Partition density-functional theory

Abstract

Partition density functional theory is a formally exact procedure for calculating molecular properties from Kohn-Sham calculations on isolated fragments, interacting via a global partition potential that is a functional of the fragment densities. An example is given and consequences discussed. PACS numbers:Kohn-Sham density functional theory (KS-DFT) [1,2] is an efficient and usefully accurate electronic structure method, because it replaces the interacting Schrödinger equation with a set of single-particle orbital equations. Calculations with several hundred atoms are now routine, but there is always interest in much larger systems. Many such systems are treated by a lower-level method, such as molecular mechanics, but a fragment in which a chemical reaction occurs must still be treated quantum mechanically. A plethora of such QM/MM approaches have been tried and tested, with varying degrees of success [3]. These are often combined with attempts at orbital-free DFT, which avoids the KS equations, but at the cost of higher error and unreliability.On the other hand, partition theory (PT) [4,5] combines the simplicity of functional minimization with a density optimization to define fragments (such as atoms) within molecules, overcoming limitations of earlier approaches to reactivity theory [6,7]. While there are now many definitions of, e.g., charges on atoms, none have the generality of PT and the associated promise of unifying disparate chemical concepts. However, previous work on PT has been either formal [4,5] or for two atom systems [8,9].In this paper, we unite KS-DFT with PT to produce an algorithm that allows a KS calculation for a molecule to be performed via a self-consistent loop over isolated fragments. Such a fragment calculation exactly reproduces the result of a standard KS calculation of the entire molecule. We demonstrate its convergence on a 12-atom example. This also shows that fragments can be calculated 'on the fly', as part of solving any KS molecular problem.Thus we present a formally exact framework within which existing practical approximations can be analyzed and, for smaller systems, compared with exact quantities. In practical terms, our method suggests new approximations that can, by construction, scale linearly[10] with the number of fragments (so-called O(N )), and allow embedding of KS calculations within cruder force-field calculations (QM/MM). It also suggests ways to improve XC approximations so as to produce correct dissociation of molecules [11].To understand the relation between DFT and PT, recall that the Hohenberg-Kohn theorem proves that for a given electron-electron interaction and statistics, the external (one-body) potential v(r) is a unique functional of the density n(r). The total energy can be written as:where F [n] is a universal functional, defined by the LevyLieb constrained search [12] over all antisymmetric wavefunctions Ψ yielding density n(r):whereT andV ee are the kinetic energy and Coulomb repulsion operators respectively. The KS equations are single-particle...

Accurate Reference Data for the Nonadditive, Noninteracting Kinetic Energy in Covalent Bonds

“…These methods, such as subsystem density-functional theory [1,2] (subsystem-DFT) and partition density-functional theory [3][4][5] (PDFT) are based around a real space partitioning of the total density into fragments such that,…”

Molecular properties via a subsystem density functional theory formulation: A common framework for electronic embedding

“…While this constraint can be avoided in the context of DFT-in-DFT embedding as shown recently by Elliot et al, 55,56 in WFT-in-DFT embedding it can only be relaxed in the DFT subsystems, 57 because wave-function-based methods can only provide accurate densities for systems with an integer number of electrons. In our general formulation, the fixed electron number approximation is applied to all subsystems, offering also the possibility of treating all subsystems with WFT.…”

Linear-Response Time-Dependent Embedded Mean-Field Theory

“…For example, methods based on localized molecular orbitals lead to complicated implementations for analytical gradients and properties, while many embedding methods place constraints on the subsystem particle numbers, spin state, and spatial extent of the excitation, or they neglect particle-number fluctuations between subsystems, or the environmental response to the excitation. Removing such constraints has motivated the recent development of embedding strategies that are formally exact in the description of subsystem interactions [25][26][27][28][29][30][31][32][33][34][35][36][37] and allow for particle-number fluctuations between subsystems via their description as open quantum systems. [35][36][37] Here, we introduce time-dependent embedded mean-field theory (TD-EMFT), a linear-response approach to describe excited electronic states using the EMFT framework.…”