On preconditioners for the Laplace double-layer in 2D

TL;DR

This paper investigates preconditioning techniques for the Laplace double-layer operator in 2D, proposing a new FMM-based preconditioner to improve iterative solver efficiency for complex geometries.

Contribution

The paper introduces a novel FMM-based spatial decomposition preconditioner for the double-layer operator, enhancing solver performance for complicated geometries.

Findings

The new preconditioner improves convergence rates.

It compares favorably with existing methods in experiments.

The approach extends to other second-kind integral equations.

Abstract

The discretization of the double-layer potential integral equation for the interior Dirichlet Laplace problem in a domain with smooth boundary results in a linear system that has a bounded condition number. Thus, the number of iterations required for the convergence of a Krylov method is, asymptotically, independent of the discretization size $N$. Using the Fast Multipole Method (FMM) to accelerate the matrix-vector products, we obtain an optimal $\mathcal{O}(N)$ solver. In practice, however, when the geometry is complicated, the number of Krylov iterations can be quite large---to the extend that necessitates the use of preconditioning. We summarize the different methodologies that have appeared in the literature (single-grid, multigrid, approximate sparse inverses) and we propose a new class of preconditioners based on an FMM-based spatial decomposition of the double-layer operator. We present an experimental study in which we compare the different approaches and we discuss the merits and shortcomings of our approach. Our method can be easily extended to other second-kind integral equations with non-oscillatory kernels in two and three dimensions.