The \(hp\)-version of the BEM with geometric meshes in 3D (Q4371004)

The solutions of three-dimensional elliptic boundary value problems in polyhedral domains or on screens have special singular forms at corners and edges. If those problems are converted via the direct method into boundary integral equations, the solutions of the latter inherit those edge and corner singularities. The singularities disturb the convergence of numerical schemes, e.g. finite element or boundary element methods. Numerical schemes converge as fast as the solution is approximated by the chosen trial functions due to the quasioptimality of the Galerkin method. Thus the rate of convergence is determined by the regularity of the solution. Even for smooth data, the regularity is reduced by edge and corner singularities, and therefore the convergence of these procedures is much slower than in case of smooth domains. In this paper, it is shown that under proper assumptions usually satisfied in applications, the \(hp\)-version of the boundary element method (BEM) with a geometric mesh for the Dirichlet problem of the Laplacian has exponential rate of convergence with respect to the number of degrees of freedom. In Section 2 we consider the Dirichlet screen problem. The countably normed spaces introduced in Section 3 are motivated by the form of the corner-edge singularities. In Section 3 we define the mesh and the appropriate countably normed spaces. For the proof of our main result, the exponential convergence of the \(hp\)-version of the BEM for 3D problems, we define an equivalent countably normed space on a reference element using Cartesian coordinates. In Section 4 we describe some numerical results. For the entire collection see [Zbl 0877.00013].