Duals of frame sequences (Q1038739)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Duals of frame sequences
scientific article

    Statements

    Duals of frame sequences (English)
    0 references
    0 references
    0 references
    0 references
    20 November 2009
    0 references
    Let \(\mathcal{H}\) be a separable Hilbert space and \(J\) a countable index set. A sequence \(\{ y_j\}_{j\in J}\) is called a {\textit{Bessel sequence}} in \(\mathcal{H}\) if the bound \(\sum_{j\in J} |\langle x,\, y_j\rangle|^2\leq C\|x\|_{\mathcal{H}}\) holds for all \(x\in \mathcal{H}\), and \(\{ x_j\}_{j\in J}\) is called a {\textit{frame sequence}} in \(\mathcal{H}\) if \(\sum_{j\in J} |\langle x,\, x_j\rangle|^2\) provides an equivalent norm on the closed span \(\mathcal{H}_X\) of \(X=\{x_j\}\). A frame sequence is a frame if its closed span is all of \(\mathcal{H}\). If \(Y=\{y_j\}\) is a Bessel sequence and \(X=\{x_j\}\) is a frame sequence in \(\mathcal{H}\) then \(Y\) is called a {\textit{generalized}} or {\textit{alternate}} dual of \(X\) or simply a dual, for short, if the restriction of \(T_X T_Y^\ast\) to \(\mathcal{H}_X\) is the identity operator on \(\mathcal{H}_X\). Here, the {\textit{synthesis operator}} \(T_X: \ell^2(J)\to \mathcal{H}\) maps \(c=\{c_j\}\) to \(\sum c_j x_j\in \mathcal{H}_X\) and \(T^\ast_X\) is its adjoint, the {\textit{analysis operator}}. \(Y\) is called a {\textit{Type I}} dual of \(X\) if it is a dual and the range of \(T_Y\) is contained in the range of \(T_X\). It is called a {\textit{Type II}} dual of \(X\) if it is a dual and the range of \(T_Y^\ast\) is contained in the range of \(T_X^\ast\). \(Y\) is called an {\textit{oblique}} dual of \(X\) if it is a dual frame sequence of \(X\) such that \(X\) is also a frame dual of \(Y\). To emphasize distinctions among these concepts the authors provide examples of frame sequences \(X\) with dual \(Y\) such that (i) \(Y\) is not itself a frame sequence, (ii) \(Y\) is a frame sequence but not a Type I dual, (iii) \(Y\) is a frame sequence but not a Type II dual, (iv) \(Y\) is a Type I dual but not a Type II dual of \(X\), (v) \(Y\) is a Type II dual but not a type I dual of \(X\), (vi) \(Y\) is an oblique dual but not a Type I dual of \(X\) and (vii) \(Y\) is an oblique dual but not a Type II dual of \(X\). The authors characterize oblique or Type II duals of \(X\) in terms of direct sum decompositions as follows. Let \(U\) and \(V\) be closed subspaces of \(\mathcal{H}\) and let \(X\) be a frame for \(U\). Then the following are equivalent: (i) \(\mathcal{H}=U+V^\bot\) where \(U\cap V=0\). (ii) There is a frame \(Y\) for \(V\) that is a Type II dual of \(X\). (iii) There is a frame \(Y\) for \(V\) that is an oblique dual of \(X\). In this event, letting \(P\) denote the restriction to \(U\) of the orthogonal projection onto \(V\), one can take \(Y\) to be the canonical dual frame of \(PX\) in \(V\). This result is refined further in the case of shift invariant spaces, that is, Hilbert function spaces defined on \(\mathbb{R}^d\) spanned by the integer shifts \(\phi_j (x-k)\) with \(j\in J\) and \(k\in \mathbb{Z}^d\). Then a uniqueness theorem for duals in this setting is formulated as follows. Let \(\Phi=\{\phi_j:\, j\in J\}\) be a countable subset of \(L^2(\mathbb{R}^d)\) and suppose that \(E(\Phi)=\{\phi_j(\cdot -k):\, j\in J,\, k\in \mathbb{Z}^d\}\). If \(E(\Phi)\) is a frame for its closed linear span and if \(V\) is a shift invariant subspace of \(L^2(\mathbb{R}^d)\) such that \(L^2(\mathbb{R}^d)\) is the direct sum of \(U\) and \(V\) where \(U\cap\, V=\{0\}\), then the following are equivalent: (i) there is a unique shift generated dual \(E(\Psi)\) of \(E(\Phi)\) such that \(|\Psi|=|\Phi|\) and \(E(\Psi)\) is a frame for \(V\), and (ii) \(E(\Phi)\) is quasi-stable in the sense that the fiber sequences \(\{\widehat{\phi}(\xi+k)\}_{k\in \mathbb{Z}^d}\) form a Riesz basis for the sequence space \(\{\{\widehat{u}(\xi+k)\}_k:\,u\in U\}\) for almost all \(\xi\) in the spectrum of \(U\), defined as those \(\xi\in \mathbb{T}^d\) such that \(\widehat{u} (\xi+k)\) is not identically equal to zero for some \(u\in U\).
    0 references
    frame
    0 references
    frame dual
    0 references
    shift-invariant space
    0 references
    0 references
    0 references
    0 references
    0 references

    Identifiers