An improved model function method for choosing regularization parameters in linear inverse problems (Q4550310)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1787293
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | An improved model function method for choosing regularization parameters in linear inverse problems |
scientific article; zbMATH DE number 1787293 |
Statements
An improved model function method for choosing regularization parameters in linear inverse problems (English)
0 references
3 June 2003
0 references
linear ill-posed problems
0 references
regularization method
0 references
a posteriori parameter choice strategies
0 references
damped discrepancy principle
0 references
Tikhonov regularization
0 references
model function method
0 references
linear inverse problems
0 references
Hilbert spaces
0 references
0.8898868560791016
0 references
0.8739184737205505
0 references
0.8640029430389404
0 references
0.8597205281257629
0 references
Ill-posed problems \[ Tf = z\tag{1} \] are considered, where \( T: X \to Y \) is a bounded linear operator between Hilbert spaces \( X \) and \( Y \). For the stable solution of equation (1), a damped version of the discrepancy principle for Tikhonov's method is considered, i.e. \[ \|Tf(\beta) - z^\delta \|_Y^2 + \beta^\gamma \|f(\beta) \|_X^2 = \delta^2\tag{2} \] Here, \( z^\delta \in Y \) denotes a perturbation of the exact right-hand side \( z \) of equation (1) with \( \|z^\delta - z \|_Y \leq \delta \), and \( 0 < \gamma \leq \infty \) is a fixed parameter. Moreover, \( f(\beta) \in X \) with \( \beta > 0 \) denotes the minimizer of the Tikhonov functional \[ J_\beta(f) = \tfrac{1}{2} \|Tf - z^\delta \|_Y^2 + \tfrac{\beta}{2} \|f \|_X^2, \quad f \in X. \] For the numerical implementation of (2), the authors consider two iterative schemes. For their derivation, (2) is rewritten in the form \[ F(\beta) +(\beta^\gamma -\beta) F^\prime(\beta) = \tfrac{1}{2} \delta^2,\tag{3} \] where the function \( F \) denotes the minimum of the Tikhonov functional, i.e., \( F(\beta) = J_\beta(f(\beta)) \). In the first iteration scheme, for a given approximate regularization parameter \( \beta_k \) the function \( F \) is approximated by a model function of the form \[ m_k(\beta) = \tfrac{1}{2} \|z^\delta \|_Y^2 + C_k/(T_k+\beta) \] where the parameters \( C_k \) and \( T_k \) are determined from the two conditions \( m_k(\beta_k) = F(\beta_k) \) and \( m_k^\prime(\beta_k) = F^\prime(\beta_k) \). The regularization parameter \( \beta_{k+1} \) is then obtained by solving equation \[ m_k(\beta) +(\beta^\gamma -\beta) m_k^\prime(\beta) = \tfrac{1}{2} \delta^2,\tag{4} \] i.e., equation (4) is obtained from equation (3) by replacing \( F(\beta) \) and \( F^\prime(\beta) \) by \( m_k(\beta) \) and \( m_k^\prime(\beta) \), respectively. Monotonicity of the parameters \( C_k \) and \( T_k \) with respect to \( \beta_k \) is proven, respectively. The second iteration scheme is similar to the first iteration scheme, with the same model function \( m_k \) and the left-hand side of equation (4) being modified somewhat. It is shown that the parameters \( \beta_k \) corresponding to the second iteration scheme converge under suitable conditions to the unique solution of equation (2). Some numerical illustrations are presented.
0 references