J. Baumeister,A.De Cezaro,A. Leitao
Abstract:We investigate iterated Tikhonov methods coupled with a Kaczmarz strategy for obtaining stable solutions of nonlinear systems of ill-posed operator equations. We show that the proposed method is a convergent regularization method. In the case of noisy data we propose a modification, the so called loping iterated Tikhonov-Kaczmarz method, where a sequence of relaxation parameters is introduced and a different stopping rule is used. Convergence analysis for this method is also provided.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the solution method for stable solutions of nonlinear ill - posed systems of equations. Specifically, the author proposes a method that combines the iterative Tikhonov method with the Kaczmarz strategy to obtain stable solutions of nonlinear ill - posed operator equations. These equations often encounter the situation of inaccurate measurement data in practical applications, so a method that can still converge to a reasonable solution under noisy data is required.
### Main problems
1. **Solution of nonlinear ill - posed systems of equations**:
- The paper focuses on how to determine the unknown physical quantity \(x\in X\) from a set of inaccurate data \((y_0,\ldots,y_{N - 1})\), where \(X\) and \(Y\) are Hilbert spaces and \(N\geq1\).
- The inaccuracy of the data is manifested as the presence of noise, that is, \(\|y_i^\delta - y_i\|\leq\delta_i\), where \(\delta_i>0\) is the noise level.
2. **Deficiencies of existing methods**:
- Existing methods such as the iterative regularization method and the Tikhonov regularization method become inefficient when \(N\) is large or the cost of calculating \(F_i(x)\) and \(F_i'(x)^*\) is high.
- Kaczmarz - type methods are faster in this case, but require further theoretical analysis and improvement to adapt to nonlinear problems.
### Proposed methods
- **Iterative Tikhonov - Kaczmarz method (iTK method)**:
- This method gradually approximates the solution by considering one equation in each iteration step.
- The iteration formula is:
\[
x_{k + 1}^\delta\in\arg\min\left\{\|F_{[k]}(x)-y_{[k]}^\delta\|^2+\alpha\|x - x_k^\delta\|^2\right\}
\]
- where \(\alpha>0\) is an appropriately selected regularization parameter, and \([k]=(k\mod N)\in\{0,\ldots,N - 1\}\).
- **Improved method for noisy data**:
- For noisy data, the author proposes the "jumping" iterative Tikhonov - Kaczmarz method (l - iTK method), introduces a sequence of relaxation parameters, and uses different stopping rules.
- The jumping strategy allows skipping updates in certain iteration steps, thereby terminating the iteration when all residuals are below a specified threshold.
### Convergence analysis
- **Convergence under exact data**:
- It is proved that for exact data, the iTK method converges to a solution of the system of equations.
- If certain additional conditions are met, the iTK method converges to the \(x_0\)-minimum - norm solution \(x^\dagger\).
- **Semi - convergence under noisy data**:
- It is proved that for noisy data, the iTK method has semi - convergence under appropriate assumptions, that is, when the noise level tends to zero, the iterative solution converges to a solution of the system of equations.
- The l - iTK method also has similar semi - convergence under noisy data.
### Applications
- **Parameter identification in elliptic partial differential equations**:
- The paper discusses how to apply the proposed iTK and l - iTK methods to the problem of parameter identification in elliptic partial differential equations, and in particular verifies the applicability of the local tangent cone condition in these problems.
In summary, the main contribution of this paper is to propose a new iterative method that can provide stable solutions when dealing with nonlinear ill - posed systems of equations and still has good convergence under noisy data.