AAA least squares solution of Helmholtz problems

Technical Summary: AAA Least Squares Solution of Helmholtz Problems

Problem Statement
The paper addresses the numerical solution of exterior "sound-soft" scattering problems governed by the Helmholtz equation ($\Delta u + k^2 u = 0$). While the Method of Fundamental Solutions (MFS) is a popular meshless technique due to its conceptual simplicity, it suffers from significant robustness and accuracy issues, particularly regarding the placement of source singularities. In classical MFS, singularities are placed manually based on heuristics, which often leads to ill-conditioned linear systems and poor performance on complex, non-star-shaped geometries or at high wavenumbers ($k \gtrsim 20$). The core challenge is that the numerical stability of MFS is intrinsically linked to the analytic continuation of the solution; singularities must be placed outside the domain but close enough to the boundary to ensure rapid convergence, a task that is difficult to automate for general domains.

Methodology
The proposed AAALS-Helmholtz algorithm reframes the MFS through the lens of rational approximation, specifically utilizing the Adaptive Antoulas-Anderson (AAA) least squares method. The methodology operates on the principle that the optimal placement of MFS sources is dictated by the singularities limiting the analytic continuation of the boundary data.

The algorithm proceeds through the following steps:

1. Continuum AAA Approximation: Instead of using static discretizations, the paper employs a "continuum" variant of the AAA algorithm (aaazp) capable of handling generic curves in the complex plane. This adaptive strategy dynamically populates sample points in regions of high error or rapid variation, ensuring sufficient resolution without manual grid refinement.
2. Conformal Mapping and Singularity Identification: The algorithm performs rational approximation on the unit circle ($\mathbb{C}$) in a computational $\zeta$-plane rather than directly on the physical boundary $\Gamma$. It approximates the mapping $M$ from the unit circle to the physical boundary and identifies the zeros of the derivative $M'$ (which correspond to branch points of the Schwarz function). Simultaneously, it approximates the incident wave data to identify poles limiting analytic extension.
3. Source Placement Strategy:
   • A "shielding curve" $\gamma$ is constructed within the conformality region, bounded by the identified zeros of $M'$ and the poles of the rational approximant.
   • Source points $\{\pi_j\}$ are placed at an optimal radial distance between $\gamma$ and the unit circle, corresponding to the support points of the rational approximation.
   • These points are mapped back to the physical domain via $M$ to generate the final source distribution $\{p_j\}$. This results in a singularity-adapted distribution that clusters naturally near geometric features (like corners) and follows the branch structure of the Schwarz function.
4. Least Squares Solution: The scattered wave is computed using a multipole expansion involving Hankel functions of the first kind. The coefficients are determined by solving an overdetermined linear system via least squares. To improve conditioning, matrix columns are scaled by the distance of sources from the boundary, and low-degree Hankel series (up to order $R=2$) are included to enhance accuracy without ill-conditioning.

Key Contributions

• Automated Singularity Placement: The paper introduces a fully automated strategy for placing MFS sources based on the analytic continuation of boundary data, eliminating the need for heuristic or manual placement.
• Continuum AAA for Generic Curves: The work extends the AAA algorithm to generic curves in the complex plane (aaazp), enabling the adaptive resolution of singularities near corners and complex geometries.
• Theoretical Connection to Laplace Problems: The authors establish a formal link between the Helmholtz and Laplace problems. They demonstrate that as the wavenumber $k \to 0$, the source placement strategy for Helmholtz naturally reduces to the optimal "double poles" technique used in Laplace problems, providing a theoretical justification for this empirical observation.
• Robustness at High Wavenumbers: The method is shown to be effective for high wavenumbers ($k$ up to 150 and beyond) and complex, non-star-shaped domains where traditional MFS often fails.

Results
Numerical experiments demonstrate that the AAALS-Helmholtz algorithm achieves high accuracy (typically 6 or more digits, often reaching $10^{-10}$ to $10^{-14}$) on a variety of test cases, including starfish-shaped domains, generalized crescents, and "sharktooth" curves with corner singularities.

• Comparison with MPSpack: In a validation case involving a trefoil domain with $k=30$, the proposed method achieved an error of $7.5 \times 10^{-11}$ using 80 sources and 720 samples in a fraction of a second. This is comparable to the MPSpack toolbox, which required 210 basis functions and 250 quadrature points to achieve an error of $8 \times 10^{-12}$.
• Handling of Singularities: The algorithm successfully handles corner singularities, though accuracy on extreme geometries (e.g., "sharktooth") is limited to approximately $10^{-6}$, attributed to the sampling limitations of the aaazp routine near sharp corners.
• Laplace Limit: For Laplace problems, the method confirms that doubling the poles (using the support points as collocation points and their images as charges) yields high accuracy, particularly for geometry-driven problems where the Schwarz function dictates the branch structure.

Significance
The paper claims to provide the first comprehensive experimental treatment of an adaptive AAALS-Helmholtz method for scattering problems with high wave numbers. Its primary significance lies in bridging the gap between rational approximation theory and the Method of Fundamental Solutions, offering a robust, meshless alternative to heuristic source placement. By interpreting MFS through rational approximation, the work validates that automated singularity placement, driven by the analytic properties of the boundary data, vastly outperforms traditional methods. Furthermore, the theoretical unification of Helmholtz and Laplace source placement strategies offers a deeper understanding of why techniques like "double poles" work, moving the field from empirical observation to theoretical justification. The authors conclude that this approach represents a promising, theoretically grounded tool for engineering computations in acoustics and scattering theory.