Vector multivariate subdivision schemes: comparison of spectral methods for their regularity analysis (Q654448): Difference between revisions

The paper deals with the characterization of the regularity of shift invariant subdivision schemes in the multivariate case. An additional hypothesis is that the dimension \(m\) of the joint \(1\)-eigenspace of the \(n\times n\) real matrices associated to the shift invariant subdivision scheme satisfies \(1\leq m \leq n\). The regularity analysis in such a setting is done in terms of the subdivision mask and usually employs either the concept of the joint spectral radius (JSR) of a finite set of matrices derived from the mask, see \textit{D.-R. Chen, R.-Q. Jia} and \textit{S. D. Riemenschneider} [Appl. Comput. Harmon. Anal. 12, No. 1, 128--149 (2002; Zbl 1006.65153)], or restricted spectral properties (RSR), sometimes referred to as contractivity, of so-called difference subdivision schemes, which are also obtained from the mask, see \textit{A. S. Cavaretta, W. Dahmen} and \textit{C. A. Micchelli} [Stationary subdivision, Mem. Am. Math. Soc. 453, (1991; Zbl 0741.41009)]. It has been believed until recently that the JSR and the RSR approaches are intrinsically different. The main contribution of this paper is that it unifies these approaches and shows that they characterize the \(W^k_p\)-regularity, \(k\in{\mathbb N}\), of subdivision in terms of the same quantity. Furthermore, it is shown that the JSR and the RSR approaches differ only by the numerical schemes they provide for the estimation of this quantity, and that the problem of obtaining such RSR estimates is equivalent to solving a problem of linear programming, when \(p=\infty\), or to a problem of convex minimization, when \(1 \leq p <\infty\).