We consider additive two-level preconditioners, with a local and a global c
omponent, for the Schur complement system arising in non-overlapping domain
decomposition methods. We propose two new parallelizable local preconditio
ners. The first one is a computationally cheap but numerically relevant alt
ernative to the classical block Jacobi preconditioner. The second one explo
its all the information from the local Schur complement matrices and demons
trates an attractive numerical behaviour on heterogeneous and anisotropic p
roblems. We also propose two implementations based on approximate Schur com
plement matrices that are cheaper alternatives to construct the given preco
nditioners but that preserve their good numerical behaviour. Through extens
ive computational experiments we study the numerical scalability and the ro
bustness of the proposed preconditioners and compare their numerical perfor
mance with well-known robust preconditioners such as BPS and the balancing
Neumann-Neumann method. Finally, we describe a parallel implementation on d
istributed memory computers of some of the proposed techniques and report p
arallel performances. Copyright (C) 2001 John Wiley & Sons, Ltd.