The approximate Dirichlet domain decomposition method. II: Applications to 2nd-order elliptic B.V.P.s (Q1180346)

[For part I see ibid. 47, No. 2, 137-151 (1991; Zbl 0741.65091).] The Dirichlet problem for the equation \(-\Delta u=f\) in an \(L\)-shaped region serves as a model problem for construction of spectrally equivalent preconditioners \(B\) for the given grid operator \(A=-\Delta_ h\). The block \(B_{22}\) is constructed on the basis of inner multigrid iterations in subregions and can be written in the form \(B_{22}=A_{22}(I_ 2-Z_ 2)^{-1}\) where \(Z_ 2\) is the reduction operator in inner iterations, \(\| Z_ 2\|_{A_{22}}<1\). Several similar approximations for \(S_ 1(A)\) are suggested and analyzed provided that spectral equivalence in a generalized sense of operators \(A_{11}\) and \(\Lambda_{11}4\) takes place: \(\sigma_ c\Lambda_{11}\leq A_{11}\leq \sigma_ 1\Lambda_{11}\), \(\sigma_ 1/\sigma_ 0=0(\ln^ 2h)\). There is also a special multigrid procedure of transforming the basis. The resulting iterative algorithms allow to find the solution of the given system \(Au=f\) with \(h\)-accuracy at the cost of \(O(h^{-2}|\ln h|^ k\)) (where \(k=2\), \(k=3\)) arithmetic operations. It means that these algorithms are almost optimal. Numerical examples involve the case of \(64\times 10^ 3\) unknowns and 4 processors.