Toward an Effective Sparse Approximate Inverse Preconditioner

From MaRDI portal





The author considers several techniques to improve the performance of so-called sparse approximate inverse preconditioners (SAIPs) \(M\) for matrices \(A\) derived from the finite difference, or the finite element discretization of partial differential equations. The SAIP matrix \(M\) should directly approximate \(A^{-1}\) in such a way that on the one hand the construction of \(M\) is not expansive, and on the other hand the preconditioning operation \(M*d\) (matrix \(\times\) vector) is cheap (i.e. \(M\) should be sparse) and the condition number of the preconditioned matrix \(MA\) is small. Since the \(i\)-th column of \(M\) should approximate the corresponding discrete Green's functions, one should incorporate some global coupling into the SAIP \(M\). The author proposes some refinement techniques that aim at a substantial improvement of the SAIPs for a wide class of problems. The numerical results given in the paper confirm these expected improvements.




Cited in
(36)


Describes a project that uses

Uses Software






This page was built for publication: Toward an Effective Sparse Approximate Inverse Preconditioner

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4702239)