A sparse approximate inverse preconditioner for parallel preconditioning of general sparse matrices (Q1855724)

The factored approximate inverse preconditioner of \textit{J.-C. Luo} [Comput. Math. Appl. 25, 113-122 (1993; Zbl 0778.65018); ibid. 25, No. 5, 83-90 (1993; Zbl 0778.65019); ibid. 25, No. 4, 95-104 (1993; Zbl 0776.65021); ibid. 25, No. 2, 73-79 (1993; Zbl 0765.65041)] is investigated and adapted for sparse parallel implementation. The exploitation of sparsity patters and threshold strategies for dropping small elements are discussed. Numerical experiments show that factored sparse approximate inverse methods are cheap in construction on sequential machines, unlike techniques based on norm minimization, but the inherent parallelism which is more prominent in these methods than in other incomplete LU factorizations is not easily exploited in general.