Shape optimisation for adaptive $r$-refinement: the one-dimensional case with residual based error estimators

Abstract

We consider $r$-refinement for the finite element discretisation of a Poisson problem. The goal of $r$-refinement is to reposition the nodes of a computational mesh in order to better approximate the finite element error. Since the mesh is being moved, it naturally becomes linked to shape optimisation methods. We propose a standard optimisation algorithm for this $r$-refinement procedure and show that if one seeks to minimise the actual error - which one cannot generally calculate - one has an algorithm which will terminate. The most novel aspect of this work is to apply shape optimisation techniques to the a standard residual error estimator, which is a functional that is differentiable with respect to the mesh, when considered for the Poisson equation in one dimension. To illustrate the approach, a number of numerical experiments are presented, which verify the efficacy of the method.