No label defined (Q4354561)

In Section 2 we first give some motivations of the new finite element scheme, then describe the procedure to generate the shape functions for the approximate concentration \(u_h\) and the related currents \(J_h(u_h)= \nabla u_h-u_h\nabla\psi\) over each element. Some computational details are then addressed, first, we suggest how to avoid overflow problems due to the presence of diverging exponentials in the entries of the element stiffness matrix. Secondly, we examine the possibility, when working in the finite arithmetic of a computer machine, of a vanishing denominator in the shape functions coefficients. It is explained how this may happen in the extreme situation of a badly shaped triangle with a very large potential difference across it. Finally, we prove the property of weak continuity of the approximate current holding for the nonconforming Petrov-Galerkin scheme. Proofs of the fulfillment of the patch test conditions on a single fully connected node triangulation, as well as a numerical example on a highly unstructured grid, are given in Section 3. This allows to establish both convergence and robustness of the method. In the concluding Section 4 we present some numerical results relative to the solution of two model problems that are quite significant both in the specific context here considered of semiconductor device modelling and in the more general framework of singularly perturbed convection-diffusion equations.