A parallel algorithm is proposed for the solution of narrow banded non-symm
etric linear systems. The linear system is partitioned into blocks of rows
with a small number of unknown ns common to multiple blocks. Our technique
yields a reduced system defined only on these common unknowns which can the
n be solved by a direct or iterative method. A projection based extension t
o this approach is also proposed for computing the reduced system implicitl
y, which gives rise to an inner-outer iteration method. In addition, the pr
oduct of a vector. with the reduced system matrix can be computed efficient
ly on a multiprocessor by concurrent projections onto subspaces of block ro
ws. Scalable implementations of the algorithm can be devized for hierarchic
al parallel architectures by exploiting the two-level parallelism inherent
in the method. Our experiments indicate that the proposed algorithm is a ro
bust and competitive alternative to existing methods, particularly for diff
icult problems with strong indefinite symmetric part. Copyright (C) 2001 Jo
hn Wiley & Sons, Ltd.