A posteriori error estimation of hierarchical type for the Schrödinger operator with inverse square potential (Q478097)

The authors consider a two-dimensional Schrödinger equation the potential of which contains singularities of inverse square type, whereas the domain is polygonal and from any of its boundary pieces to the next the type may change from Dirichlet to Neumann or vice-versa. Moreover, the singularities may be located also at corner points of the boundary. For the discretization, a family of one-parameter graded meshes is proposed along with linear finite elements. The theoretical investigation of well-posedness, regularity and optimal order convergence is undertaken in a weighted variant of \(H_1\) and based on the work of \textit{H. Li} and \textit{V. Nistor} [J. Comput. Appl. Math. 224, No. 1, 320--338 (2009; Zbl 1165.65079)]. For the error estimates, a second finite element space is introduced with quadratic elements disappearing in the singular points. These error estimates are shown to be cheap in the sense that the corresponding stiffness matrix is proved to be spectrally equivalent to its diagonal. This allows, along with the algebraic multigrid of Bank's PLTMG for solving the basic linear finite element equations, to use some Jacobi-preconditioned steps of CG for the second order finite element equations. The authors also prove that the approximate error computed is efficient and reliable: it is asymptotically identical to the true error in the weighted energy norm. A series of numerical experiments concludes the paper. These not only illustrate the theoretical results but also provide comparison with the adaptively driven refinement of PLTMG. The latter turns out to be somewhat better but theoretical results backing this fact are lacking or correspond to basic cases like that of \textit{W. Dörfler} [SIAM J. Numer. Anal. 33, No. 3, 1106--1124 (1996; Zbl 0854.65090)].