Radial positive definite functions and Schoenberg matrices with negative eigenvalues

Abstract

The main object under consideration is a class Φ n ∖ Φ n + 1 \Phi _n\backslash \Phi _{n+1} of radial positive definite functions on R n \mathbb {R}^n which do not admit radial positive definite continuation on R n + 1 \mathbb {R}^{n+1} . We find certain necessary and sufficient conditions on the Schoenberg representation measure ν n \nu _n of f ∈ Φ n f\in \Phi _n for f ∈ Φ n + k f\in \Phi _{n+k} , k ∈ N k\in \mathbb {N} . We show that the class Φ n ∖ Φ n + k \Phi _n\backslash \Phi _{n+k} is rich enough by giving a number of examples. In particular, we give a direct proof of Ω n ∈ Φ n ∖ Φ n + 1 \Omega _n\in \Phi _n\backslash \Phi _{n+1} , which avoids Schoenberg’s theorem; Ω n \Omega _n is the Schoenberg kernel. We show that Ω n ( a ⋅ ) Ω n ( b ⋅ ) ∈ Φ n ∖ Φ n + 1 \Omega _n(a\cdot )\Omega _n(b\cdot )\in \Phi _n\backslash \Phi _{n+1} for a ≠ b a\not =b . Moreover, for the square of this function we prove the surprisingly much stronger result Ω n 2 ( a ⋅ ) ∈ Φ 2 n − 1 ∖ Φ 2 n \Omega _n^2(a\cdot )\in \Phi _{2n-1}\backslash \Phi _{2n} . We also show that any f ∈ Φ n ∖ Φ n + 1 f\in \Phi _n\backslash \Phi _{n+1} , n ≥ 2 n\ge 2 , has infinitely many negative squares. The latter means that for an arbitrary positive integer N N there is a finite Schoenberg matrix S X ( f ) := ‖ f ( | x i − x j | n + 1 ) ‖ i , j = 1 m \mathcal {S}_X(f) := \|f(|x_i-x_j|_{n+1})\|_{i,j=1}^{m} , X := { x j } j = 1 m ⊂ R n + 1 X := \{x_j\}_{j=1}^m \subset \mathbb {R}^{n+1} , which has at least N N negative eigenvalues.

Representations of hypergeometric functions for arbitrary parameter values and their use

“…The class of n-RPD functions is denoted by Φ n . The above definition and many further details can be found, for instance, in the two recent papers [14,15]. The class Φ n has been characterized by Schoenberg in 1938: Classes Φ n are known to be nested: Φ n+1 ⊂ Φ n , and the inclusion is proper.…”

Optimal Bandwidth for Geographically Weighted Regression to Model the Spatial Dependency of Land Prices in Manado, North Sulawesi Province, Indonesia

“…Manganelli et al (2013) stated that extrinsic factors in the form of infrastructure, accessibility of specificpublic services, and availability of basic commercial services are very important in determining land prices. Grace and Saberi (2018) measured the accessibility of infrastructure for a certain location using spatial autocorrelation (Moran's I) followed by an Ordinary Least Square model. Chen (2018) studied land price differentiation and its factors in Guangdong, China using Exploratory Spatial Data Analysis (ESDA) and GWR methods.…”

Mathematical Developments on Isotropic Positive Definite Functions on Spheres

“…, Z d ) T be a zero-mean Gaussian vector-valued random field on S d , with the matrix covariance operator K Z of Z, given by The difference between the result in (Gneiting, 2013a, corollary 4) and that problem is there the walk dimension with step 2, but here the step is 1. Also, Golinskii et al (2015) studied that problem on the Euclidean Space R d . Problem 5.2.2 is important to solve the second problem in Gneiting (2013b).…”