On the spectra of sums of orthogonal projections with applications to parallel computing (Q2639599): Difference between revisions

For elliptic problems the Schwarz alternating method is a possibility to compute the solution on subdomains in a sequential fashion, but it is not attractive in parallel computing. A parallel version of this method is described in this paper by computing the iterates on all subdomains (or on all grids) simultaneously. In the concrete problem there is given a Hilbert space H with two inner products (\(\cdot,\cdot)\) and \(<\cdot,\cdot >\), respectively, \(f\in H'\). The task is to find \(u\in H\) with \((u,v)=<f,v>\) for all \(v\in H\). Let be \(V=\sum^{n}_{i=1}V_ i\subset H\) and \(V_ i\) closed subspaces of H, \(i=1,...,n\). The classical Schwarz alternating method decomposes the problem into a series of subproblems on subspaces \(V_ i\) \((=additive\) method). The parallel version of it is transformed into a multiplicative view. The authors describe an additive algorithm for computing the composite solution on a grid having one level of refinement, too. The convergence of these methods depends on properties of the spectrum of certain sums of orthogonal projections. Known results about those spectra are collected and complemented by the authors' own theorems. The case \(H=V_ 1+V_ 2\) (where \(V_ 1\cap V_ 2\neq \{0\}\) is possible) is discussed. Applications of the theory to finite element spaces are given.