Adaptive refinement with truncation error injection (Q1095718): Difference between revisions

In the context of finite difference approximation, the difference between a differential equation and the discrete (difference) equation that models it is the truncation error, and it can be easily shown that if the truncation error were known, solving the discrete equation would yield the exact solution at nodal points. An adaptive procedure for improving the accuracy of a numerical solution on a fixed grid, which we call the base, through the approximation of the truncation error by subdomain grid refinements is introduced. Regions where refinements are needed are identified using an estimate of the truncation error. Local solutions on grids tailored to each of the regions are constructed and used to form approximations to the truncation error, which is then `injected' into the base grid to improve the base solution. A one-dimensional model of the convection-diffusion equation is used to demonstrate the basic ideas behind this method; two other examples which imply extensions of this method to multidimensional problems are also studied.