Nonsingularity of unsymmetric Kansa matrices: random collocation by MultiQuadrics and Inverse MultiQuadrics

R. Cavoretto,F. Dell'Accio,A. De Rossi,A. Sommariva,M. Vianello
2024-03-27
Abstract:Unisolvence of unsymmetric Kansa collocation is still a substantially open problem. We prove that Kansa matrices with MultiQuadrics and Inverse MultiQuadrics for the Dirichlet problem of the Poisson equation are almost surely nonsingular, when the collocation points are chosen by any continuous random distribution in the domain interior and arbitrarily on its boundary.
Numerical Analysis
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the invertibility problem of **asymmetric Kansa configuration matrices**. Specifically, it is to prove that when using MultiQuadrics (MQ) and Inverse MultiQuadrics (IMQ) radial basis functions (RBF) for meshless collocation, for the Dirichlet problem of the Poisson equation, these matrices are almost surely nonsingular. ### Problem Background 1. **Background of the Kansa Method**: - The Kansa method is a commonly used meshless method, widely applied in engineering and scientific computing to solve partial differential equations (PDEs). It is discretized by radial basis functions (RBF). - Although the Kansa method performs well in practical applications, its mathematical theoretical foundation is still not perfect. In particular, whether the selection of collocation points can ensure the nonsingularity and unisolvence of the collocation matrix remains an open problem. 2. **Limitations of Existing Research**: - The research of Hon and Schaback shows that there are certain "rare" point configurations that can make the Kansa matrix singular, so it is necessary to find sufficient conditions to ensure the nonsingularity of the matrix. - In previous work, the unisolvence in the case of random Kansa collocation using Thin - Plate Splines (TPS) has been proved, but this is limited to the two - dimensional domain and the case where the boundary is an analytic curve, with more limitations. ### Main Contributions of the Paper This paper proves that the asymmetric Kansa configuration matrices using MultiQuadrics (MQ) and Inverse MultiQuadrics (IMQ) are almost surely nonsingular in any dimension by introducing the concept of complex singularities. Specifically: - **Main Result**: When the collocation points are selected by any continuous random distribution inside the domain and arbitrarily on the boundary, the Kansa matrix is almost surely nonsingular. - **Proof Method**: Using the radial properties of the Laplace operator and the analytic properties of MQ and IMQ, combined with the induction method, this conclusion is proved. ### Formula Summary - MultiQuadrics (MQ) function: \[ \varphi(r)=\sqrt{1 + (\varepsilon r)^2} \] - Inverse MultiQuadrics (IMQ) function: \[ \varphi(r)=\frac{1}{\sqrt{1 + (\varepsilon r)^2}} \] - Poisson equation: \[ \begin{cases} \Delta u(P)=f(P), & P\in\Omega, \\ u(P)=g(P), & P\in\partial\Omega, \end{cases} \] - The expression of the Laplacian in polar coordinates: \[ \Delta\varphi_A=\frac{\partial^2\varphi}{\partial r^2}+\frac{1}{r}\frac{\partial\varphi}{\partial r} \] - Specific Laplacian values for MQ and IMQ: \[ \Delta\varphi_A = \begin{cases} \frac{\varepsilon^2(2 + (\varepsilon r)^2)}{(1 + (\varepsilon r)^2)^{3/2}}, & \text{for MQ}, \\ \frac{-\varepsilon^2(2 - (\varepsilon r)^2)}{(1 + (\varepsilon r)^2)^{5/2}}, & \text{for IMQ}. \end{cases}