ISSN:
0271-2091
Keywords:
domain decomposition
;
line Gauss-Seidel
;
conjugate gradient
;
Engineering
;
Engineering General
Source:
Wiley InterScience Backfile Collection 1832-2000
Topics:
Mechanical Engineering, Materials Science, Production Engineering, Mining and Metallurgy, Traffic Engineering, Precision Mechanics
Notes:
The numerical discretization of the equations governing fluid flow results in coupled, quasi-linear and non-symmetric systems. Various approaches exist for resolving the non-linearity and couplings. During each non-linear iteration, nominally linear systems are solved for each of the flow variables. Line relaxation techniques are traditionally employed for solving these systems. However, they could be very expensive for realistic applications and present serious synchronization problems in a distributed memory parallel environment. In this paper the discrete linear systems are solved using the generalized conjugate gradient method of Concus and Golub. The performance of this algorithm is compared with the line Gauss-Seidel algorithm for laminar recirculatory flow in uni- and multiprocessor environments. The uniprocessor performances of these algorithms are also compared with that of a popular iterative solver for non-symmetric systems (the GMRES algorithm).
Additional Material:
8 Ill.
Type of Medium:
Electronic Resource
Permalink