Extension of inverse-based dropping techniques for ILU preconditioners
In this paper we present a safe and efficient way of dropping for those class of ILU factorizations that their factors are extracted as by-products of AINV process. The drop tolerance parameters of W and Z factors of AINV are selected the same as the drop tolerance parameters of L and U factors respectively. The infinity norms of the columns of W and Z are used to drop entries of L and U. The new dropping technique on L and U is based on monitoring the information of the inverses of L and U. Different dropping strategies for W and Z affect the efficiency of dropping for L and U.
Use and reproduction:
All rights reserved