Adjoint recovery of superconvergent linear functionals from Galerkin approximations. The one-dimensional case (Q2385511)

The approximation for a value of an integral functional corresponding to the solution of a partial differential equation is improved using the associated adjoint problem by \textit{N. A. Pierce} and \textit{M. B. Giles} [SIAM Rev. 42, No. 2, 247--264 (1998; Zbl 0948.65119)]. Thereby, the order of the approximation increases significantly. This phenomenon is called superconvergence. In the paper at hand, the authors extend the previous approach to discontinuous Galerkin methods in the one-dimensional case. Consequently, integral functionals corresponding to the solution of ordinary differential equations (ODEs) are considered. Using piecewise polynomials of degree \(k\) in the discontinuous Galerkin method yields an approximation of the functional with order around \(2k\). Solving the ODEs as well as the adjoint problem enables the construction of an approximation of order around \(4k\). The authors propose two strategies for this postprocessing. Firstly, polynomial interpolation of a high degree achieves better approximations by the values at the given nodes. Secondly, a Galerkin method based on polynomials of degree \(2k\) is applied in each subinterval separately. Two model problems are discussed in the paper. Firstly, an initial value problem of a scalar linear ODE is considered. Secondly, a one-dimensional convection-diffusion problem is investigated, which represents a boundary value problem of a linear ODE of second order. Thereby, the corresponding system of ODEs of first order is employed. The authors construct the enhanced approximations of corresponding integral functionals for both problems. The proofs of error bounds are given in detail. Furthermore, numerical results confirm the validity of the proved orders of convergence in both cases. Finally, the authors outline an improvement of the approximation of the solution corresponding to the ODE itself in case of the convection-diffusion problem, where a relatively low additional effort is required.