A block FSAI-ILU parallel preconditioner for symmetric positive definite linear systems

A novel parallel preconditioner for symmetric positive definite matrices is developed coupling a generalized factored sparse approximate inverse (FSAI) with an incomplete LU (ILU) factorization. The generalized FSAI, called block FSAI, is derived by requiring the preconditioned matrix to resemble a block-diagonal matrix in the sense of the minimal Frobenius norm. An incomplete block Jacobi algorithm is then effectively used to accelerate the convergence of a Krylov subspace method. The block FSAI-ILU preconditioner proves superior to both FSAI and the incomplete block Jacobi by themselves in a number of realistic finite element test cases and is fully scalable for a given number of blocks.