Analysis of FETI methods for multiscale PDEs (Q998636)

This paper is concerned with the convergence of a variant of the finite element tearing and interconnecting (FETI) methods for a model of elliptic problem with highly heterogeneous (multiscale) coefficients \(\alpha(x)\); in particular, \(\alpha\) can have strong variation within subdomains and/or jumps that are not aligned with the subdomain interfaces. Thanks to a minimisation and cut-off arguments, the authors show rigorously that for an arbitrary (positive) coefficient function \(\alpha \in L^\infty(\Omega)\) the condition number of the preconditioned FETI system can be bounded by \(C(\alpha)(1+\log(H/h))^2\), where \(H\) is the subdomain diameter and \(h\) is the mesh size. In addition to this, the coefficient \(C(\alpha)\) depends only on the coefficient variation in the vicinity of subdomain interfaces. This yields the important particular case when \(\alpha|_{\Omega_i}\) varies only midly in a layer \(\Omega_{i,\eta}\) of width \(\eta\) near the boundary of each of the subdomain \(\Omega_i\), then \(C(\alpha)={\mathcal{O}}((H/\eta)^2)\), independent of the variation of \(\alpha\) in the remainder \(\Omega_i\backslash \Omega_{i,\eta}\) of each subdomain and independent of any jumps of \(\alpha \) across subdomain interfaces. The authors show that this quadratic dependence of \(C(\alpha)\) on \(H/\eta\) can be relaxed to a linear dependence under stronger assumptions on the behavior of \(\alpha\) in the interior of the subdomains. Numerical tests justifying the theoretical results are presented.