Regularization of ill-posed problems by iteration methods (Q4940108)

Regularization methods for ill-posed operator equations can be broadly classified into two groups, discrete and continuous, depending on the nature of the regularization parameter. Continuous methods, such as Tikhonov regularization and mollification depend on a continuous regularization parameter, while for discrete methods, such as expansion methods (e.g., truncated singular value decomposition), and iteration methods, the regularization parameter is an integer. Of course, hybrid methods, such as iterated Tikhonov regularization, are possible. This book is detailed and extensive investigation of iterative regularization methods. Chapter 1 sets the scene by establishing a general Hilbert space context for the study of linear ill-posed problems and reviewing a general spectral-based regularization technique for such problems. The key issue here is the relation of the regularization parameter to the data error. In Chapter 2, steepest descent type methods, including minimum error methods and minimum residual methods for linear ill-posed problems are treated. A general residual principle for choosing the stopping index of the iteration is provided. Similar issues for conjugate direction methods are taken up in Chapter 3. In Chapter 4 the method of steepest descent is investigated for nonlinear ill-posed operator equations in Hilbert space. The analysis is based on certain local conditions for the Fréchet derivative of the operator. Regularizing conjugate gradient methods for ill-posed constrained minimization problems are investigated in Chapter 5. In Chapter 6 regularization methods in incorporating a priori information, that is descriptive regularization methods, are treated. Numerical investigations of regularization algorithms applied to ill-posed problems in thermophysics are provided. The book concludes with an extensive bibliography.