High order finite difference numerical methods for time-dependent convection-dominated problems (Q2573277): Difference between revisions

The paper continues the investigation of specific implicit finite difference methods applied to initial value problems of convection-diffusion equations. \textit{M. A. Pirozzi} [High-accuracy finite difference methods for systems of conservation laws, II, Tech. Rep. No. 6, Instituto di Matematica, Fisica e Applicazioni, Universita di Napoli `Parthenope' (2002)] has constructed a parameter-dependent implicit discretisation of the linear convection-diffusion equation, which represents a parabolic PDE in case of one space dimension. The investigation of the corresponding modified equation allows to set the parameters such that high order schemes are achieved for the convection part in space. Furthermore, the parameters can be used to reduce dispersion and dissipative behaviour in the approximations. Based on a linearisation technique, the method can be used for solving nonlinear convection-diffusion equations, too. In the paper at hand, this class of implicit finite difference methods is generalised to two and three space dimensions. A splitting scheme arises, where just subsequent tridiagonal linear systems with the complexity of one space dimension have to be solved. Nevertheless, the parameters can be set such that the local truncation error in the convection part exhibits order two or three. Furthermore, the choices produce linear systems with strictly diagonal dominant matrices. The criterion of von Neumann does not restrict the size of time steps, which indicates good stability properties. In the main part of the paper, the authors present numerical simulations, which verify the resulting orders and indicate the magnitude of errors in amplitude and phase for corresponding approximations. However, the used equations are just the one- and two-dimensional linear convection-diffusion equation and the one-dimensional viscous Burgers' equation. Several choices for the parameters in the schemes and for the Courant number, which couples space and time step size, are used. Since only the convection part exhibits a high order approximation, the diffusion coefficient is reduced according to the space step size, where a hyperbolic partial differential equation arises in the limit. Thus the high order approximation is just given in the case of dominating convection, i.e., the diffusion part does not influence the complete error significantly. The results for the one-dimensional linear equation with a certain choice of parameters show that the resolution of amplitude and phase is as good as for the Crank-Nicolson mass operator scheme. Moreover, tests with discontinuous initial data determine sharp fronts correctly without oscillations or too much dissipation at later times.