\(\alpha\)-Molecules (Q285551): Difference between revisions

In the paper, a general framework of \(\alpha\)-molecules is introduced. Most multiscale systems known from applied harmonic analysis, such as wavelets, ridgelets, curvelets, and shearlets, as well as their extensions, are instances of \(\alpha\)-molecules. The parameter \(\alpha\) measures the degree of anisotropy. The construction allows a common treatment of multiscale systems. In particular, the authors list the following goals they want to achieve: {\parindent=6mm \begin{itemize}\item[--] `Encompass most known multiscale systems within the area of applied harmonic analysis.' \item[--] `Allow the construction of novel multiscale systems.' \item[--] `Allow a categorization of systems with respect to their approximation behavior.' \item[--] `Enable a transfer of (sparse approximation) results between the systems within this framework.' \item[--] `Enable the derivation of approximation results by easy-to-verify conditions on certain parameters associated with the system.' \end{itemize}} The first step is the definition of a common parameter space, consisting of parameters that describe a scale, an orientation, and a location. In order to describe orientation, a rotation parameter is chosen, more convenient than shearing. Further, generating functions are chosen, in such a way that they can vary with each parameter, however, they satisfy some time-frequency localization properties, governed by a set of control parameters. Based on those two ingredients, a system of \(\alpha\)-molecules is introduced. All the previously listed multiscale systems (as well as some of their generalizations using different types of scaling) are then described in the framework of \(\alpha\)-molecules. Their parameter spaces are mapped onto the common one. Most difficulties occur in the case of shearlets, since shearing must be replaced by rotation. Further, generating functions for \(\alpha\)-molecules are chosen in such a way that for a special choice of the parameters they yield the classical wavelets, ridgelets, curvelets, or shearlets. Further, control parameters of those \(\alpha\)-molecules are computed. It is also shown that two systems of \(\alpha\)-molecules are almost orthogonal with respect to the \(\alpha\)-scaled index distance in the sense of a strong off-diagonal decay of the associated cross-Gramian matrix. This result provides a base for the classification and analysis of \(\alpha\)-molecules with respect to their approximation properties. Next, \(\alpha\)-molecules are categorized by their approximation behavior by the notion of sparsity equivalence as well as the novel notion of \((\alpha,k)\)-consistency. Further, it is shown how sparse approximation results can be transferred from one system of \(\alpha\)-molecules to another and how sparse approximation results can be concluded from the order of a system of \(\alpha\)-molecules. The paper is well-organized and well-written. It contains many literature hints, and the basic definitions and some proofs are included in the paper. More complicated proofs are postponed to the last section in order not to disturb the fluency of reading.