Single determinant approximation for ground and excited states with accuracy comparable to that of the configuration interaction

Abstract

It was realized from the early days of Chemical Physics (Rev. Mod. Phys. 35, 496 (1963)) that the energy $E_{HF}$ of the Slater determinant (SlDet) $|Φ_{HF}\rangle$, obtained by the single particle Hartree-Fock (HF) equation, does not coincide with the minimum energy of the functional $\langleΦ|H|Φ\rangle$ where $|Φ\rangle$ is a SlDet and $H$ is the many particle Hamiltonian. However, in most textbooks, there is no mention of this fact. In this paper, starting from a Slater determinant $|Φ\rangle$ with its spin orbitals calculated by the standard HF equation or other approximation, we search for the maximum of the functional $|\langleΦ^{\prime }|H|Φ\rangle|$, where $|Φ^{\prime }\rangle$ is a SlDet and $H$ is the exact Hamiltonian of an atom or a molecule. The element $|\langleΦ_{1}|H|Φ\rangle|$ with $|Φ_{1}\rangle$ the maximizing $|Φ^{\prime }\rangle$ gives a value larger than $\langleΦ|H|Φ\rangle$. The next step is to calculate the corresponding maximum overlap $\langleΦ_{2}|H|Φ_{1}\rangle|$ and subsequently $|\langleΦ_{n+1}|H|Φ_{n}\rangle|$ until $|\langleΦ_{m+1}|H|Φ_{m}\rangle -\langleΦ_{m-1}|H|Φ_{m}\rangle|\leq\varepsilon$, where $\varepsilon $ determines the required numerical accuracy. We show that the sequence $a_{n}=|\langleΦ_{n+1}|H|Φ_{n}\rangle|$ is ascending and converges. We applied this procedure for determining the eigenstate energies of several configurations of H$_{3}$, the Lithium atom, LiH and Be. After comparing our values with those of the configuration interaction we found that our deviations are in the range 10$^{-5}~$to $10^{-8}$ and the ground state energy is significantly below that of the standard HF calculations.

We want to recheck the CI results obtained by an old version of Gamess US