Data Reduction Schemes in Davidson Subspace Diagonalization for MR-CI

摘要

In this paper, we investigate several data reduction schemes to improve the computational efficiency in the multi reference configuration interaction (MR-CI) method, one of the main quantum chemical approaches for solving the electronic Schroedinger equation. The basic idea is to take advantage of the often relatively low accuracy requirements on the solution of the resulting large eigenvalue problem, whose dimension may reach several hundred millions or even more. We will discuss some approaches to reduce the amount of data to be accessed and to be transferred within the Davidson subspace diagonalization method. We also show experimental results achieved with the COLUMBUS code.