Density functional theory: Foundations reviewed

• Published in: Physics reports Vol. 544; no. 2; pp. 123 - 239
• Main Authors: Kryachko, Eugene S., Ludeña, Eduardo V.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 10.11.2014
• DOI: 10.1016/j.physrep.2014.06.002
• ISSN: 0370-1573

Guided by the above motto (quotation), we review a broad range of issues lying at the foundations of Density Functional Theory, DFT, a theory which is currently omnipresent in our everyday computational study of atoms and molecules, solids and nano-materials, and which lies at the heart of modern many-body computational technologies. The key goal is to demonstrate that there are definitely the ways to improve DFT. We start by considering DFT in the larger context provided by reduced density matrix theory (RDMT) and natural orbital functional theory (NOFT), and examine the implications that N-representability conditions on the second-order reduced density matrix (2-RDM) have not only on RDMT and NOFT but, also, by extension, on the functionals of DFT. This examination is timely in view of the fact that necessary and sufficient N-representability conditions on the 2-RDM have recently been attained. In the second place, we review some problems appearing in the original formulation of the first Hohenberg–Kohn theorem which is still a subject of some controversy. In this vein we recall Lieb’s comment on this proof and the extension to this proof given by Pino et al. (2009), and in this context examine the conditions that must be met in order that the one-to-one correspondence between ground-state densities and external potentials remains valid for finite subspaces (namely, the subspaces where all Kohn–Sham solutions are obtained in practical applications). We also consider the issue of whether the Kohn–Sham equations can be derived from basic principles or whether they are postulated. We examine this problem in relation to ab initio DFT. The possibility of postulating arbitrary Kohn–Sham-type equations, where the effective potential is by definition some arbitrary mixture of local and non-local terms, is discussed. We also deal with the issue of whether there exists a universal functional, or whether one should advocate instead the construction of problem-geared functionals. These problems are discussed by making reference to ab initio DFT as well as to the local-scaling-transformation version of DFT, LS-DFT. In addition, we examine the question of the accuracy of approximate exchange–correlation functionals in the light of their non-observance of the variational principle. Why do approximate functionals yield reasonable (and accurate) descriptions of many molecular and condensed matter properties? Are the conditions imposed on exchange and correlation functionals sufficiently adequate to produce accurate semi-empirical functionals? In this respect, we consider the question of whether the results reflect a true approach to chemical accuracy or are just the outcome of a virtuoso-like performance which cannot be systematically improved. We discuss the issue of the accuracy of the contemporary DFT results by contrasting them to those obtained by the alternative RDMT and NOFT. We discuss the possibility of improving DFT functionals by applying in a systematic way the N-representability conditions on the 2-RDM. In this respect, we emphasize the possibility of constructing 2-matrices in the context of the local scaling transformation version of DFT to which the N-representability condition of RDM theory may be applied. We end up our revision of HKS-DFT by considering some of the problems related to spin symmetry and discuss some current issues dealing with a proper treatment of open-shell systems. We are particularly concerned, as in the rest of this paper, mostly with foundational issues arising in the construction of functionals. We dedicate the whole Section  4 to the local-scaling transformation version of density functional theory, LS-DFT. The reason is that in this theory some of the fundamental problems that appear in HKS-DFT, have been solved. For example, in LS-DFT the functionals are, in principle, designed to fulfill v- and N-representability conditions from the outset. This is possible because LS-DFT is based on density transformation (local-scaling of coordinates proceeds through density transformation) and so, because these functionals are constructed from prototype N-particle wavefunctions, the ensuing density functionals already have built-in N-representability conditions. This theory is presented in great detail with the purpose of illustrating an alternative way to HKS-DFT which could be used to improve the construction of HKS-DFT functionals. Let us clearly indicate, however, that although appealing from a theoretical point of view, the actual application of LS-DFT to large systems has not taken place mostly because of technical difficulties. Thus, our aim in introducing this theory is to foster a better understanding of its foundations with the hope that it may promote a cross-hybridization with the already existing approaches. Also, to complete our previous discussion on symmetry, in particular, spin-symmetry, we discuss this issue from the perspective of LS-DFT. Finally, in Section  6, we discuss dispersion molecular forces emphasizing their relevance to DFT approaches.