Optimal a priori estimates for higher order finite elements for elliptic interface problems (Q969296): Difference between revisions

The authors consider a homogeneous Dirichlet boundary value problem of an elliptic partial differential equation (PDE). The corresponding two- or three-dimensional domain \(\Omega\) is decomposed into two subdomains along an interface. In the PDE, a coefficient function is assumed to be constant in each subdomain, i.e., a discontinuity appears at the interface. The authors apply a finite element method with polynomials up to degree \(p\) in each element. In the Galerkin approach, the bilinear form of the weak formulation is replaced by an approximative bilinear form, which is more convenient to evaluate. Although the triangulations are given on the complete domain, a certain property of resolution is assumed with respect to a neighbourhood \(S_\delta\) of the interface. The width of the neighbourhood \(S_\delta\) is characterised by a parameter \(\delta > 0\). A priori error estimates are shown in the norms of \(L^2(\Omega)\) as well as \(H^1(\Omega)\). It follows that the optimal order \(p\) of convergence in \(H^1(\Omega)\) is only achieved for a high resolution of the interface satisfying \(\delta = O(h^{2p})\). However, the optimal order \(p\) is given in \(H^1(\Omega \backslash S_\delta)\) for the weaker condition \(\delta = O(h^{p+1})\). The proofs are based on the construction of a quasi-interpolation operator, where the resolution property of the triangulation is crucial. The authors prove error bounds for this quasi-interpolation, which depend on the triangulation only. The shown convergence results also yield statements on special cases, which involve approximations of the interface by interpolating splines of arbitrary degree. Finally, the authors present results of numerical simulations, where four examples in two space dimensions and an example in three space dimensions are examined. Just the two-dimensional example of an elastic interface problem is physically relevant, whereas the other test cases are chosen artificially. The theoretical results on the order of convergence are confirmed by the numerical simulations.