Orthogonality constrained gradient reconstruction for superconvergent linear functionals (Q2185875)

Linear functionals of the solution of variational problems discretized with Galerkin finite elements are important in engineering applications and goal-oriented adaptivity. The authors show that the non-optimal convergence properties induced by the superconvergent patch recovery (SPR) technique are related to the loss of the Galerkin orthogonality condition. An enhanced variant of the SPR method is proposed to remedy that, characterized by the addition of an orthogonality constraint to the discrete least squares problem that represents the SPR approach. In this way, the reconstructed gradient satisfies the Galerkin orthogonality condition. The new gradient reconstruction technique not only increases the accuracy but also preserves the convergence properties of functionals of the solution. The constraint yields simultaneously a more accurate approximation of the gradient and linear functionals as observed analyzing numerical results of the solutions of generic second-order elliptic problems in different dimensions.