A-posteriori error estimates for inverse problems

Vishwas Rao,Adrian Sandu

DOI: https://doi.org/10.48550/arXiv.1409.7129

2015-01-30

Abstract:Inverse problems use physical measurements along with a computational model to estimate the parameters or state of a system of interest. Errors in measurements and uncertainties in the computational model lead to inaccurate estimates. This work develops a methodology to estimate the impact of different errors on the variational solutions of inverse problems. The focus is on time evolving systems described by ordinary differential equations, and on a particular class of inverse problems, namely, data assimilation. The computational algorithm uses first-order and second-order adjoint models. In a deterministic setting the methodology provides a posteriori error estimates for the inverse solution. In a probabilistic setting it provides an a posteriori quantification of uncertainty in the inverse solution, given the uncertainties in the model and data. Numerical experiments with the shallow water equations in spherical coordinates illustrate the use of the proposed error estimation machinery in both deterministic and probabilistic settings.

Numerical Analysis

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is to quantify the impact of different errors on the optimal solutions of variational inverse problems. Specifically, the authors focus on how to estimate the influence of observational errors and model errors on the accuracy of the optimal solutions. This research centers on time - evolving systems described by ordinary differential equations (ODEs) and pays special attention to the data assimilation problem.

### Problem Background

In practical applications, inverse problems usually use physical measurements and computational models to estimate the parameters or states of a system. However, due to errors in measurements and uncertainties in computational models, these estimates are often not accurate enough. To improve the reliability of the estimates, researchers need a method to evaluate the impact of these errors on the final results.

### Research Objectives

The main objective of this paper is to develop a methodology for estimating the impact of different types of errors (such as observational errors and model errors) on the optimal solutions of variational inverse problems. Specifically, the authors hope:

1. **Quantify Errors**: Provide a method for post - hoc error estimation to quantify the impact of different error sources on the optimal solutions.
2. **Apply to Time - Evolving Systems**: Especially for time - evolving systems described by ordinary differential equations.
3. **Cover Data Assimilation Problems**: Pay special attention to the data assimilation problem, which is an important category of inverse problems.

### Methodology

To solve the above problems, the authors adopt the following methods:

- **First - and Second - Order Adjoint Models**: Use adjoint models to calculate the impact of errors on the optimal solutions.
- **Hyper - Lagrangian Technique**: By introducing the hyper - Lagrangian function, derive the necessary conditions for error estimation.
- **Perturbed Inverse Problems**: Consider the forward model and cost function with errors and derive the perturbed optimality conditions.
- **Error Estimation Formula**: Based on the perturbed optimality conditions, derive the specific formula for error estimation.

### Main Contributions

The main contributions of this paper are:

- Provide a systematic method to estimate the impact of observational errors and model errors on the optimal solutions of variational inverse problems.
- Apply this method to specific numerical experiments and verify its effectiveness and accuracy.
- Provide theoretical support and technical means for practical problems such as data assimilation.

Through these works, the authors provide a powerful tool for improving the reliability and accuracy of solving inverse problems.