Linear scaling conjugate gradient density matrix search: Implementation, validation, and application with semiempirical molecular orbital methods

TL;DRAbstract

Given the need to extend current quantum-chemical treatments to large molecules with thousands of atoms, the main goal of this work was the implementation of linear scaling approaches in the context of semiempirical molecular orbital theory. At the outset of this project, three such approaches had already been suggested in the literature. We have chosen the conjugate gradient density matrix search (CG-DMS) because it employs reliable and well-established minimization procedures and offers a transparent route towards linear scaling through the use of cutoffs in the density matrix and the Fock matrix.<br>Three versions of the CG-DMS code have been implemented. The full-matrix versions with precomputation (ITERCG) and on-the-fly computation (DIRCG) of the required integrals serve mainly for testing purposes, whereas the sparse-matrix integral-direct version (DIRCGS) is designed for linear scaling production work on large molecules. DIRCGS employs the compressed sparse row format and subro

Chat with Paper

AI Agents for this Paper