Maximum norm error estimates for Neumann boundary value problems on graded meshes

Thomas Apel, Sergejs Rogovs, Johannes Pfefferer, Max Winkler

TL;DR

This paper provides the first sharp $L^\infty$ (maximum norm) error estimates for finite element solutions of Neumann boundary value problems on polygonal domains. By using locally refined "graded" meshes near domain corners, the authors prove that numerical solutions can achieve the optimal $h^2 |\ln h|$ convergence rate, extending these results to semilinear problems through a "supercloseness" argument.

What Problem Does This Paper Address?

When solving partial differential equations (PDEs) using the Finite Element Method (FEM), we usually assume the domain is smooth. However, real-world engineering problems often involve polygonal shapes with sharp corners. These corners create "singularities"—points where the solution's derivatives blow up—which significantly degrade the accuracy of standard numerical methods.

While the "Dirichlet" case (where values are fixed on the boundary) is well-documented, the "Neumann" case (where the derivative or "flux" is fixed on the boundary) is more technically challenging. Specifically, there was a gap in the literature regarding "pointwise" ($L^\infty$) error estimates on graded meshes for Neumann problems. Without these estimates, researchers cannot guarantee the maximum possible error in a simulation, which is critical for applications like optimal control or safety-critical engineering.

Methods

The authors employ several sophisticated mathematical techniques to bridge this gap:

  • Graded Meshes: Instead of a uniform grid, the mesh is systematically refined (made denser) as it approaches the corners. This "grading" compensates for the loss of regularity in the solution.
  • Weighted Sobolev Spaces: The authors analyze the problem in specialized function spaces where derivatives are weighted by their distance from the corners, allowing them to precisely capture the singular behavior.
  • Dyadic Decomposition: To prove the estimates, the authors partition the domain into a series of concentric "shells" (like the rings of a tree) surrounding the corners. They then apply local estimates to each shell and sum them up.
  • Supercloseness for Semilinear Problems: For nonlinear (semilinear) equations, they demonstrate that the finite element solution is "superclose" to the Ritz projection of the continuous solution. This allows them to transfer the high-quality linear estimates to the nonlinear regime.

Datasets

As a theoretical mathematics paper, it does not use external datasets. Instead, it utilizes a benchmark numerical example to verify the theory:

  • Domain: A circular sector defined by an interior angle $\omega$.
  • Scenarios: They test a convex domain ($\omega = 3\pi/4$) and a non-convex "L-shaped" domain ($\omega = 3\pi/2$).
  • Exact Solution: The problem is constructed such that the exact solution is known ($y = r^\lambda \cos(\lambda \phi)$), allowing for precise measurement of the numerical error.

Key Insights and Ideas

  • The Logarithmic Term: The authors show that the exponent of the logarithmic term in the error estimate ($|ln h|$) is exactly one. Previous attempts at similar problems often resulted in a sub-optimal $|ln h|^{3/2}$.
  • Neumann vs. Dirichlet: The proof for Neumann conditions is fundamentally different because standard tools like the Poincaré inequality do not apply directly, and the boundary data requires more careful handling in weighted spaces.
  • Linear to Nonlinear Transfer: The insight that semilinear problems can inherit the "sharp" convergence rates of linear problems via supercloseness is a powerful extension, making the results applicable to a wider range of physical phenomena (like heat radiation or chemical reactions).

Key Results

  • Optimal Convergence: The paper proves the error estimate $\|y - y_h\|_{L^\infty(\Omega)} \leq ch^2 |\ln h|$, where $h$ is the mesh size. This is the "Gold Standard" for piecewise linear finite elements.
  • Numerical Confirmation: In the L-shaped domain test ($\omega = 3\pi/2$):
  • On a uniform mesh, the convergence rate was a poor 0.67.
  • On a graded mesh ($\mu = 0.3$), the rate jumped to 1.96, nearly reaching the theoretical limit of 2.0.
  • Regularity Specification: The authors precisely define the required regularity of the input data ($f$ and $g$) needed to achieve these rates, which is essential for practitioners implementing these methods.

Limitations and Future Work

  • 2D Focus: The results are derived for two-dimensional polygonal domains. While the authors mention 3D polyhedral domains, the technical complexity of 3D edge and corner singularities remains a significant hurdle for future research.
  • Constant Dependence: The constant $c$ in the error estimates depends on the Hӧlder norms of the input data. In cases with very "rough" data, this constant could be large.
  • Application to Optimal Control: The authors suggest that these $L^\infty$ estimates are a prerequisite for proving the convergence of optimal control problems involving Neumann PDEs, which they intend to document in a forthcoming paper.

Summary updated

Abstract

Abstract This paper deals with a priori pointwise error estimates for the finite element solution of boundary value problems with Neumann boundary conditions in polygonal domains. Due to the corners of the domain, the convergence rate of the numerical solutions can be lower than in the case of smooth domains. As a remedy, the use of local mesh refinement near the corners is considered. In order to prove quasi-optimal a priori error estimates, regularity results in weighted Sobolev spaces are exploited. This is the first work on the Neumann boundary value problem where both the regularity of the data is exactly specified and the sharp convergence order $h^{2} \lvert \ln h \rvert $ in the case of piecewise linear finite element approximations is obtained. As an extension we show the same rate for the approximate solution of a semilinear boundary value problem. The proof relies in this case on the supercloseness between the Ritz projection to the continuous solution and the finite element solution.
BibTeX
@inproceedings{apel2018maximum,
  author = {Thomas Apel and Sergejs Rogovs and Johannes Pfefferer and Max Winkler},
  title = {{Maximum norm error estimates for Neumann boundary value problems on graded meshes}},
  booktitle = {IMA},
  year = {2018},
  doi = {10.1093/imanum/dry076},
}
Figure