Algebraic study of multigrid methods for symmetric, definite problems (Q1098229)

The convergence of the algebraic multigrid method with pre-smoothing is shown to follow from a single condition involving the orthogonal projection onto the complement of \(H_{k-1}\) in \(H_ k\) (from the sequence of Hilbert spaces connected with the different multigrid levels). The mentioned condition is next shown to be replaceable by an approximation and a smoothing assumption. Equivalent formulations of both assumptions are derived, and the smoothing assumption is proved to hold for Jacobi-like iterations \((u_{k+1}=u_ k-\omega B^{-1}(Au_ k- f),\rho (B^{-1}A)=1\), \(\omega =3/2\) being optimal), SOR, Gauss-Seidel, steepest descent. [Somewhat strange is the late appearance of this paper already cited in \textit{W. Hackbusch}'s book [Multi-grid methods and applications (1985; Zbl 0595.65106)].