Box splines revisited: Convergence and acceleration methods for the subdivision and the cascade algorithms (Q2472366): Difference between revisions

First some terminology: A mask is a family \(a = (a_k)_{k \in \mathbb{Z}^d}\) of complex numbers. The cascade operator associated to the mask is defined (say on \(L_1(\mathbb{R}^d)\)) as \[ (D \varphi)(x) = \sum_{k \in \mathbb{Z}^d} a_k \varphi(2x-k). \] The function \(\varphi\) is called refinable with mask \(a\) if it is a fixed point of the cascade operator associated to \(a\). Refinable functions occur, for instance, in the theory of wavelets: The scaling functions of a multiresolution analysis is a refinable function fulfilling certain extra conditions, and it is the chief tool for the construction of wavelet orthonormal bases. As the authors point out, the customary approach to the construction of refinable functions is via iteration of the cascade operator. This paper studies the convergence of fixed point iteration in the \(p\)-norm, with particular emphasis on the speed of convergence, depending on the seed of the iteration and suitable additional conditions. The chief result of the paper provides an estimate of the speed of convergence for the case of box-splines, \(\varphi = M_{\Xi}\). The sufficient conditions for this estimate are of a purely algebraic nature, and rather easy to check. The proof relies on the so-called double-tree theorem, and thus also provides an improved understanding of the associated subdivision process.