New Quantum Chemistry Methods for Open-Shell Systems and Their Applications in Spin-Polarized Conceptual Density Functional Theory

Abstract

Motivated by our frustration with the lack of quantum chemistry methods for strongly-correlated open-shell systems, we develop quantitative methods for computing the electronic structure of such systems and qualitative tools for analyzing their chemical properties and reactivity. Specifically, we present a modern framework for performing sparse configuration interaction (CI) computations with arbitrary (Slater determinant) N-electron basis sets, using restricted or generalized spin-orbitals, and including computation of spin-polarized 1- and 2- electron reduced density matrices (RDMs). This framework is then used to implement the flexible ansätze for N-electron CI (FanCI) method more efficiently, via increased vectorization in the FanCI equations and use of sparse CI algorithms. We also extended the FanCI approach, including least-squares and stochastic optimization techniques, the computation of spin-polarized 1- and 2- electron RDMs, and transition energies (ionization potentials, electron affinities, and excitation energies). We use these tools to compare various open-shell CI methods and FanCI methods based on various antisymmetrized product of nonorthogonal geminals ansätze. To translate the vast amount of quantitative data present in the energies and (spin-polarized) density matrices of multiple open-shell states, we present a new, internally consistent and unambiguous framework for spin-polarized conceptual density-functional theory (SP-DFT) that reduces to a sensible formulation of spin-free CDFT in an appropriate limit. Using this framework, we were able to generalize the (non-spin-polarized) Parr function. We can also, using this framework, construct promolecules with proatoms having non-integer charges and multiplicities. Finally, we describe an equations-of-motion-based method for computing spin-polarized reactivity descriptors of a chemical system from only the ground state energy and the 1- and 2- electron RDMs from a single-point electronic structure computation, and show some benchmark computations for this method based on various CI and FanCI electronic structure methods.