Quality assessment of the a-posteriori error estimation in finite elements (Q1205249)

A posteriori error estimators are often derived by purely heuristic reasons without mathematical analysis. Usually the estimators are numerically analyzed or verified on a set of examples, so called benchmarks, which are selected at best by heuristic reasoning. These benchmark computations are used for the design of a correction factor which is used to improve the quality of the estimators for this set of benchmarks. Each benchmark computation should be a representative of a more or less precisely defined class of computational problems and the conclusions used for this class of problems. The authors address various aspects of the relation between theoretical understanding of an estimator and the benchmark selection. Certain estimators which are used in practice are investigated and their properties are discussed. The selection of benchmarks is directed to a numerical verification of basic theoretical features of the selected estimators. The authors do not intend to compare these selected estimators with any other error estimator.