A posteriori error estimation in least-squares stabilized finite element schemes (Q1299285)

The author is concerned with identifying optimally sized and configured meshes for resolving local features of solutions of stabilized discrete problems. The approach is an iterative one, based on constructing a provisional solution using a particular mesh, then constructing the solution to a dual problem based on that provisional solution, and finally employing both solutions to estimate errors and identify mesh regions that are candidates for refinement. The focus of the paper is on formulating the dual problem and establishing the error estimates. The analysis includes the usual case that the error is based on the \(L^2\) norm but also applies to error measurements based on solution values or derivatives evaluated at a chosen point and to other functionals of the solution. The context of the paper is least-squares stabilized finite element methods. Examples discussed in the paper include the Poisson equation expressed using a mixed variational formulation, the Stokes problem for velocity and pressure in fluid mechanics, and an advection-diffusion-reaction problem stabilized using streamline diffusion. For each of these examples, a solution-dependent stabilized dual problem is identified and rigorous a posteriori error estimates constructed. Numerical results are presented for the Poisson equation and illustrate that mesh refinement based on the error estimates is appropriate for various error criteria and also that the error converges with increasing numbers of mesh elements at optimal rates.