An adaptive BDDC algorithm for three dimensional problems with an enhanced edge eigenvalue problem

Hyea Hyun Kim,Junxian Wang

2020-01-01

Abstract:An adaptive BDDC (Balancing Domain Decomposition by Constr aints) preconditioner is considered for three dimensional elliptic problems with coeffi cients of high contrast and random variation. For such problems, the classical primal unknown s in the BDDC preconditioner, such as values at subdomain vertices, and average values over sub domain edges or faces, are not enough to give a scalable coarse problem. Specially designe d eig nvalue problems related to the estimate of condition numbers are considered and proble matic eigenvectors are selected from the eigenvalue problems to enrich the set of primal unkn owns so as to make the resulting coarse problem in the preconditioner more scalable. In thre e dimensions, eigenvalue problems are formed on each face or edge nodal equivalence classes. Fo r the case of edges, proposed eigenvalue problems in previous studies are not satisfacto ry while those for faces perform very effectively. For the case of edges, the eigenvalues are not c learly separated in contrast to the case of faces and most eigenvectors are thus often selected i n practice, see [1–3]. The eigenvalue problems can be enhanced by utilizing prior selected p rimal constraints as proposed in [4]. In this paper, this new idea is adopted when forming edge eige nvalue problems. The primal unknowns at subdomain vertices and the adaptive primal unknow ns chosen from face eigenvalue problems are included when forming edge eigenvalue problem s. For the BDDC preconditioner with the enhanced edge eigenvalue problems, the bound of con dition numbers is shown to be dependent on the given tolerance value, that is used to selec t the problematic eigenvectors from face and edge eigenvalue problems, but independent of the co ntrast in the coefficients of the model problem. In addition, numerical results are provided to verify that the new edge eigenvalue problem gives a more optimal set of primal unknowns com pared to those in the previous studies.