The Galerkin gradient least-squares method (Q913465)

The Galerkin gradient least-squares method introduced by \textit{T. J. R. Hughes} and \textit{A. Brooks} [Finite element methods for convection dominated flows, Winter ann. Meet. ASME, New York 1979, AMD Vol. 34, 19- 35 (1979; Zbl 0423.76076) and Proc. 3rd int. Conf. Finite elements in flow problems, Banff/Alberta 1980, Vol. II, 283-292 (1980; Zbl 0446.76077)] has allowed to substantially improve the Galerkin approximations of some engineering problems like thin structures, incompressible media, fluid flows etc., where spurious oscillations, locking and other undesirable features appeared. By noting that the direct application of such a method to axisymmetric shell problems cannot overcome the dependence on the shell thickness, the authors analyze in this paper a simpler scale model equation, i.e., a singular diffusion problem given by \(\sigma^ 2u-\epsilon^ 2\Delta u=f.\) The direct applications of standard Galerkin method and Galerkin/least-squares method do not overcome the spurious oscillations which appear when the ratio \(\epsilon^ 3/\sigma^ 2\) becomes too small. To cure this shortcoming, the authors add to the Galerkin method a least-squares form of the gradient of the Euler-Lagrange equation; in this way, they obtain stability in the \(H^ 1\) seminorm instead of \(L_ 2\)-stability in previous works. This new method is analyzed in detail for the one-dimensional model and then generalized to the multi-dimensional equation. Finally, numerical experiments confirm the good stability and accuracy of this method.