Random tight frames (Q421199)

Frame theory has proven to be useful in several areas of analysis such as spherical codes, signal processing, compressed sensing, wavelets and multiresolution analysis. In particular, finite frames for \(\mathbb{R}^d\) have widely been studied. In this paper, the author studies a generalization of the notion of finite frames for \(\mathbb{R}^d\): probabilistic frames. Given \(K\subseteq \mathbb{R}^d\), a probabilistic measure \(\mu\) is a probabilistic frame for \(\mathbb{R}^d\) if there exist \(0<A\leq B\) such that \[ A\|x\|^2\leq\int_K|\langle x, y\rangle|^2\,d\mu(y)\leq B\|x\|^2, \] for all \( x\in\mathbb{R}^d\). This extends the notion of finite frames in \(\mathbb{R}^d\) since if \(\{x_k\}_{k=1}^n\subseteq\mathbb{R}^d\) is a frame for \(\mathbb{R}^d\), then the normalized counting measure is a probabilistic frame with respect to any subset \(K\) containing \(\{x_k\}_{k=1}^n\). The author shows that many concepts (tight and Parseval frames, synthesis, analysis and frame operators) and properties (characterization of tight frames) of classical frame theory can be extended to the probabilistic setting. Then, M. Ehler proves the main theorem of this paper which is a generalization of the following result of \textit{V. K. Goyal}, \textit{M. Vetterli} and \textit{N. T. Thao} [IEEE Trans. Inf. Theory 44, No. 1, 16--31 (1998; Zbl 0905.94007)]: \(n\) random points identically and uniformly distributed on the sphere converge towards a finite unit norm tight frame. The generalization allows for a more flexible choice of the \(n\) points: The points do not need to be identically nor uniformly distributed, and even less, they do not have to lie on a sphere. In the present paper, the \(n\) points only have to be independently distributed according to \(n\) probabilistic tight frames with finite 4th moments. Another important result of this work is related to the frame potential, first introduced by Benedetto and Fickus. The author shows that the minimizers of the probabilistic version of the frame potential are probabilistic tight frames. This provides an extension to the probabilistic setting of a result of Waldrom.