“…They show that under certain conditions on the continuous problem and on the finiteelement mesh, the approximations are all good. They proposed to use L to construct a combinatorial preconditioner rather then apply multigrid to L. This proposal was based on the observation that over the last decade, several provably good combinatorial graph algorithms for constructing preconditioners for SDD matrices have been developed [46,21,20,7,43,44,18,45,29]. Some of them, as well as some combinatorial heuristics, have been shown to be effective in practice [15,25,36,38,19,35].…”