Asymptotic estimate in an algorithm for the wavelet transform. Connection with the regularity of the wavelet (Q1901334)

Let \(\phi\) be the scaling function of a multiresolution analysis, defined by a refinement equation with filter coefficients \((h_k)\). Then in every step of the corresponding wavelet transform, one has to perform a filter operation characterized by the operator \(T_0\) defined as \((T_0 x)(n)=2^{-1/2}\sum_k h_k x(2n-k)\), \(x\in\ell^2({\mathbb{Z}})\). Thus at level \(j\), one has computed implicitly or explicitly \(T_0^j x\). This paper links the speed of convergence of \(T_0^j\) as \(j\to\infty\) with the regularity of the scaling function \(\phi\). The main result can be formulated as follows. Let \(\hat{\phi}\) be the Fourier transform of \(\phi\) and \(s_1=\sup\{s\in{\mathbb{R}}: \int_{-\infty}^\infty |\lambda|^s|\hat{\phi}(\lambda)|d\lambda<+\infty\}\). Then, in a subspace of \(\ell^2({\mathbb{Z}})\), the operator sequence \((2^{j/2}T_0^j)_{j\geq1}\) converges in norm if \(s_1>0\). The convergence is exponential with order \(2^{-js_1}\) if \(s_1<1\) and with order \(2^{-j}\) if \(s_1>1\).