Optimal grids for anisotropic problems (Q871271)

Boundary value problems of anisotropic partial differential equations (PDEs) are considered in two space dimensions, i.e., elliptic equations and wave equations. Finite differences yield a \hbox{(semi-)} discretisation of the PDEs on a grid in space. The aim is to choose the step sizes of the grid optimal in the sense that the difference between the continuous and the discrete Neumann-to-Dirichlet (NtD) operator becomes minimal. This strategy corresponds to rational approximations of the discrete operator. The authors show that the convergence properties of the NtD operator in the anisotropic case coincide with the order of convergence of the NtD operator in the isotropic case on a ray in the complex plane. Corresponding error estimates are proved. Using specific rational approximations, an exponential order of convergence is achieved. The authors analyse problems on finite as well as semi-infinite intervals. Numerical simulations are presented for scalar anisotropic wave equations in bounded squares, where different grid types are combined in a domain decomposition. The results indicate that relatively large errors in regions with coarse grids do not interfere with the accuracy in regions with fine grids, which allows for efficient selections of grids.