Optimization over finite frame varieties and structured dictionary design (Q413650)

A finite \((\mu,S)\)-frame variety \(\mathcal F_{\mathbb E}(\mu,S)\) consists of real (for \(\mathbb E=\mathbb R\)) or complex matrices \(F=[f_1,\dots,f_N]\) satisfying \(FF^*=S\) and \(\| f_n\|=\mu_n\) for all \(n=1,\dots,N\) with a \(d\) by \(d\) Hermitian positive operator \(S.\) The paper introduces an approximate geometric gradient descent procedure over these varieties, which is powered by minimization algorithms for the frame operator distance and recent characterizations of these varieties' tangent spaces. For almost all compatible pairings \((\mu,S),\) the author demonstrates that minimization of the frame operator distance converges linearly under a threshold, derives a process for constructing the orthogonal projection onto these varieties' tangent spaces, and finally demonstrates that the approximate gradient descent procedure converges. Finally, he applies this procedure to numerically constructed Grassmannian frames and Welch bound equality sequences with low mutual coherence.