Abstract
We present an approach to preconditioning for large, relatively dense linear systems and verify the validity of our method. We restrict the target of our method to Molecular Orbital (MO) calculations. Sparse Approximate Inverse (SAI) is typically less effective at accelerating the convergence and requires a huge computational cost in its construction when a large number of nonzero entries are kept in the approximate inverse matrix. We explain a construction of Block SAI and a cutoff strategy to reduce the number of nonzero elements, and investigate the efficiency of a cutoff strategy and Block SAI.
