"A Different Pivoting Strategy in Gaussian
Elimination for Parallel Machines"
ABSTRACT
For most matrices Gaussian elimination with partial pivoting (GEPP) is an
accurate method to solve systems of equations. To improve GEPP on
parallel machines a new pivoting strategy was introduced by Onaga and
Takechi. This method, GEPPA, makes GEPP cheaper on parallel machines by
eliminating one row transport per pivoting step. Theoretically, it
appears as if GEPPA may produce worse results than GEPP. Results from
matrices formed from two-point boundary value problems show both GEPP
and GEPPA can produce bad results. Results from matrices where GEPP
produces good results and GEPPA produces bad results will also be shown.
However, for randomly generated matrices, GEPP and GEPPA are comparable.
|
Home|