used the specific structure of these problems both for a priori construction of supernodes and to avoid all dynamic permutations during factorization. For the latter, we present a technique we call "leaf swapping" which performs permutations of the supernodes in the elimination tree without any reference to numerical values. The results compare favorably with currently available high performance codes on our problem sets because of the high overhead necessary to process very large problems on increasingly complex supercomputers.