Nonharmonic multivariate Fourier transforms and matrices: condition numbers and hyperplane geometry (Q6932286)

For a finite set \(\aleph\subseteq[-\frac{1}{2}, \frac{1}{2})^d\), \(d\in\mathbb{N}\), consider the arbitrary complex valued, discrete, in \(\aleph\) supported measure \(\mu= \sum^s_{k=1}u_k\delta_{x_k}\), where \(u\in\mathbb{C}^s\). Let \(F_{\Omega,\aleph}:\mathbb{C}^s\rightarrow L^2(\Omega)\) be the restriction of the Fourier transform \(\widehat{\mu}(\omega):=\sum^s_{k=1}u_ke^{-2\pi i\omega x_k}\) on a compact \(\Omega\subseteq\mathbb{R}^d\). Consider a nonharmonic multivariate Fourier matrix \(\Phi_{\Omega,\aleph}:=[e^{-2\pi i\omega\cdot x}]_{\omega\in \Omega\cap\mathbb{Z}^d,x\in\aleph}\) and \(\Phi_{\Omega,\aleph}u=[\widehat{\mu}(\omega)]_{\omega \in \Omega\cap \mathbb{Z}^d}\). Let \(\sum\Omega^p_m:=\{x\in \mathbb{R}^d: ||x||_{l^p}\leq m\},\ p\in [1,\infty]\). Some accurate estimates for the quantities \(\sigma_ {\min } (F_{\Omega^p_m,\aleph}):=\inf_{u\in \mathbb{C}^s, |u|_2=1}||F_{\Omega^p_m,\aleph}u||_{L^2(\Omega^p_m)}\) and \(\sigma_ {\min } (\Phi_{\Omega^p_m,\aleph}):=\inf_{u\in \mathbb{C}^s, |u|_2=1}||\Phi_{\Omega^p_m,\aleph}u||_{L^2(\Omega^p_m)}\) are obtained. These estimates show how the choice of metric and geometry of \(\aleph \) relative to the bandwidth \(2m\) influences their singular value. The three types of estimates proceed in increasing complexity of the geometric assumption on \(\aleph \), from minimum separation, to local sparsity, and to hyperplane decomposition.