An ILU preconditioner for nonsymmetric positive definite matrices by using the conjugate Gram-Schmidt process (Q817473)

The authors consider preconditioners for iterative methods to solve linear systems \(A x = b\), where \(A\) is a large sparse matrix. The Introduction contains a classification and characterization of preconditioners, especially the ILU factorization and the approximate inverse factorization methods. The incomplete factorization preconditioners are very efficient in reducing the number of iteration steps, but for arbitrary matrices the ILU could face breakdown. Otherwise, there are less effective approximate inverse factorization methods which guarantee the existence of preconditioners even for nonsymmetric positive definite matrices. Following \textit{M. Benzi}'s and \textit{M. Tuma}'s [Numer. Linear Algebra Appl. 10, No. 5--6, 385--400 (2003; Zbl 1071.65528)] idea, the authors develop an ILU factorization preconditioner based on the conjugate Gram-Schmidt process using a relation between the inverse approximate technique and the incomplete factorization. The existence of the new effective preconditioner is guranteed for any symmetric and nonsymmetric positive definite matrix. The results are confirmed by numerical experiments.