A posteriori error estimators, gradient recovery by averaging, and superconvergence (Q2491148)

An error estimator based upon an appropriate gradient recovery by local averaging is derived for the linear finite element solution to a linear elliptic model problem with strong anisotropy (more specifically, orthotropy) in the coefficients. In contrast to popular variants like the Zienkiewicz-Zhu estimator, the proposed estimator contains some additional terms that ensure reliability also on coarse meshes. Moreover, the enhanced estimator is proved to be (locally) efficient and asymptotically exact whenever the recovered gradient is superconvergent. An adaptive refinement algorithm directed by this estimator is formulated, and the afore-mentioned properties as well as their importance are illustrated by a few fairly severe numerical tests. In particular, it is observed that the smoothed gradient proposed here is superconvergent even in the presence of a crack-type singularity.