Recent experiences with error estimation and adaptivity. I: Review of error estimators for scalar elliptic problems (Q1205653)

Two classes of methods are reviewed for the a posteriori estimation of the error in finite element approximations of elliptic boundary value problems, namely residual and flux-projection error estimators. Any meaningful adaptive method has two major components, namely: error estimation, including the methods and algorithms used for the estimation of the error in approximate solutions, and mesh optimization which refers to the methods which are used to predict the optimal distribution of the parameters of the approximation based on the estimated error and the procedures which are employed in the adaptation of the approximation. The mesh is modified using an adaptive scheme for such parameters as the mesh size and the degree of the local polynomial order. Several versions of the residual method and the popular version of the flux-projection method were implemented in detail for approximations of a scalar elliptic model problem defined on uniform grids of triangles with hierarchic shape functions of polynomial order \(p\), \(1\leq p\leq 7\). Some numerical experiments indicate that flux projection schemes, which are now in use in several comercial finite element program codes, are not asymptotically exact and may lead to very poor predictions of the error for even-order finite element approximations. Fortunately residual estimates, which are theoretically justified, provide asymptotically exact estimators of the global energy norm of the error. Both classes of error estimates may fail to give uniformly accurate estimates of the local distribution of errors over the grid.