Nonorthogonal multiresolution analyses, commutation between projectors and differentiation and divergence-free vector wavelets (Q1202515): Difference between revisions

This paper explores the connections between biorthogonal wavelet bases and differentiation. Suppose that \(g,g^*\in L^2(\mathbb{R})\) satisfy the following properties: (a) \(\langle g(x- k), g^*(x- 1)\rangle= \delta_{k,l}\), \(k,l\in\mathbb{Z}\), (b) \(g,g^*\in C^{\varepsilon}\) for some \(\varepsilon>0\) and with compact support, (c) \(g(x/2)\) is a finite linear combination of \(g(x- k)\), \(k\in\mathbb{Z}\), and similarly for \(g^*\), (d) \(\int_{\mathbb{R}} g(x)dx= \int g^*dx=1\). Then \(g\) and \(g^*\) are called conjugate scaling functions. Let the closed linear span of \(\{g_{jk}(x)= 2^{j/2}g(2^jx-k): k\in\mathbb{Z}\}\) be \(V_j\) and let \(P_j= \sum_{k\in\mathbb{Z}}\langle f,g^*_{jk}\rangle g_{jk}\) be the nonorthogonal projection onto \(V_j\) perpendicular to \(V^*_j\). Then \((V_j)\), \(j\in\mathbb{Z}\), is a multiresolution analysis. The first results concern the properties of this multiresolution analysis and the construction of a Riesz basis for \(L^2(\mathbb{R})\) of the form \(\{2^{j/2}\gamma(2^jx-k): j,k\in\mathbb{Z}\}\), where the function \(\gamma\in C^\varepsilon\) is derived from \(g\) and has compact support. The biorthogonal basis has the same structure and is derived from \(g^*\). The next question treated is how differentiation is compatible with the structure of such nonorthogonal multiresolution analysis. Proposition: If \(g, g^*\) are conjugate scaling functions and if \(g\in C^{1+\varepsilon}\), \(\varepsilon>0\), then there exist two conjugate scaling functions \(h, h^*\) and corresponding multiresolution analyses, such that \(g'(x)= h(x)-h(x-1)\), \(g^*(x+1)- g^*(x)=(h^*)'(x)\). The nonorthogonal projections \(P_j\) and \(\widetilde{P_j}= \sum_{k\in\mathbb{Z}}\langle f,h^*_{jk}\rangle h_{jk}\) satisfy \((d/dx)\circ P_j=\widetilde{P_j}\circ d/dx\). By taking tensor products and Cartesian products, one obtains vector-valued multiresolution analyses \(\vec{V_j}\) for \(L^2(\mathbb{R}^n)\times\cdots\times L^2(\mathbb{R}^n)= L^2(\mathbb{R}^n)^n\) with nonorthogonal projections \(\vec{P_j}\) onto \(\vec{V_j}\). This construction is carried out in such a way that, similar to the one-dimensional case, \(\text{div}(\vec{P_j}\vec{f})= P_j(\text{div }\vec{f})\) holds, where \(P_j\) corresponds to a one-dimensional multiresolution analysis. The main theorem asserts the existence of compactly supported vector-valued functions \(\vec\Gamma_s\), \(s\) being in some index set \(S\), such that \(\{2^{l n/2}\vec\Gamma_s(2^l\vec{x}- \vec{k})\), \(l\in\mathbb{Z}\), \(\vec{k}\in \mathbb{Z}^n\), \(s\in S\}\) is a Riesz basis for \(\vec{H}= \{\vec{f}\in L^2(\mathbb{R}^n)^n:\text{div }\vec{f}= 0\}\). Moreover, the biorthogonal basis has a similar structure.