Superconvergence analysis of approximate boundary-flux calculations (Q1326417)

``Superconvergence'' in the finite element method implies local accuracy of higher order than predicted by the standard optimal error estimates. Since in many applications the quantity of interest is the flux across the boundary, it is natural to use the superconvergence of the gradient in order to get better flux approximation. \textit{G. F. Carey, S. S. Chow}, and \textit{M. Seager} [Comput. Meth. Appl. Mech. Eng. 50, 107-120 (1985; Zbl 0546.73057)] proposed a procedure for computing the boundary flux for two-dimensional problems based on \textit{G. F. Carey's} ideas [ibid. 35, 1- 14 (1982; Zbl 0478.73052)] and on ideas due to \textit{M. F. Wheeler} [SIAM J. Numer. Anal. 11, 764-768 (1974; Zbl 0292.65046)]. Numerical experiments showed that the order of convergence is almost \(O(h^ 2)\). In the recent work the authors prove that the accuracy is of order \(O(h^ 2\ln (1/h))\) in the maximum norm for Sobolev-space solutions resp. \(O(h^{3/2})\) in the \(L^ 2\)-norm for Hilbert-space solutions. (The similar analysis of \textit{J. Douglas jun., T. Dupont}, and \textit{M. F. Wheeler} [Revue Franç. Autom. Inform. Rech. Opérat. 8, R-2, 47-59 (1974; Zbl 0315.65063)] has been restricted to differential operators without mixed derivatives instead of the general elliptic operators in the plane studied here).