A framework for robust a posteriori error control in unsteady nonlinear advection-diffusion problems (Q2840378)

This paper is devoted to practically useful error estimations for a rather general problem: a (possibly degenerate) nonlinear convection-diffusion equation in a \(d\)-dimensional (\(d\geq 2\)) polyhedral domain along with Dirichlet boundary data. For this equation and its solution by some discretization method (in their numerical experiments they take an (in space possibly non-conform) inner penalty discontinuous Galerkin (IPDG) method combined with Crank-Nicolson and upwind numerical flux), the authors are able to construct and prove a computable guaranteed upper bound and an error lower bound, reaching in this way a robust, effective a posteriori error estimation. Their approach contains a linear dual problem with the residual as right hand side -- which may be difficult to solve, they work with a flux reconstruction satisfying (space-time elementwise) mass conservation, take account of the initial error, the possible non-conformity, and quadrature errors. They also prove a computable error lower bound (following Verführt in taking here bubble functions but use them in space and time). For the IPDG method they further prove that both needed assumptions on the flux reconstruction are satisfied. Finally, they show results of numerical experiments in which moving inner layers or degeneracy of the nonlinear diffusion are present. These results exhibit effectivities not far from 1, show further that the errors of the initial data are more and quadrature errors are less important, and that the local error estimators are less close to the real local errors but sufficient to drive an adaptive mesh refinement or coarsening.