An accelerated direct solver for scalar wave scattering by multiple transmissive inclusions in two dimensions

TL;DR

This paper introduces a fast direct solver for 2D scalar wave scattering by multiple inclusions, leveraging low-rank approximations to significantly reduce computational costs in transmission problems.

Contribution

The proposed solver efficiently handles multiple disjoint scatterers using proxy-based low-rank approximations, especially for the PMCHWT formulation, offering speed and system size reductions.

Findings

The solver compresses the linear system to size O(ωD).

Computational cost scales as O(N^{1.5}) for fixed frequency.

PMCHWT formulation is six times faster than Burton--Miller in tests.

Abstract

This paper discusses a fast direct solver using boundary integral equations for Helmholtz transmission problems involving multiple inclusions in two dimensions. Efficiently addressing scattering problems in the presence of numerous inclusions remains a key challenge for various practical applications. For problems involving a large number of scatterers, the number of iterations in Krylov subspace methods is known to increase significantly. This occurs even when using second-kind boundary integral equations, which are typically recognized for their rapid convergence. We consider a fast direct solver as an alternative, an approach that has been less commonly explored for transmission problems with disjoint multiple inclusions. The low-rank approximation based on the proxy method achieve speedup by calculating interactions between disjoint scatterers without the terms derived from the internal integral representation. Notably, this advantage applies to the Poggio--Miller--Chang--Harrington--Wu--Tsai (PMCHWT) formulation but breaks down in the Burton--Miller case. Numerical examples demonstrate that the proposed solver can compress the system of linear algebraic equations to a size of $O(\omega D)$, where $\omega$ is the frequency of the incident wave and $D$ is the diameter of the (smallest) bounding box enclosing the multiple inclusions. The total computational cost scales as $O(N^{1.5})$ $(= O(\sqrt{N}^3))$ at most for a fixed $\omega$ when the inclusions are arranged on a grid. Moreover, the PMCHWT formulation, that omits the interior term in the proxy method, is approximately six times faster than the Burton--Miller formulation when treating each inclusion as a cell. Furthermore, in the same setting, the former can compress the size of the system of linear algebraic equations by half compared to the latter.