Goal-oriented error estimation for transient parabolic problems

Author (s): Pedro Díez, Giovanni Calderón
Journal: Computational Mechanics

Volume: 39, Issue 5
Pages: 631 – 646
Date: 2007

This work focuses on controlling the error and adapting the discretization in the context of parabolic problems. In order to obtain a sound mathematical framework, the time domain is discretized using a Discontinuous Galerkin (DG) approach. This
allows to formulate the time stepping procedure in a variational format. The error is measured in the basis of an output of interest of the solution, defined by a linear functional. A dual problem, associated with this linear output is introduced.
The dual problem has to be solved backward in time.
An error representation is introduced, based on the weak residual of the primal error applied to the dual solution. Two different alternatives are studied to estimate the error in the dual solution: 1) recovery based error estimators and 2) implicit
residual type estimators. Once the error assessment is performed implicitly in the dual problem, the obtained estimate is plugged into the primal residual to obtain the error in the quantity of interest. The implementation of the estimator
is drastically simplified by using the weak version of the residual instead of the strong version used in previous works.
Thus, the output error is assessed using a mixed technique, explicit for the primal problem and implicit for the dual. In the framework of adaptive computations of transient problems, this approach is very attractive because it allows using first
the implicit scheme for the dual problem and then integrating the primal problem, estimating the error explicitly and eventually adapting the space-time grid. Thus, at every time step of the time marching scheme, the estimate of the dual error is injected into the primal residual (explicit estimate for the primal problem).



author="D{\'i}ez, P. and Calder{\'o}n, G.",
title="Goal-oriented error estimation for transient parabolic problems",
journal="Computational Mechanics",