On the abstract theory of additive and multiplicative Schwarz algorithms (Q1893484)

The authors consider additive (ASM) and multiplicative (MSM) Schwarz methods for solving discrete (e.g. finite element) symmetric positive definite problems of the form: \(\text{Find }u \in V\) such that \(a(u,v) = \Phi (v)\) \(\forall v \in V\), where \(V\) is some fixed, finite-dimensional (e.g. finite element) Hilbert space. Starting with a splitting of the space \(V\) into \((J + 1)\) subspaces such that \(V = \sum^J_{j = 0} V_j\) and with \((J + 1)\) preconditioning forms \(b_j (.,.)\) for the original form \(a(.,.)\) restricted to the subspaces \(V_j\) (\(j = 0, \dots, J\)), one can construct the corresponding ASM and MSM. The authors first review the convergence results for the ASM and give then a quite pretty connection of the ASM and the MSM analysis to the classical matrix analysis of the classical Jacobi-/Richardson- and Gauss- Seidel-/SOR-like iterations. Finally, they present new bounds for the convergence rate \(\rho_{ms}\) of the appropriately scaled MSM in terms of the condition number \(\chi\) of the corresponding ASM-operator. The rigorous bound is given by the estimate \(\rho_{ms} \leq \sqrt {1- 1/(\log_2 (4J) \times \chi)}\).