Abstract
We discuss preconditioning strategies for solving large Electric Field Integral Equation systems. We consider several algebraic preconditioners for solving the dense linear system arising from the Galerkin discretization of the pertinent integral equation. We show that approximate inverse methods based on Frobenius-norm minimization techniques can be very effective to reduce the number of iterations of Krylov subspace solvers for this problem class. We describe the implementation of the preconditioner within the Fast Multipole Algorithm and we illustrate how to reduce the construction cost by using static pattern selection strategies. Finally, we present deflating techniques based on low-rank matrix updates to enhance the robustness of the approximate inverse on tough problems. Experiments are reported on the numerical behavior of the proposed method on a set of realistic industrial problems.