Optimal frame designs for multitasking devices with weight restrictions (Q1987758): Difference between revisions

This paper is on the design of optimal frames with weight restriction. Given a dimension sequence \(\mathfrak{d}=(d_1,\dots, d_m)\in\mathbb{N}^m\) and a positive weight sequence \(\alpha = (\alpha_1,\dots,\alpha_n)\), the paper considers the so-called \((\alpha,\mathfrak{d})\)-design. That is, the \(m\)-tuples \(\Phi=(\mathcal{F}_1,\dots,\mathcal{F}_m)\) such that each \(\mathcal{F}_j = \{f_{ij}\}_{i=1,\dots,n}\) is a finite sequence of vectors in \(\mathbb{C}^{d_j}\) for \(j=1,\dots, m\) and \[ \sum_{j=1}^m {\|f_{ij}\|^2}=\alpha_i,\quad i=1,\ldots,n. \] Using the notation of majorization between vectors, the paper continues to focus on two tasks: 1) Characterization for the existence of \((\alpha,\mathfrak{d})\)-designs with prescribed spectral structure \(\mathcal{M}=(\mu_1,\ldots,\mu_m)\), where each \(\mu_j=(\mu_{ij})_{i=1}^{d_j}\) is a nonnegative vector in \(\mathbb{R}^{d_j}\). This is with respect to the \((\alpha,\mathcal{M})\)-admissible pair; i.e., there exists an \((\alpha,\mathfrak{d})\)-design \(\Phi=(\mathcal{F}_j)_{j=1}^m\) such that the spectra of the frame operator \(S_{\mathcal{F_j}}\) satisfy \[ \lambda(S_{\mathcal{F}_j}) = \mu_j, \quad j =1,\ldots, m. \] Such a characterization is given in Theorem 3.5. 2) The construction of an optimal \((\alpha,\mathfrak{d})\)-design \(\Phi^{opt}=(\mathcal{F}_j^{opt})_{j=1}^m\) in the sense that it is the minimizer of the joint convex potential \[ P_\varphi(\Phi):=\sum_{j=1}^m P_\varphi(\mathcal{F}_j)=\sum_{j=1}^m \mathrm{tr}\varphi(S_{\mathcal{F}_j}), \] for any convex function \(\varphi:[0,\infty)\rightarrow [0,\infty)\). That is, \[ \Phi^{opt} =\arg\min_{\Phi} P_\varphi(\Phi). \] Such a construction is based on Theorem 3.11. The main conclusion is given in Theorem 3.15, which in particular says that for an \((\alpha,\mathfrak{d})\)-design \(\Phi=(\mathcal{F}_j)_{j=1}^m\), if it is a global minimizer of \(P_{\varphi}\) for a specific \(\varphi\) and \(P_\varphi(\Phi)=P_{\varphi}(\Phi^{opt})\), then each \(\mathcal{F}_j\) is a frame for \(\mathbb{C}^{d_j}\). Examples are given in the last part of the paper.