Fourth-Order Compact FDMs for Steady and Time-Dependent Nonlinear Convection-Diffusion Equations

TL;DR

This paper introduces high-order compact finite difference methods for steady and time-dependent nonlinear convection-diffusion equations, demonstrating improved accuracy, reduced pollution effects, and satisfying maximum principles, with extensive numerical validation.

Contribution

It develops fourth-order compact FDMs for steady equations and second to fourth-order FDMs for time-dependent equations, with proven stability and accuracy improvements over existing methods.

Findings

FDMs satisfy the discrete maximum principle for small mesh sizes.

Numerical results show high accuracy and convergence rates in space and time.

Proposed methods outperform DG in error magnitude with coarse time steps.

Abstract

In this paper, we discuss the steady and time-dependent nonlinear convection-diffusion (advection-diffusion) equations with the Dirichlet boundary condition. For the steady nonlinear equation, we use an iteration method to reformulate the nonlinear equation into its linear counterpart, and derive a fourth-order compact 9-point finite difference method (FDM) to solve the reformulated equation on a uniform Cartesian grid. To increase the accuracy, we modify the FDM to reduce the pollution effect. The linear system of the FDM generates an M-matrix, provided the mesh size $h$ is sufficiently small. For the time dependent nonlinear equation, we discrete the temporal domain using the Crank-Nicolson (CN), BDF3, BDF4 time stepping methods, and apply a similar iterative method to rewrite the nonlinear equation as the same linear convection-diffusion equation. Then we propose the second-order to fourth-order compact 9-point FDMs with the reduced pollution effects on a uniform Cartesian grid. We prove that all FDMs satisfy the discrete maximum principle for sufficiently small $h$. Several examples with the variable and time-dependent diffusion coefficients and challenging nonlinear terms (not limited to the Burgers equation) are provided to verify the accuracy and the desired convergence rates in the $l_2$ and $l_{\infty}$ norms in space and time. We also compare our second-order CN method with the third-order BDF3 method and the discontinuous Galerkin (DG) method, and the numerical results demonstrate that our FDM with the coarse time step generates the small error. Especially, if the same BDF3 scheme is applied, our error is 1.6\% of that obtained from the DG method. The proposed methods can be easily extended to a 3D spatial domain and more general nonlinear convection-diffusion-reaction equations.