In this article, we formulate and analyze a two-level preconditioner for optimized Schwarz and 2-Lagrange multiplier methods for PDEs with highly heterogeneous (multiscale) diffusion coefficients. The preconditioner is equipped with an automatic coarse space consisting of lowfrequency modes of approximate subdomain Dirichlet-to-Neumann maps. Under a suitable change of basis, the preconditioner is a 2 × 2 block upper triangular matrix with the identity matrix in the upper-left block. We show that the spectrum of the preconditioned system is included in the disk having center z = 1/2 and radius r = 1/2 - ∈, where 0 < ∈ < 1/2 is a parameter that we can choose. We further show that the GMRES algorithm applied to our heterogeneous system converges in O(1/∈) iterations (neglecting certain polylogarithmic terms). The number ∈ can be made arbitrarily large by automatically enriching the coarse space. Our theoretical results are confirmed by numerical experiments.
- Adaptive coarse space enrichment
- Coefficient dependent coarse space
- Dirichlet to Neumann generalized eigenproblem
- Domain decomposition
- Heterogeneous media
- Multiscale PDEs