Interior Estimates for Ritz-Galerkin Methods

Abstract

Interior a priori error estimates in Sobolev norms are derived from interior Ritz-Galerkin equations which are common to a class of methods used in approximating solutions of second order elliptic boundary value problems.The estimates are valid for a large class of piecewise polynomial subspaces used in practice, which are defined on both uniform and nonuniform meshes.It is shown that the error in an interior domain can be estimated with the best order of accuracy that is possible locally for the subspaces used plus the error in a weaker norm over a slightly larger domain which measures the effects from outside of the domain ii.Additional results are given in the case when the subspaces are defined on a uniform mesh.Applications to specific boundary value problems are given.0. Introduction.There are presently many methods which are available for computing approximate solutions of elliptic boundary value problems which may be classified as Ritz-Galerkin type methods.Many of these methods differ from each other in some respects (for example, in how they treat the boundary conditions) but have much in common in that they have what may be called "interior Ritz-Galerkin equations" which are the same.Here we shall be concerned with finding interior estimates for the rate of convergence for such a class of methods which are consequences of these interior equations.Let us briefly describe, in a special case, the type of question we wish to consider.Let 2 be a bounded domain in R^ with boundary 92 and consider, for simplicity, the problem of finding an approximate solution of a boundary value problem (0.1) Am =/ in 2, (0.2) Au =g on 92, where A is some boundary operator.Suppose now that we are given a one-parameter family of finite-dimensional subspaces Sh (0 < A < 1) of an appropriate Hilbert space in which u lies and that, for each A, we have computed an approximate solution un GSh to u using some Ritz-Galerkin type method.Here we have in mind, for example, methods such as the "engineer's" finite element method [8], [22], the Aubin-BabusTca penalty method [2], [4], the methods of Nitsche [12], [13] or the