Close evaluation of layer potentials in three dimensions

TL;DR

This paper introduces a spherical coordinate-based quadrature method for accurately evaluating layer potentials near boundaries in 3D Laplace problems, achieving quadratic error rates with improved accuracy over existing techniques.

Contribution

The authors develop a novel quadrature approach using spherical coordinates and specific rules to evaluate layer potentials close to boundaries with high accuracy.

Findings

Quadratic error in double-layer potential evaluation near the boundary.

Linear error in single-layer potential evaluation, improved to quadratic with perturbation expansion.

Method aligns with asymptotic behavior of layer potentials at close distances.

Abstract

We present a simple and effective method for evaluating double-and single-layer potentials for Laplace's equation in three dimensions close to the boundary. The close evaluation of these layer potentials is challenging because they are nearly singular integrals. The method we propose is based on writing these layer potentials in spherical coordinates where the point at which their kernels are peaked maps to the north pole. An N-point Gauss-Legendre quadrature rule is used for integration with respect to the the polar angle rather than the cosine of the polar angle. A 2N-point periodic trapezoid rule is used to compute the integral with respect to the azimuthal angle which acts as a natural and effective averaging operation in this coordinate system. The numerical method resulting from combining these two quadrature rules in this rotated coordinate system yields results that are consistent with asymptotic behaviors of the double- and single-layer potentials at close evaluation distances. In particular, we show that the error in computing the double-layer potential, after applying a subtraction method, is quadratic with respect to the evaluation distance from the boundary, and the error is linear for the single-layer potential. We improve upon the single-layer potential by introducing an alternate approximation based on a perturbation expansion and obtain an error that is quadratic with respect to the evaluation distance from the boundary.