Domain decomposition methods in scattered data interpolation with conditionally positive definite radial basis functions

Scattered data interpolation using conditionally positive definite radial basis functions (RBFs) requires the solution of a symmetric saddle-point system. Based on an approximation of the system matrix as a hierarchical matrix, we solve the system iteratively using the GMRes algorithm and a domain decomposition preconditioner. The novelty of our work lies in the proposed solution of the subdomain problems using the nullspace method with an orthogonal basis represented as a sequence of Householder reflectors. The resulting positive definite subdomain systems are solved either directly or using an inner GMRes iteration with H-Cholesky preconditioning. Numerical tests demonstrate the effectiveness of this solution process for up to N = 160000 centers in two and three dimensions. (C) 2018 Elsevier Ltd. All rights reserved.