A basic convergence result for conforming adaptive finite elements

We consider the approximate solution with adaptive finite elements of a class of linear boundary value problems, which includes problems of `saddle point' type. For the adaptive algorithm we suppose the following framework: refinement relies on unique quasi-regular element subdivisions and generates locally quasi-uniform grids, the finite element spaces are conforming, nested, and satisfy the inf-sup conditions, the error estimator is reliable as well as locally and discretely efficient, and marked elements are subdivided at least once. Under these assumptions, we give a sufficient and essentially necessary condition on marking for the convergence of the finite element solutions to the exact one. This condition is not only satisfied by D"orfler's strategy, but also by the maximum strategy and the equidistribution strategy.