Wavelet-type decompositions and approximations from shift-invariant spaces (Q5961553)

A principal shift-invariant space \(S=S(\phi)\) is the closed linear span of the translates of \(\phi\) in \(L_p({\mathbf R}^d)\), \(1<p<\infty\). Define the scaling spaces \(S^h=\{f(\cdot/h): f\in S\}\), \(h>0\). The approximating quality of \(S^h\) is measured by its approximation order (for some class of functions \(F\)). This order is \(r\) if \(\inf_{s\in S^h}|f-s|_{L_p}=O(h^r)\) as \(h\to 0\) for every function \(f\in F\). Conditions on \(\phi\) monitor the smoothness required for the class \(F\). Weakening the conditions on \(\phi\) reduces the set \(F\). In this paper, conditions on \(\phi\) are considered which allow for \(F\) to choose Besov spaces \(B_p^{r,q}\), \(0<q\leq 1\). In the classical approaches, it is usually assumed that the Fourier transform \(\hat{\phi}\) vanishes at the origin: \(\hat{\phi}(0)=0\). An important contribution of this paper is that this condition is avoided. This is obtained by a technique based on wavelets: First \(f\) is approximated by a truncated wavelet expansion and then each wavelet basis function is approximated by an element from \(S^h\). Since wavelets have mean zero, \(\hat{\phi}(0)=0\) can be easily incorporated.