Representation of polynomials by linear combinations of radial basis functions (Q1943997)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Representation of polynomials by linear combinations of radial basis functions
scientific article

    Statements

    Representation of polynomials by linear combinations of radial basis functions (English)
    0 references
    3 April 2013
    0 references
    Let \(d\in\mathbb{N}\), and denote by \(C_n^{d/2}\) the Gegenbauer polynomials of degree \(n\in \mathbb{N}_0\) associated with the parameter \(d/2\). The properly normalized Gegenbauer polynomials form a complete orthogonal system \(\{U_n : n\in\mathbb{N}_0\}\) for \(L^2([-1,1],\sigma)\), where \(\sigma\) is the weight function \((1-t^2)^{(d-1)/2}\). Let \(\mathbf{I} :=\{I := (i,j,k) : i \in\mathbb{N}_0, j \in \{0,1,\dotsc, i\},\;j = i (\mod 2),\;k = 1, \dotsc, l_j\}\), where \(l_j\) is the dimension of the space of spherical harmonics of degree \(j\). For \(I\in \mathbf{I}\), define \[ P_I(x) := \nu_{ij}\,\int_{\mathbb{S}^{d-1}} U_i (x\cdot \xi) Y_{j,k}(\xi)d\xi, \] where \(Y_{j,k}\) is a spherical harmonic polynomial and \(\nu_{ij}\) is chosen so that \(\|P_I\|_{L^2} = 1\). The author defines a new space of polynomials \(\mathcal{Q}_{2n}^d\) as \[ \mathcal{Q}_{2n}^d := \operatorname{span} \{P_I : I\in \mathbf{I}_n\}, \] where \(\mathbf{I}_n := \{I = (i,j,k)\in \mathbf{I} : i+j\leq 2n\}\). The main result of the paper is the following statement: Let \(\mathcal{A}\) be a finite set of points in \(\mathbb{R}^d\) and \(h\) a fixed univariate even polynomial of degree \(n\). Then, every polynomial \(P\in \mathcal{Q}_{2n}^d\) can be represented as a linear combination of shifted radial basis functions of the form \(h(\|\bullet + a\|)\), \(a\in \mathcal{A}\), if and only if \(\mathcal{A}\) is a uniqueness set for \(\mathcal{Q}_{2n}^d\). (Here, a uniqueness set for \(\mathcal{Q}_{2n}^d\) is any finite set \(\mathcal{A}\) such that, for any pair of polynomials \(p_1,p_2\in \mathcal{Q}_{2n}^d\) with \(p_1(a_i) = p_2 (a_i)\), \(i = 1, \dotsc, \operatorname{card}(\mathcal{A})\), the implication \(p_1(a) = p_2(a)\) for all \(a\in \mathbb{R}^d\) holds.)
    0 references
    radial basis functions
    0 references
    polynomials
    0 references
    linear representation
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references