Complementary error bounds for foolproof finite element mesh generation (Q1062742)

By using the Laplace equation as a basic example, the authors formulate the corresponding complementary variational principles whose solutions are respectively potential function \(\phi\) and stream function \(\psi\). The approximation by finite element methods give approximations of \(\phi\) and \(\psi\) that are still denoted \(\phi\) and \(\psi\) ; from these approximations, compute \(\bar A_ 1=-\nabla \phi\), \(\bar A_ 2=\nabla \times \psi\) and the global error \[ E(\bar A_ 1,\bar A_ 2)=\int_{\Omega}| \bar A_ 1-\bar A_ 2|^ 2dx=\sum^{N}_{i=1}E_ i(\bar A_ 1,\bar A_ 2), \] where it is supposed that the domain \(\Omega\) is partitioned into N finite elements. To accelerate the convergence of the finite element process, the authors propose a selective refinement procedure of the finite element mesh in order to only refine the elements which have led to the largest errors \(E_ i(\bar A_ 1,\bar A_ 2)\). This refinement uses the triangulation algorithm due to Delaunay based on the technique of Voronoï polygons. (In this way, I recommend the excellent book by \textit{F. P. Preparata} and \textit{M. I. Shamos}: Computational Geometry, Springer Verlag (1985)). The results are illustrated on a L-shaped problem. It is worth to note that the computational requirements of the new algorithm are relatively low.