Efficient in-score sparse direct solution of large finite element problems

0430422 - ÚT 2015 RIV CZ eng A - Abstract
Pařík, Petr - Plešek, Jiří
Efficient in-score sparse direct solution of large finite element problems.
Modelling 2014. Ostrava: Ústav geoniky AV ČR, 2014 - (Blaheta, R.; Starý, J.; Sysalová, D.). s. 66-67. ISBN 978-80-86407-47-0.
[Modelling 2014. 02.06.2014-06.06.2014, Rožnov pod Radhoštěm]
Institutional support: RVO:61388998
Keywords : finite element method * sparse direct solution
Subject RIV: JC - Computer Hardware ; Software

Sparse direct solvers represent one of the possible approaches to the solution of large problems in FEM. Our approach focuses on minimizing the amount of storage space and computational time required for the solution of a large problem. This is achieved particularly by using a modified minimum ordering algorithm to reduce the fill-in, therefore, reducing the number of numerical operations needed to compute the solution. Both the factorization and the backsubstitution algorithms are parallelized using the OpenMP library to fully exploit today's multi-core and multi-processor computers. The global stiffness matrix is assembled and stored in-core throughout the computation, using an efficient sparse matrix storage format. We present some details of our sparse direct solver implementation, realized within the framework of our in-house finite element code PMD. We also show the performance results for several large real-world problems taken from the engineering practice and assess the scalability of the solver.
Permanent Link: http://hdl.handle.net/11104/0239235