A refined convergence estimate for a fourth order finite difference numerical scheme to the Cahn-Hilliard equation

Jing Guo,Cheng Wang,Yue Yan,Xingye Yue

2024-04-06

Abstract:In this article we present a refined convergence analysis for a second order accurate in time, fourth order finite difference numerical scheme for the 3-D Cahn-Hilliard equation, with an improved convergence constant. A modified backward differentiation formula temporal discretization is applied, and a Douglas-Dupont artificial regularization is included to ensure the energy stability. In fact, a standard application of discrete Gronwall inequality leads to a convergence constant dependent on the interface width parameter in an exponential singular form. We aim to obtain an improved estimate, with such a singular dependence only in a polynomial order. A uniform in time functional bounds of the numerical solution, including the higher order Sobolev norms, as well as the associated bounds for the first and second order temporal difference stencil, have to be carefully established. Certain recursive analysis has to be applied in the analysis for the BDF-style temporal stencil. As a result, we are able to apply a spectrum estimate for the linearized Cahn-Hilliard operator, and this technique leads to the refined error estimate. A three-dimensional numerical example of accuracy check is presented as well.

Numerical Analysis

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the convergence estimate of the numerical solution of the Cahn - Hilliard equation. Specifically, the author proposes a long - stencil finite - difference numerical method with second - order accuracy in time and fourth - order accuracy in space, and performs time discretization through the improved backward difference formula (BDF2), while adding a Douglas - Dupont artificial regularization term to ensure energy stability. The application of the standard discrete Gronwall inequality will lead to a convergence constant in exponential form that depends on the interface width parameter \(\varepsilon\). The goal of this paper is to obtain an improved estimate so that this singular dependence only appears in polynomial form.

### Main contributions of the paper
1. **Improved convergence estimate**: Through spectral analysis techniques, the author is able to obtain a convergence constant that depends on \(\varepsilon\) in polynomial form rather than exponential form. This is an important improvement in theoretical analysis.
2. **Uniform estimate of high - order Sobolev norms**: To overcome the difficulties brought by the discrete Gronwall inequality, the author performs a unified analysis of the numerical errors of the nonlinear part, the concave linear part, and the surface diffusion part. This requires a consistent estimate of the high - order Sobolev norms of the numerical solution in time.
3. **Detailed error analysis**: Through precise control of the time step \(\Delta t\) and the space step \(h\), the author obtains an error estimate in the discrete \(H^{-1}\) norm, and its convergence constant depends on \(\varepsilon\) in polynomial form.

### Mathematical background
The Cahn - Hilliard equation describes the phase separation process of a two - phase system, and its energy functional is:
\[ E(\phi)=\int_{\Omega}\left(\varepsilon^{-1}\left(\frac{1}{4} \phi^{4}-\frac{1}{2} \phi^{2}\right)+\frac{\varepsilon}{2}|\nabla \phi|^{2}\right) d x, \]
where \(\varepsilon\) is the interface width parameter. The Cahn - Hilliard equation can be expressed as:
\[ \phi_{t}=\Delta \mu, \quad \mu:=\delta_{\phi} E=\varepsilon^{-1}\left(\phi^{3}-\phi\right)-\varepsilon \Delta \phi. \]

### Numerical method
The author uses the BDF2 time discretization with second - order accuracy in time and the long - stencil finite - difference method with fourth - order accuracy in space. To ensure energy stability, a Douglas - Dupont artificial regularization term is introduced. The specific form of the numerical scheme is:
\[ \frac{3}{2} \phi^{n + 1}-2 \phi^{n}+\frac{1}{2} \phi^{n - 1}=\Delta t \Delta_{h}^{(4)} \mu^{n + 1}, \]
\[ \mu^{n + 1}=\varepsilon^{-1}\left(\left(\phi^{n + 1}\right)^{3}-2 \phi^{n}+\phi^{n - 1}\right)-\varepsilon \Delta_{h}^{(4)} \phi^{n + 1}-A \varepsilon^{-2} \Delta t \Delta_{h}^{(4)}\left(\phi^{n + 1}-\phi^{n}\right). \]

### Convergence analysis
1. **Energy stability and discrete \(H^{1}\) stability**: Through energy stability analysis, the uniform boundedness of the numerical solution in the discrete \(H^{1}\) norm is proved.
2. **Estimation of high - order Sobolev norms**: Through a consistent estimate of the high - order Sobolev norms of the numerical solution in time, an improved error estimate is obtained.
3. **Spectral analysis**: Using the spectral estimate of the linearized Cahn - Hilliard operator,