Convergence of cascade algorithms introduced by I. Daubechies (Q1801401)

The author examines the way the scaling function \(\varphi\) of a multiresolution analysis of \(L^ 2(\mathbb{R}^ n)\) with filter \(m_ 0(\xi)=\sum c_ k \exp(-ik\xi)\) (all sums are over \(k\in \mathbb{Z}^ n\) unless otherwise stated) is related to the properties of \(\varphi\) and of the recursive sequence \(f_ j\) used to construct \(\varphi\). The sequence \(\{c_ k\}\) is assumed to be rapidly decreasing and \(\sum c_ k=1\). Define \(T:L^ 2 (\mathbb{R}^ n)\to L^ 2(\mathbb{R}^ n)\) by \(T(f)(x)=2^ n \sum c_ k f(2x-k)\). Then let \(f_ j=T(f_{j-1})=T^ j(f_ 0)\), where \(f_ 0\in L^ 2(\mathbb{R}^ n)\) is arbitrary. Define the function \(\varepsilon_{p,j}\) by \[ f_ j(x)=\sum_{0\leq |\alpha|\leq p} 2^{-|\alpha|_ j} \partial_ x^ \alpha \varphi(x) \mu_ \alpha (2^ j x)+\varepsilon_{p,j}(x), \] where \(\mu_ \alpha(x)=((- 1)^ \alpha/ \alpha!) \sum P_ \alpha(x-k)f_ 0(x-k)\), \(P_ \alpha(x)=\int_{\mathbb{R}^ n} (x-y)^ \alpha \overline{\varphi} (y)dy\), and \(\alpha\) is a multi-index. Under rather general, but slightly cumbersome, conditions on \(\varphi\) and \(f_ 0\), the author shows that \(\varepsilon_{p,j}\to 0\) in \(L^ 2(\mathbb{R}^ n)\) faster than \(2^{-pj}\) as \(j\to\infty\). Under more restrictive conditions on \(\varphi\) and \(f_ 0\), he shows that \(\varepsilon_{p,j}\to 0\) in the Sobolev space \(H^ p(\mathbb{R}^ n)\). Finally, the author extends a result of \textit{P. Verlinden} and \textit{A. Haegemans} [Numer. Algorithms 2, No. 3/4, 287-298 (1992; Zbl 0772.65095)] concerning the ``dual problem'' to the complex case. Reviewer's remark: ``Entire part'' on p. 309 should presumably be ``integer part''. Other similar uses of language also occur.