A systematic strategy for simultaneous adaptive \(hp\) finite element mesh modification using nonlinear programming (Q5938292): Difference between revisions

It is already well established that when applying the finite element method, results that are both reliable and accurate can be achieved by means of adaptive \(hp\) methods, because they combine good approximation with the ability of mesh refinement based methods that allow to deal with singularities. The present paper reviewed here addresses the task of optimal mesh design in the adaptive \(hp\) context by formulating it as an optimization problem. This formulation aims either at (P1) minimizing error by generating a convenient mesh for a given computational cost, or at (P2) producing a mesh with the lowest computational cost for a given error level. Such formulation is made possible by applying a priori error estimates for the \(hp\) discretizations obtained by \textit{I. Babuska} and \textit{M. Suri} [RAIRO, Modélisation Math. Anal. Numér. 21, 199-238 (1987; Zbl 0623.65113)]. Some convenient relaxations allow P1 to be formulated as a nonlinear program. A further reduction in computational work is obtained by employing simplified analytical gradients derived from initial mesh calculations. Finally, an algorithm is implemented in order to actually sove P1 or P2, which uses the commercial calculation optimizer MINOS. The mesh design strategy thus developed in the paper is tested on three examples. Two of them are based on the Poisson problem, one with a sharp internal layer in the solution, the other with a competing bubble type feature. In all cases, \(hp\)-adaptive grids meeting the error/cost constraints were obtained in one step starting from a uniform coarse grid.