Abstract:We prove a.s. (almost sure) unisolvency of interpolation by continuous random sampling with respect to any given density, in spaces of multivariate a.e. (almost everywhere) analytic functions. Examples are given concerning polynomial and RBF approximation.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to prove the unisolvency of interpolation by continuous random sampling in the space of multivariate almost - everywhere analytic functions. Specifically, the authors aim to prove that, given any probability density, the interpolation matrix obtained by random sampling is almost surely nonsingular. This provides a general framework for different methods such as polynomial approximation and radial basis function (RBF) approximation.
### Main Problems
1. **Unisolvency of Random Sampling Interpolation in the Space of Multivariate Almost - everywhere Analytic Functions**
- The authors are concerned with how to perform interpolation by random sampling in the space of multivariate almost - everywhere analytic functions and prove its unisolvency.
2. **Nonsingularity of the Interpolation Matrix**
- The core problem of the paper is to prove that the interpolation matrix \( V_m \) composed of random sampling points is almost surely nonsingular, that is, the probability that \( \det(V_m)\neq 0 \) is 1.
### Solutions
The authors solve the above problems through the following steps:
1. **Definitions and Assumptions**
- Let \( \Omega \) be an open connected subset in \( \mathbb{R}^d \), and \(\{f_j\}_{j \geq 1}\) be a set of functions defined on \( \Omega \), and each \( f_j \) is real - analytic except for a zero - measure set \( I_j\subset \Omega \).
- Assume that these functions are linearly independent on each connected component.
2. **Random Sampling**
- Let \( \{x_i\}_{i \geq 1} \) be a sequence of points randomly distributed on \( \Omega \) according to any given probability density \( \sigma(x) \).
3. **Proof by Induction**
- Prove by mathematical induction that for any \( m\geq 1 \), the interpolation matrix \( V_m = [f_j(x_i)]_{1\leq i,j\leq m} \) is almost surely nonsingular.
- For \( m = 1 \), prove that the probability that \( \det(V_1)=f_1(x_1)\neq 0 \) is 1.
- Assume that \( V_m \) is nonsingular, and then prove that \( V_{m + 1} \) is also nonsingular.
4. **Application Examples**
- Specific applications of polynomial approximation, trigonometric polynomial approximation, and rational function approximation are given.
- It is proved that the random sampling interpolation matrices of radial basis functions (RBFs) such as the Gaussian function, multiquadrics, inverse multiquadrics, and thin - plate splines are almost surely nonsingular.
### Formula Representation
The main formulas involved in the paper include:
- Interpolation matrix \( V_m = [f_j(x_i)]_{1\leq i,j\leq m} \)
- Calculation of matrix determinant: \( \det(V_m) \)
- Determinant expansion in induction: \( \det(U_{m + 1}(x))=\det(V_m)f_{m + 1}(x)+\alpha_m f_m(x)+\cdots+\alpha_1 f_1(x) \)
### Conclusion
Through strict mathematical proofs, this paper establishes the unisolvency of interpolation by random sampling in the space of multivariate almost - everywhere analytic functions. This result is not only applicable to polynomial approximation, but also widely applicable to various numerical methods such as radial basis function approximation, providing a solid theoretical foundation for these methods.