Abstract:This paper concerns quantitative analysis of errors generated by incompletely known data in convex minimization problems. The problems are discussed in the mixed setting and the duality gap is used as the fundamental error measure. The influence of the indeterminate data is measured using the worst case scenario approach. The worst case error is decomposed into two computable quantities, which allows the quantitative comparison between errors resulting from the inaccuracy of the approximation and the data uncertainty. The proposed approach is demonstrated on a paradigm of a nonlinear reaction-diffusion problem together with numerical examples.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the quantitative analysis of errors arising from incomplete knowledge of data in convex minimization problems. Specifically, the author is concerned with the fact that in mathematical models, parameters (data) are usually not completely known but are obtained based on measurements or assumptions. Therefore, it is very important for analysts to quantify the impact of data uncertainty on model results. In particular, in the case of data uncertainty, there is a limit of precision, beyond which any (numerical or analytical) efforts become meaningless or even misleading.
### Main problems
1. **Impact of data uncertainty**: How to quantify and evaluate the errors caused by incompletely known data.
2. **Worst - case analysis**: Measure the impact of data uncertainty through the worst - case scenario method, rather than assuming that some data are more likely than others.
3. **Error decomposition**: Decompose the worst - case error into two computable quantities so that the errors caused by approximation error and data uncertainty can be quantitatively compared.
4. **Practical applications**: Demonstrate the practical application value of the proposed method through specific nonlinear reaction - diffusion problems and their numerical examples.
### Method overview
- **Duality theory**: Based on the duality theory developed by Fenchel, Moreau and Rockafellar, use the duality gap as the basic error metric.
- **Worst - case method**: Unlike the probability method, the worst - case method does not assume the probability distribution of unknown parameters, but looks for the most serious event that may occur.
- **Error quantification**: Propose computable quantities (or two - sided estimates) related to the approximation error and the error caused by incompletely known data, which are of great significance for practical applications.
### Mathematical formulas
The main formulas involved in the paper include:
- Definition of duality gap:
\[
E(u, p^*) := J(u) - I^*(p^*) = 0
\]
- Decomposition of duality gap:
\[
E(v, y^*) = D_F(v, \Lambda^* y^*) + D_G(\Lambda v, -y^*)
\]
- Composite functional in nonlinear reaction - diffusion problems:
\[
D_F(v, -\text{div} y^*) = \int_{\Omega} \left( \frac{1}{p} \gamma |v|^p + \frac{1}{q} \left( \frac{1}{\gamma} \right)^{q - 1} |-\text{div} y^* + f|^q - (-\text{div} y^* + f)v \right) dx
\]
\[
D_G(\nabla v, -y^*) = \frac{1}{2} \| \nabla v + y^* \|_{L^2(\Omega, \mathbb{R}^d)}^2
\]
Through these methods and formulas, the paper aims to provide an effective way to evaluate and control the errors caused by incompletely known data, thereby providing guidance for solving practical problems.