Super-convergence and post-processing for mixed finite element approximations of the wave equation

Herbert Egger,Bogdan Radu

DOI: https://doi.org/10.1007/s00211-018-0966-2

2016-08-12

Abstract:We consider the numerical approximation of acoustic wave propagation problems by mixed BDM(k+1)-P(k) finite elements on unstructured meshes. Optimal convergence of the discrete velocity and super-convergence of the pressure by one order are established. Based on these results, we propose a post-processing strategy that allows us to construct an improved pressure approximation from the numerical solution. Corresponding results are well-known for mixed finite element approximations of elliptic problems and we extend these analyses here to the hyperbolic problem under consideration. We also consider the subsequent time discretization by the Crank-Nicolson method and show that the analysis and the post-processing strategy can be generalized to the fully discrete schemes. Our proofs do not rely on duality arguments or inverse inequalities and the results therefore apply also for non-convex domains and non-uniform meshes.

Numerical Analysis

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is how to achieve the optimal convergence of the discrete velocity and the super - convergence of the pressure when using the mixed finite element method (BDM \(k + 1\)-\(P_k\) element) to numerically approximate the acoustic wave propagation problem on unstructured grids, and based on these results, propose a post - processing strategy to construct an improved pressure approximation. Specifically:

- **Research Background and Problem Description**:
- The propagation of acoustic waves in Newtonian fluids or elastic solids can be modeled by hyperbolic systems, such as:
\[
a\frac{\partial p}{\partial t}+\text{div}u = 0,\quad b\frac{\partial u}{\partial t}+\nabla p = 0
\]
where \(p\) and \(u\) represent the pressure field and the velocity field respectively, and the parameters \(a\) and \(b\) encode the physical properties of the medium (such as density or sound speed).

- **Existing Methods and Their Limitations**:
- Standard Galerkin approximations usually focus on the approximation of the pressure variable \(p\), but in many cases, the velocity field \(u\) is the main quantity of interest.
- The semi - discretization error estimate using the BDM \(k + 1\)-\(P_k\) element is:
\[
\|p(t)-p_h(t)\|_{L^2(\Omega)}+\|u(t)-u_h(t)\|_{L^2(\Omega)}\leq Ch^{k + 1}
\]
This estimate is optimal for pressure but sub - optimal for velocity.

- **Main Contributions of the Paper**:
- It is proved that for the BDM \(k + 1\)-\(P_k\) approximation, the error is actually of the optimal order:
\[
\|\pi_hp(t)-p_h(t)\|_{L^2(\Omega)}+\|u(t)-u_h(t)\|_{L^2(\Omega)}\leq Ch^{k+2}
\]
where \(\pi_h\) represents the \(L^2\) projection onto the discrete pressure space. This shows that the projected pressure error is even super - convergent by one order.
- A post - processing strategy is proposed. By local calculation, an improved pressure approximation \(\tilde{p}_h\) is constructed from the discrete solution \((p_h, u_h)\) and the problem data, such that:
\[
\|u(t)-u_h(t)\|_{L^2(\Omega)}+\|p(t)-\tilde{p}_h(t)\|_{L^2(\Omega)}\leq Ch^{k+2}
\]

- **Time Discretization**:
- The paper also considers the subsequent time discretization (such as the Crank - Nicolson method) and proves that the super - convergence estimate and the post - processing strategy can be extended to the fully discrete scheme, so that the optimal convergence order is achieved in both space and time.

In conclusion, this paper aims to improve the accuracy and efficiency of the numerical solutions of the acoustic wave propagation problem through the improved mixed finite element method and post - processing techniques.