Domain decomposition for parallel row projection algorithms (Q1181532): Difference between revisions

It is shown that for nonsymmetric matrices arising from finite differences or finite elements applied to partial differential equations a domain decomposition approach gives row projection methods and allows parallelism in computing orthogonal projections \(P_ iv=A_ i(A^ T_ iA_ i)^{-1}A^ T_ iv\). (\(A\) is partitioned into block rows \(A^ T_ i\)). Criteria for suitable row partitioning are discussed. Suitable row partitioning allows parallelism in the computation, creates subproblems that can be solved efficiently and gives numerically well behaved subproblems. The domain decomposition approach provides row partitionings that satisfy all of the criteria. Four test problems involving second order elliptic partial differential equations are computed and the results tabulated.