Iranian Journal of Numerical Analysis and Optimization

Iranian Journal of Numerical Analysis and Optimization

A stable and high-order accurate D-RBF-PU method via hybrid kernels

Document Type : Research Article

Authors
1 Faculty of Mathematical and Statistical Sciences, Malayer University, Malayer, Iran.
2 Department of Mathematics “Giuseppe Peano”, University of Torino, Torino, Italy.
Abstract
The Direct Radial Basis Function Partition of Unity (D-RBF-PU) method is a powerful mesh-free technique for solving Partial Differential Equations (PDEs). However, its performance is inherently constrained by a critical trade-off between accuracy and numerical stability. Finite-order smoothness kernels, such as Thin Plate Splines (TPS) and Mat{'e}rn, ensure well-conditioned systems but severely limit the convergence rate. Conversely, infinitely smooth kernels, such as Gaussians, yield high accuracy but suffer from severe ill-conditioning, leading to computational breakdown at finer discretizations. In this paper, we propose a novel approach using hybrid kernels, strategically combining Gaussian kernels with finitely smooth kernels to effectively bridge this gap.
Numerical experiments on 2D benchmark elliptic PDEs demonstrate that the proposed hybrid formulations achieve an optimal balance. Specifically, hybrid kernels exhibit robust, high-order convergence rates that remain highly competitive with, and in certain structural combinations significantly surpass, the accuracy of pure infinitely smooth bases. Simultaneously, they maintain manageable local and global condition numbers, allowing the method to successfully scale to finer spatial resolutions where traditional infinitely smooth kernels fail. These results confirm that the hybrid kernel approach significantly enhances the robustness and accuracy of the D-RBF-PU method, offering a highly practical solution for large-scale simulations.
Keywords
Subjects

[1] Bleecker, D. and Csordas, G. Basic partial differential equations, Chapman and Hall/CRC,
1995.
[2] Cavoretto, R. A numerical algorithm for multidimensional modeling of scattered data points,
Comput. Appl. Math. 34 (2015), 65–80.
[3] Cavoretto, R., De Marchi, S., De Rossi, A., Perracchione, E. and Santin, G. Partition of
unity interpolation using stable kernel-based techniques, Appl. Numer. Math. 116 (2017),
95–107.
[4] Cavoretto, R., De Rossi, A. and Haider, A. A shape-parameterized RBF-partition of unity
technique for PDEs, Appl. Math. Lett. 163 (2025), 109453.
[5] Cavoretto, R., De Rossi, A. and Lancellotti, S. Bayesian approach for radial kernel parameter
tuning, J. Comput. Appl. Math. 441 (2024), 115716.
[6] Esmaeilbeigi, M. and Cheraghi, M. Hybrid kernel approach to improving the numerical
stability of machine learning for parametric equations with Gaussian processes in the noisy
and noise-free data assumptions, Eng. Comput. 40 (2024), 761–794.
[7] Fasshauer, G.E. Meshfree approximation methods with MATLAB, Interdisciplinary Mathe-
matical Sciences, Vol. 6, World Scientific Publishing Co., 2007.
[8] Fasshauer, G.E. and McCourt, M. Kernel-based approximation methods using MATLAB, In-
terdisciplinary Mathematical Sciences, Vol. 19, World Scientific Publishing Co., Singapore,
2015.
[9] Fornberg, B. and Flyer, N. A primer on radial basis functions with applications to the
geosciences, Society for Industrial and Applied Mathematics, Philadelphia, PA, 2015.
[10] Fornberg, B. and Flyer, N. Solving PDEs with radial basis functions, Acta Numer. 24 (2015),
215–258.
[11] Fornberg, B., Larsson, E. and Flyer, N. Stable computations with Gaussian radial basis
functions, SIAM J. Sci. Comput. 33(2) (2011), 869–892.
[12] Fornberg, B. and Wright, G. Stable computation of multiquadric interpolants for all values
of the shape parameter, Comput. Math. Appl. 48(5–6) (2004), 853–867.
[13] Garmanjani, G., Banei, S., Shanazari, K. and Azari, Y. An RBF-PUM finite difference
scheme for forward–backward heat equation, Comput. Appl. Math. 42 (2023), 231.
[14] Hussain, M., Ghafoor, A., Hussain, A., Haq, S., Ali, I. and Ul Arifeen, S. A hybrid kernel-
based meshless method for numerical approximation of multidimensional Fisher’s equation,
Math. Comput. Simul. 223 (2024), 130–157.
[15] Larsson, E., Shcherbakov, V. and Heryudono, A. A least squares radial basis function parti-
tion of unity method for solving PDEs, SIAM J. Sci. Comput. 39(6) (2017), A2539–A2564.
[16] Melenk, J.M. and Babuška, I. The partition of unity finite element method: Basic theory
and applications, Comput. Methods Appl. Mech. Eng. 139(1) (1996), 289–314.
[17] Mirzaei, D. The direct radial basis function partition of unity (D-RBF-PU) method for
solving PDEs, SIAM J. Sci. Comput. 43(1) (2021), A1223–A1249.
[18] Mishra, P.K., Fasshauer, G.E., Sen, M.K. and Ling, L. A stabilized radial basis-finite differ-
ence (RBF-FD) method with hybrid kernels, Comput. Math. Appl. 77(9) (2019), 2354–2368.
[19] Mohammadi, M., Sommariva, A. and Vianello, M. Unisolvence of Kansa collocation for
elliptic equations by polyharmonic splines with random fictitious centers, Appl. Math. Lett.
168 (2025), 109571.
[20] Wendland, H. Fast evaluation of radial basis functions: Methods based on partition of unity,
in Approximation Theory X: Wavelets, Splines, and Applications, Vanderbilt University
Press, Nashville, 2002, 473–483.
[21] Wendland, H. Scattered Data Approximation, Cambridge University Press, 2005.
Send comment about this article
Enter Name.
Enter a valid email address.
Enter a vaid affiliation.
Enter comments (At leaset 10 words)
CAPTCHA Image
Enter Security Code Correctly.