An analysis of a multi-level projected steepest descent iteration for nonlinear inverse problems in Banach spaces subject to stability constraints (Q2514242)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | An analysis of a multi-level projected steepest descent iteration for nonlinear inverse problems in Banach spaces subject to stability constraints |
scientific article |
Statements
An analysis of a multi-level projected steepest descent iteration for nonlinear inverse problems in Banach spaces subject to stability constraints (English)
0 references
3 February 2015
0 references
The subject of this paper is a projected steepest descent iteration for solving nonlinear inverse problems in Banach spaces. We give below some of the details. The authors consider an operator equation \( F(x) = y\), \(x \in \mathcal{D}(F)\), \(y \in Y \), with domain \(\mathcal{D}(F) \subset X \), where \(X\) is a \( p \)-convex and \( q \)-smooth Banach space (\( p, q > 1\), \(\tfrac{1}{p} + \tfrac{1}{q} = 1 \)), and \( Y \) denotes a Banach space. In the first part of the paper, it is assumed that \( \Delta_p(x,\tilde{x}) \leq \mathfrak{C}^p \| F(x) - F(\tilde{x}) \|^p \) for \( x, \tilde{x} \in Z \) holds, where \( \Delta_p \) denotes the Bregman distance, \( \mathfrak{C} > 0 \) is a constant, and \( Z \subset X \) is a compact convex set. This means conditional Lipschitz stability of the considered inverse problem. The considered projected steepest descent iteration is of the form \( \tilde{x}_{k+1} = J_q^*(J_p(x_k)-\mu_kT_k), x_{k+1} = P_Z(\tilde{x}_{k+1}) \) for \( k = 0,1, \ldots \), where \( J_p: X \to X^* \) denotes the single-valued duality mapping of \( X \) with gauge function \( t \mapsto t^{p-1} \). In addition, \( P_Z(x) = \) arg min \( \{\Delta_p(y,x) | y \in Z \} \) is the Bregman projection of \( x \in X \) onto \( Z \). Moreover, the notation \( T_k = DF(x_k)^* j_p(F(x_k)-y) \) is used, where \( j_p: Y \to Y^* \) is a single-valued selection of the duality mapping of \( Y \) with gauge function \( t \mapsto t^{p-1} \). The considered iteration is stopped when \( \| F(x_k) - y \| \leq \hat{\eta} \) is satisfied for the first time, where \( \hat{\eta} \) is an appropriately chosen tolerance. It is shown that this algorithm stops after a finite number of steps. For the approximations \( x_k \), a strict monotonicity result with respect to the Bregman distance is presented. In the second part of the paper, a multi-level extension of this projected steepest descent iteration is considered. It is based on a nested family of compact subsets \( Z_\alpha \subset X\), \(\alpha \geq 0 \). The considered algorithm continues from level to level until the iterate satisfies a desired discrepancy criterion. It is shown that this multi-level extension stops after a finite number of steps.
0 references
ill-posed problem
0 references
nonlinear inverse problem
0 references
projected steepest descent method
0 references
Banach space
0 references
multi-level algorithm
0 references
\( p \)-convex Banach space
0 references
\( q \)-smooth Banach space
0 references
Bregman distance
0 references
discrepancy principle
0 references
duality mapping
0 references
nonlinear operator equation
0 references
0 references
0 references
0 references
0 references
0 references
0 references
0 references