Abstract:We present a novel probabilistic finite element method (FEM) for the solution and uncertainty quantification of elliptic partial differential equations based on random meshes, which we call random mesh FEM (RM-FEM). Our methodology allows to introduce a probability measure on standard piecewise linear FEM. We present a posteriori error estimators based uniquely on probabilistic information. A series of numerical experiments illustrates the potential of the RM-FEM for error estimation and validates our analysis. We furthermore demonstrate how employing the RM-FEM enhances the quality of the solution of Bayesian inverse problems, thus allowing a better quantification of numerical errors in pipelines of computations.
What problem does this paper attempt to address?
### What problems does this paper attempt to solve?
This paper aims to solve problems related to **the numerical solution of elliptic partial differential equations (PDEs) and uncertainty quantification**, especially in the random mesh - based probability finite element method (RM - FEM). Specifically, the paper mainly focuses on the following aspects:
1. **Introducing probability measures to quantify numerical errors**:
- Traditional numerical methods usually use deterministic methods to solve PDEs and evaluate the accuracy of solutions through standard error estimates. However, this method cannot fully reflect the uncertainty introduced by numerical discretization.
- The paper proposes a random mesh - based probability finite element method (RM - FEM). By introducing probability measures in the finite element method, these uncertainties can be quantified more accurately.
2. **Construction of posterior error estimators**:
- The paper develops posterior error estimators based entirely on probability information to evaluate the errors of numerical solutions. These estimators can not only provide error estimates but also be used to guide mesh adaptation, thereby improving the accuracy of solutions.
- Specifically, the paper proposes two posterior error estimators (\(\tilde{E}_{h,1}\) and \(\tilde{E}_{h,2}\)) based on RM - FEM and proves that they are reliable and efficient under certain conditions.
3. **Application in Bayesian inverse problems**:
- In computational pipelines (such as Bayesian inverse problems), traditional deterministic methods may lead to over - confident posterior distributions, thereby underestimating the uncertainty.
- The paper shows how to use RM - FEM to construct empirical distributions and calculate random posterior distributions, thereby better quantifying the uncertainty in inverse problems.
### Main contributions
- **Proposing a new probability finite element method (RM - FEM)**. This method introduces probability measures by randomizing the mesh, so that each sample is a finite element solution, and thus uncertainty can be transferred in subsequent calculations.
- **Constructing posterior error estimators based on probability information**. These estimators are simple to calculate and do not require significant additional computational costs.
- **Verifying the effectiveness of RM - FEM in Bayesian inverse problems**, indicating that this method can improve the quality of numerical solutions, especially when the forward model is relatively rough.
### Conclusion
By introducing the random mesh - based probability finite element method, the paper provides a new idea for solving the numerical solution of elliptic PDEs and uncertainty quantification problems. This method can not only effectively quantify numerical errors but also provide more reliable uncertainty estimates in complex computational pipelines.