Abstract
Recently Lu and Gao [J. Phys. Chem. Lett. 13, 33, 7762-7769 (2022)] published a new, rigorous density functional theory for excited states and proved that the projection of the kinetic and electron repulsion operators into the subspace of the lowest electronic states are a universal functional of the matrix density D(r). This is the first attempt to find an approximation to the multistate universal functional F[D(r)]. It is shown that F (i) does not explicitly depend on the number of electronic states and (ii) is an analytic matrix functional. The Thomas-Fermi-Dirac-von Weizsäcker model and the correlation energy of the homogeneous electron gas are turned into matrix functionals guided by two principles: That each matrix functional should transform properly under basis set transformations, and that the ground state functional should be recovered for a single electronic state. Lieb-Oxford-like bounds on the average kinetic and electron repulsion energies in the subspace are given. When evaluated on the numerically exact matrix density of LiF, this simple approximation reproduces the matrix elements of the Hamiltonian in the basis of the exact eigenstates accurately for all bond lengths. In particular the off-diagonal elements of the effective Hamiltonian that come from the interactions of different electronic states can be calculated with the same or better accuracy than the diagonal elements. Unsurprisingly, the largest error comes from the kinetic energy functional. More exact conditions that constrain the functional form of F are needed to go beyond the local density approximation.
Supplementary materials
Title
Supporting Information
Description
properties of matrix density functionals, comparison of paramagnetic and ferromagnetic
correlation energy, proof of Klein’s trace inequality, incorporation of exact constraints into
von-Weizsäcker functional, Lieb-Oxford-like lower bounds for kinetic and electron repulsion
energy of subspace.
Actions
Supplementary weblinks
Title
Github repository with code and data
Description
All calculations can be repeated on a desktop computer using the python scripts in the github repository.
Actions
View