Is the Pollution Effect of the FEM Avoidable for the Helmholtz Equation Considering High Wave Numbers? (Q4377544)

The authors consider numerical solution of the Helmholtz equation \(-\Delta u-k^2u=f\) on a region exterior to some domain \(D\) and interior to a large ball \(B\). Dirichlet conditions are assumed on \(D\) and Robin conditions on \(B\). It is well known that the Galerkin finite element method suffers from progressively greater loss of accuracy for greater \(k\). This loss of accuracy is called the pollution effect and can be viewed as a loss of stability because the constant in the inf-sup condition depends on \(k\), becoming small when \(k\) becomes large. The authors describe their results in the following manner: \(\bullet\) ``In one dimension, the pollution effect can be eliminated completely by a suitable modification of the discrete bilinear form.'' \(\bullet\) ``In two dimensions, the pollution effect can be reduced substantially, but cannot be avoided in principle. Hence, one-dimensional results are not fully representative for the two- and higher-dimensional cases.'' The authors define a generalized finite element method (GFEM) as a difference scheme with the same sparsity and symmetry properties as the usual Galerkin finite element method, but whose coefficients are otherwise arbitrary. In the one-dimensional (ordinary differential equation) case, they exhibit a difference scheme (with tridiagonal stiffness matrix) whose error estimates do not deteriorate for large \(k\) so that the method is pollution-free. The proofs are based on showing that a certain class of right hand sides yield GFEM solutions which interpolate the exact solution, and that homogeneous solutions are included among the ones that are exactly interpolated. They illustrate their theory with numerical examples. In two dimensions, the class of homogeneous solutions is infinite, making it impossible to construct an analog to the pollution-free one-dimensional GFEM. It is possible to choose a GFEM that has minimal pollution. The authors examine a model problem with rectangular mesh elements and show that no choice of GFEM with a constant nine-point stencil can be pollution-free. This proof is based on weighted-norm estimates of a certain homogeneous solution.