On convergence and convergence rates for Ivanov and Morozov regularization and application to some parameter identification problems in elliptic PDEs

Barbara Kaltenbacher,Andrej Klassen

DOI: https://doi.org/10.1088/1361-6420/aab739

2018-01-31

Abstract:In this paper we provide a convergence analysis of some variational methods alternative to the classical Tikhonov regularization, namely Ivanov regularization (also called method of quasi solutions) with some versions of the discrepancy principle for choosing the regularization parameter, and Morozov regularization (also called method of the residuals). After motivating nonequivalence with Tikhonov regularization by means of an example, we prove well-definedness of the Ivanov and the Morozov method, convergence in the sense of regularization, as well as convergence rates under variational source conditions. Finally, we apply these results to some linear and nonlinear parameter identification problems in elliptic boundary value problems.

Numerical Analysis

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: when solving the ill - posed elliptic partial differential equation (PDE) parameter identification problem, explore the convergence and convergence rate of Ivanov regularization (quasi - solution of the method) and Morozov regularization (residual method). Specifically:

1. **Introduction of background**:
- Inverse problems are usually expressed as operator equations \(F(x) = y\), where \(F: D(F)\subseteq X\rightarrow Y\), \((X, T_X)\) and \((Y, T_Y)\) are topological spaces.
- Such problems are usually ill - posed, that is, \(F\) is not continuously invertible.

2. **Noise data processing**:
- The actual measurement data \(y^{\delta}\) usually contains noise, and its level \(\delta\) satisfies \(S(y, y^{\delta})\leq\delta\).
- Due to the above ill - posedness, the problem needs to be regularized.

3. **Regularization methods**:
- **Morozov regularization (residual method)**:
\[
x^{\delta}_{\text{Mo}}=\arg\min_{x\in D(F)}R(x)\quad\text{s.t.}\quad S(F(x), y^{\delta})\leq\tau\delta
\]
where \(\tau\geq1\) is a fixed constant independent of \(\delta\).

- **Ivanov regularization (quasi - solution method)**:
\[
x^{\delta}_{\rho}=\arg\min_{x\in D(F)}S(F(x), y^{\delta})\quad\text{s.t.}\quad R(x)\leq\rho
\]
where \(\rho\) is a regularization parameter that needs to be appropriately selected.

4. **Main research contents**:
- Prove the well - definedness of Ivanov and Morozov methods and their convergence as regularization methods.
- Derive the convergence rate under variational source conditions.
- Apply these results to parameter identification problems in some linear and nonlinear elliptic boundary value problems.

5. **Key issues**:
- Avoid relying on Tikhonov regularization equivalence, because this equivalence may fail due to non - convexity.
- Provide specific counterexamples to illustrate the differences between Ivanov and Morozov regularizations and Tikhonov regularization.

Through these studies, the paper aims to provide theoretical support for solving ill - posed elliptic PDE parameter identification problems and explore the effectiveness and applicable scope of different regularization methods.