The correlation factor approach: Combining density functional and wave function theory

The electron-centric approach to the exchange-correlation energy

The Kohn-Sham theory addresses the challenge of representing the kinetic energy by re-quantizing density functional theory at a level of non-interacting electrons. It transforms the many-electron problem into a fictitious non-interacting electron problem, with the many-electron effects concealed within the exchange-correlation (XC) energy, which is expressed in terms of the electron density ρ(r). Unlike the wave function, ρ(r) can be viewed as a classical quantity, and expressing the XC energy in terms of it circumvents the need for correlated wave functions. In this work, we once again employ the re-quantization strategy and determine the XC energy using a local one-particle Schrödinger equation. The ground-state eigenfunction of the corresponding Hamiltonian is a reference point (r) dependent orbital φr,σ(u, σ') which is subsequently used to generate the XC hole and the XC energy. The spin coordinate is denoted by σ and u is the electron-electron separation. The one-particle equation for φr,σ(u, σ') includes a local potential vr,σ(u, σ') that we approximate using two simple physical constraints. We assess the approximation by applying it to the helium iso-electronic series, the homogeneous electron gas, and the dissociation of the hydrogen molecule.

Generating Exchange-Correlation Functionals with a Simplified, Self-Consistent Correlation Factor Model

We focus on the spherically averaged exchange-correlation hole ρ_XC(r, u) of density functional theory, which describes the reduction in the electron density at a distance u due to the reference electron localized at position r. The correlation factor (CF) approach, where the model exchange hole ρ_X^model(r, u) is multiplied by a CF (f_C(r, u)) to yield an approximation to the exchange-correlation hole ρ_XC(r, u) = f_C(r, u) ρ_X^model(r, u), has proven to be a powerful tool for the development of new approximations. One of the remaining challenges within the CF approach is the self-consistent implementation of the resulting functionals. To address this issue, here we propose a simplification of the previously developed CFs such that self-consistent implementations become feasible. As an illustration of the simplified CF model, we develop a new meta-GGA functional, and using only a minimum of empiricism, we provide an easy derivation of an approximation that is of an accuracy similar to more involved meta-GGA functionals.

The fourth-order expansion of the exchange hole and neural networks to construct exchange–correlation functionals

The curvature Q σ of spherically averaged exchange (X) holes ρX, σ(r, u) is one of the crucial variables for the construction of approximations to the exchange–correlation energy of Kohn–Sham theory, the most prominent example being the Becke–Roussel model [A. D. Becke and M. R. Roussel, Phys. Rev. A 39, 3761 (1989)]. Here, we consider the next higher nonzero derivative of the spherically averaged X hole, the fourth-order term T σ. This variable contains information about the nonlocality of the X hole and we employ it to approximate hybrid functionals, eliminating the sometimes demanding calculation of the exact X energy. The new functional is constructed using machine learning; having identified a physical correlation between T σ and the nonlocality of the X hole, we employ a neural network to express this relation. While we only modify the X functional of the Perdew–Burke–Ernzerhof functional [Perdew et al., Phys. Rev. Lett. 77, 3865 (1996)], a significant improvement over this method is achieved.

Solid-state performance of a meta-GGA screened hybrid density functional constructed from Pauli kinetic enhancement factor dependent semilocal exchange hole

The semilocal form of the exchange hole is highly useful in developing non-local range-separated hybrid density functionals for finite and extended systems. The way to construct the conventional exact exchange hole model is based on either the Taylor series expansion or the reverse engineering technique from the corresponding exchange energy functional. Although the latter technique is quite popular in context of generalized gradient approximation (GGA) functionals, the same for the meta-GGA functionals is not so much explored. Thus, in this study, we propose a reverse-engineered semilocal exchange hole of a meta-GGA functional, that depends only on the meta-GGA ingredient α (also known as the Pauli kinetic energy enhancement factor). The model is used subsequently to design the short-range-separated meta-GGA hybrid density functional. We show that the present method can be successfully applied for several challenging problems in the context of solids, especially for which the GGA based hybrid fails drastically. This assessment proves that the present functional is quite useful for materials sciences. Finally, we also use this method for several molecular test cases, where the results are also as comparative as its base semilocal functional.

The factorization ansatz for non-local approximations to the exchange-correlation hole

Among the various types of approximations to the exchange-correlation energy (E_XC), the completely non-local approach is one of the lesser explored approximation schemes. It has not yet reached the predictive power of the widely used generalized gradient approximations, meta-generalized gradient approximations, hybrids, etc. In non-local functionals pursued here, the electron density at every point in space is employed to express the exchange-correlation energy per particle ϵ_XC(r) at a given position r. Here, we use the non-local, spherical-averaged density ρ(r,u)=∫dΩ_u4πρ(r+u) as a starting point to construct approximate exchange-correlation holes through the factorization ansatz ρ_XC(r, u) = f(r, u)ρ(r, u). We present upper and lower bounds to the exchange energy per particle ϵ_X(r) in terms of ρ(r, u). The factor f(r, u) is then designed to satisfy various conditions that represent important exchange and correlation effects. We assess the resulting approximations and find that the complex, oscillatory structure of ρ(r, u) makes the construction of a corresponding f(r, u) very challenging. This conclusion, identifying the main issue of the non-local approximation, is supported by a detailed analysis of the resulting exchange-correlation holes.

Constructing and representing exchange-correlation holes through artificial neural networks

One strategy to construct approximations to the exchange-correlation (XC) energy E_XC of Kohn-Sham density functional theory relies on physical constraints satisfied by the XC hole ρ_XC(r, u). In the XC hole, the reference charge is located at r and u is the electron-electron separation. With mathematical intuition, a given set of physical constraints can be expressed in a formula, yielding an approximation to ρ_XC(r, u) and the corresponding E_XC. Here, we adapt machine learning algorithms to partially automate the construction of X and XC holes. While machine learning usually relies on finding patterns in datasets and does not require physical insight, we focus entirely on the latter and develop a tool (ExMachina), consisting of the basic equations and their implementation, for the machine generation of approximations. To illustrate ExMachina, we apply it to calculate various model holes and show how to go beyond existing approximations.