An embedding theorem on reducing subspace frame multiresolution analysis (Q418267)

Denote \(F(L^2(\Omega))=\{f\in L^2(\mathbb{R}^d): \text{supp}\,\widehat{f}\subset \Omega\}\). A {\textit{reducing}} subspace of \(L^2(\mathbb{R}^d)\) is one that is invariant under the dilation \(f\mapsto Df=|\det{M}|^{1/2} f(M\cdot)\) in which \(M\) is an expansive matrix, meaning its entries are integers and all eigenvalues exceed one in modulus. Reducing subspaces all have the form \(F(L^2(\Omega))\) such that \(M^t\Omega=\Omega\). Given an expansive matrix, a frame multiresolution analysis (FMRA) is a sequence \(\{V_j\}\) of nested subspaces of \(F(L^2(\Omega))\) whose union is dense in \(F(L^2(\Omega))\), whose intersection is zero, and satisfying \(V_j=D^j V_0\), \(j\in \mathbb{Z}\), such that there is a \(\varphi\in F(L^2(\Omega))\), called a scaling function, whose shifts \(T_k\varphi(\cdot)=\varphi(\cdot-k)\), \(k\in\mathbb{Z}^d\), have dense span in \(F(L^2(\Omega))\). After proving the existence of \(M\)-admissible scaling sets, that is, measurable subsets of \(S\subset\Omega\), the inverse Fourier transform, of which the characteristic functions are scaling functions, the authors show that FMRAs admit MRA embeddings. That is, if \(M\) is expansive and \(F(L^2(\Omega))\) is a reducing subspace with \(\Omega\) having nonempty interior, and such that \(\{V_j\}\) is an FMRA for \(F(L^2(\Omega))\) associated with \(M\), then there exists an MRA \(\{\mathcal{V}_j\}\) for \(F(L^2(\Omega))\) -- that is, the shifts of its scaling function form a Riesz basis -- associated with \(M\) such that \(\mathcal{V}_j\subset V_j\). It is also shown that for any reducing subspace there exists an MRA that cannot be properly embedded in an FMRA, that is, whose scaling function is a frame but not a Riesz basis.