Paperlane

no. 17 · Wednesday 18 November 2026

High-Order Regularization for Singular and Nearly Singular Laplace and Stokes Layer Potentials in Two Dimensions

Han Zhou

math.NA · code · 36 pp

Abstract

Layer potentials represent the solutions of many elliptic boundary value problems, but their kernels have the singularity of the fundamental solution, so standard quadrature loses accuracy at targets on or near the boundary. In two dimensions, Gaussian regularization of the logarithmic kernel gives a smooth kernel involving the exponential integral $E_1$, but for the single layer its regularization error is only first order in the smoothing radius $δ$. We construct modified kernels for two-dimensional Laplace and Stokes single- and double-layer potentials that achieve regularization error $O(δ^p)$, $p=3,5,7$, uniformly for targets on and near the curve. The resulting method is both high-order accurate and easy to implement: the integrals are evaluated directly by the periodic trapezoidal rule on a single fixed set of boundary nodes for all targets. Choosing $δ=c h^q$ with $q<1$, where $h$ is the node spacing, keeps the discretization error smaller than the regularization error as $h\to0$, and numerical experiments confirm the predicted $O(h^{pq})$ convergence. Finally, we combine the near-curve evaluation with a fast Poisson solver to compute the potentials on the entire Cartesian grid.

code
github.com/han-zhou-math/high-order-reg-2d
comments
36 pages, 11 figures; code and data available at https://github.com/han-zhou-math/high-order-reg-2d
fields
Numerical Analysis
pdf ↗ page arXiv:2611.21991 ↗