Well-posedness and convergence rates for sparse regularization with sublinear \(l^q\) penalty term (Q2268306)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Well-posedness and convergence rates for sparse regularization with sublinear \(l^q\) penalty term
scientific article

    Statements

    Well-posedness and convergence rates for sparse regularization with sublinear \(l^q\) penalty term (English)
    0 references
    0 references
    10 March 2010
    0 references
    The article deals with the special regularization procedure of solving an operator equation \(F(u) = v\), where \(F: \;U \to V\) is a possibly nonlinear operator between Banach spaces \(U\) and \(V\). More precisely, the author considers the problem of minimization for the following functional \[ {\mathcal T}_{\alpha,v}(u) = \|F \circ Lu -v\|_V^p + \alpha \sum_{i \in {\mathbb N}} |u_i|^q: \;\;\ell^1 \to {\mathbb R}, \] where \(U = \sum_{i \in {\mathbb N}} u_i\phi_i\), \(\phi_i\), \(i \in {\mathbb N}\), is a bounded sequence in \(U\), \(\alpha > 0\), \(p \geq 1\), \(0 < q < 1\). It is assumed that \(V\) is a reflexive Banach space, \(F \circ L: \;\ell^1 \to V\) is sequentially closed. It is proved (1) the functional \({\mathcal T}_{\alpha,v}\) has a minimizer for every \(\alpha > 0\) and \(v \in V\); (2) if a sequence \((v^{(k)})\) from \(V\) converges to \(v\), then every sequence \((u^{(k)})\) from \(\ell^1\) with \(u^{(k)} \in \text{arg min} \, \{{\mathcal T}_{\alpha,v}(u): \;u \in \ell^1\}\) has a subsequence converging to a minimizer of \({\mathcal T}_{\alpha,v}\). The last statement (3) of the corresponding theorem is vague; it contains the assumption \({\mathcal R}(\hat{u}) < \infty\) but the functional \({\mathcal R}(\cdot)\) in the article is not defined. Further, the author states that every local minimizer \(u\) of \({\mathcal T}_{\alpha,v}\) is sparse (\(u\) has only finitely many non-zero coefficients \(u_i\)) under the assumption that \(F \circ L\) is locally Lipschitz. At last, under some additional assumptions (with the use of \({\mathcal R}(\cdot)\)) it is proved the linear convergence rate.
    0 references
    0 references
    0 references
    0 references
    0 references
    Tikhonov regularization
    0 references
    sparsity
    0 references
    convergence rate
    0 references
    nonlinear operator equation
    0 references
    reflexive Banach space
    0 references
    0 references