A frequency accurate spatial derivative finite difference approximation

From MaRDI portal





Approximating partial differential equations using finite methods usually introduces grid-dependent errors in the solution. The magnitude of the error and its nature are determined by both the individual approximations and their interaction in approximating the partial differential equation. The most common measure of these errors is the truncation error order: the smallest exponent of the grid spacing term in the error equation. This measure of error has great practical utility in defining how rapidly the approximation error will decrease when the grid spacing is decreased. However, for situations in which the grid resolution is limited by computational resources, the truncation error order is not always a satisfying estimator of the error. This is especially the case in simulating systems with dominant dynamics, systems where significant energy is concentrated near distinct temporal and spatial frequencies. In this situation, the reducing error in the regions of the dominant dynamics, at the expence of an increasing error in regions where there is little energy, is an attractive strategy in selecting an approximation. With this point of view in mind, the authors present a method for designing partial derivative approximations that achieves a priori accuracy in the spatial frequency domain. They use a general average value approximation with undetermined coefficients together with a set of constraints that ensure convergence and consistency to formulate a constrained optimal fitting problem. These constraints lead to a linear matrix formulation. The authors apply the method to the design of spatial approximations for simulating equations with wavelike solutions using both an explicit central difference approximation, which has no phase error, and an upwind design where the level of dissipation can be specified by the designer.











This page was built for publication: A frequency accurate spatial derivative finite difference approximation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4354660)