[Paper Review] Variants of an explicit kernel-split panel-based Nyström discretization scheme for Helmholtz boundary value problems

Johan Helsing, Anders G. Holst|arXiv (Cornell University)|Nov 25, 2013

Electromagnetic Scattering and Analysis16 references4 citations

TL;DR

This paper presents and evaluates explicit kernel-split panel-based Nyström schemes for solving planar Helmholtz boundary value problems, leveraging kernel splitting, high-order polynomial interpolation, on-the-fly product integration, and adaptive coarse/fine grids to achieve rapid and stable convergence in both near and far fields. The key contribution is demonstrating that such schemes are competitive with state-of-the-art methods, particularly in accuracy and convergence speed, when properly implemented with advanced numerical techniques.

ABSTRACT

The incorporation of analytical kernel information is exploited in the construction of Nyström discretization schemes for integral equations modeling planar Helmholtz boundary value problems. Splittings of kernels and matrices, coarse and fine grids, high-order polynomial interpolation, product integration performed on the fly, and iterative solution are some of the numerical techniques used to seek rapid and stable convergence of computed fields in the entire computational domain.

Motivation & Objective

• To investigate the performance of explicit kernel-split panel-based Nyström discretization schemes for planar Helmholtz boundary value problems.
• To compare these schemes with split-free panel-based schemes in terms of convergence speed and achievable accuracy in near and far fields.
• To develop and demonstrate numerical tools—such as on-the-fly product integration and mixed coarse/fine mesh strategies—that enhance scheme performance beyond naive implementations.
• To show that kernel-splitting, despite its apparent complexity, leads to simple and implementable formulations when combined with efficient quadrature techniques.

Proposed method

• The scheme uses an explicit splitting of the singular kernels in the combined field integral equation for the Helmholtz problem into smooth and singular parts, enabling higher accuracy.
• High-order polynomial interpolation is applied on panels to achieve spectral convergence, with quadrature nodes and weights derived from Gauss-Legendre rules.
• On-the-fly product integration is used to compute corrections for logarithmic and Cauchy-type singularities arising when evaluating fields near or on the boundary.
• The method employs a combination of coarse and fine meshes to improve efficiency and accuracy, particularly in regions with high gradients or near singularities.
• Iterative solvers are used to solve the resulting linear systems, with convergence monitored across both near-field and far-field points.
• Specialized MATLAB routines are provided for computing weight corrections and compensation weights for singular integrals, enabling accurate evaluation of the solution in the entire domain.

Experimental results

Research questions

• RQ1 How does the convergence rate of explicit kernel-split panel-based Nyström schemes compare to split-free schemes for the high-frequency exterior Helmholtz Dirichlet problem?
• RQ2 Can the use of on-the-fly product integration and adaptive meshing significantly improve accuracy and stability in near-field and far-field evaluations?
• RQ3 What is the impact of combining coarse and fine grids on the performance of kernel-split Nyström schemes in terms of convergence speed and solution accuracy?
• RQ4 To what extent do kernel-splitting techniques simplify implementation when analytical kernel information is used, despite the apparent complexity of Hankel function expansions?
• RQ5 How do the spectral properties and iterative convergence behavior of the linear systems differ between kernel-split and non-split schemes?

Key findings

• The explicit kernel-split panel-based Nyström scheme achieves rapid and stable convergence for both the boundary density and the solution field in the entire computational domain, including near and far fields.
• The scheme outperforms or matches the performance of all competitive schemes listed in Table 2 of reference [6], particularly in terms of accuracy and convergence speed.
• On-the-fly product integration enables accurate evaluation of singular integrals without precomputation, significantly improving robustness and efficiency.
• The use of mixed coarse and fine meshes enhances performance, especially in regions with high-frequency oscillations or boundary layer effects.
• The implementation is straightforward once kernel splits are derived, with the final expressions involving only functions also needed in split-free schemes.
• The method is robust for high wave numbers and maintains accuracy even when evaluating the solution close to the boundary, where standard schemes often fail.