|
General Description Most forms of numerical analysis in science and engineering produce a sparse set of linear system of equations. Solving a large linear system of equations requires an efficient and robust linear equation solver. Direct methods tend to be more robust than iterative methods. Direct sparse solvers are the solution for solving large scale linear system of equations.Specifications: PVSS is a general parallel sparse linear equation solver. It is a robust and efficient sparse solver. It uses minimum memory when solving the linear system of equations. Memory is dynamically allocated within the solver, thus the user does not have to worry about allocation of memory . It is packaged so that it is easy to use and implement within any code. Users and developers can use it as a black box. It consist of 4 different sections:A- Matrix Reordering Reordering is included with PVSS to minimize the fill. At least 2 reordering algorithms are use and the best is chosen. In order to minimize reordering time and storage, the reordering is done on a smaller size matrix.B- Factorization: Factorization of the system of equations is done is an efficient way where the equations are blocked in order to reduce the factorization time and minimize indirect addressing.C- Forward/Backward substitution: After factorization is done forward/backward substitution can be called as many times as required.D- Free Memory: A subroutine is given to free all the memory allocated by PVSS Solver. For nonlinear analysis, where it is known that the matrix does not change in geometry, the reordering is saved and used over and over - which saves time. PVSS has been implemented on many computer platforms such as: IBM, SGI, CRAY, SUN, HP, DEC. Limitations: PVSS can solve symmetric definite linear system of equations using real and complex arithmetic.Future release Work is underway to develop new reordering techniques for the solver. |