High order compact schemes for gradient approximation (Q625760)

The authors present three high-order compact schemes for gradient recovery: (1) a weighted linear combination of the gradient on the neighborhood elements; (2) a geometric average of two cubic interpolations on macro elements; (3) fitting a polynomial of degree 3 at some sampling points on element patches by a least-squares argument. It is shown that these schemes can approximate the gradient of the exact solution on symmetric points to fourth order. In particular, for a uniform mesh, these three schemes are the same on the considered points. Though the first scheme is easy to implement, it only goes well on uniform grids. Besides uniform grids, the second and third schemes can perform on strongly regular grids as well. These results are applied to the Poisson problem to get a superconvergence error estimate. Some numerical experiments justify the theoretical results.