"Analysis of a Pivoting Strategy for Gaussian Elimination in Parallel"
ABSTRACT
A pivoting strategy was introduced by Onaga and Takechi to improve
Gaussian elimination with partial pivoting on parallel machines. This
method is cheaper on parallel
machines becuase one row transport is eliminated per pivoting step when
rows are added instead of pivoted. It is shown that this pivoting
strategy results in growth bounded
above by 3^(N-1) rather than 2^(N-1) in the usual case. Growth factors
near this bound are found for near singular matrices where the growth
is small with the usual pivoting
strategy. Numerical results verify this algorithm mail fail when
partial pivoting is stable even though in the average case the two
methods are comparable.
|
Home|