Abstract: The electron localization function (ELF) is a universal measure of electron localization that allows for, e.g., an effective characterization of physical bonds in molecular and solid state systems. In the context of the widely used Kohn-Sham density-functional theory (KS-DFT) and its generalizations, ELF is given in terms of the single-particle electron density as well as the non-interacting kinetic energy density (KED) of the KS system. Starting from the notion of an edge electron gas put forth by Kohn and Mattsson, we here use an exactly soluble, strongly correlated few-electron model of a harmonically confined electron gas in order to parameterize the positive-definite non-interacting KS KED in terms of the density and its reduced second-order gradients. We arrive at a simple, yet generally applicable functional approximation to ELF expressed in the electron density and its derivatives. To demonstrate the validity of our approach, we use the obtained parameterization to perform topological analysis of the bonds in solid Al and Si, and to study physical and chemical adsorption of graphene on a Ni surface. We find that while the expression does not provide a quantitatively accurate approximation of absolute ELF values, the most essential qualitative features are captured. Hence, the expression is useful in contexts where the electron density is available, but not the KS orbitals or the KED, and one desires a qualitative picture of the electron localization that mimics ELF.
Introduction
In computational chemistry and materials science, the Electron Localization Function (ELF) is the standard tool for visualizing chemical bonds and atomic shell structures. However, calculating the standard ELF requires the non-interacting kinetic energy density (KED), which is derived from individual Kohn-Sham orbitals. This requirement creates a significant bottleneck in high-throughput materials databases—which often store only total electron density to save space—and in orbital-free Density Functional Theory (DFT) simulations where orbitals are never calculated.
This work addresses this gap by developing a "semi-local" functional that approximates the ELF using only the electron density ($n$), its gradient ($\nabla n$), and its Laplacian ($\nabla^2 n$).
Theoretical Framework and Methodology
The researchers utilized an "exactly soluble" model system to derive a general approximation: the Harmonic Oscillator (HO) model, representing a few-electron gas confined by a parabolic potential. By deriving the exact mathematical relationship between KED and density gradients for simple systems, they established a template for more complex scenarios.
Through a least-squares fit on this model, the authors determined parameters that allow the functional to work for many-electron systems. A central insight of this approach is the importance of the term $s^2 - q$ (the difference between the squared reduced gradient and the Laplacian), which acts as a sensitive indicator of electronic shell structure and confinement.
Application and Validation
The new functional was validated against three distinct physical systems to test its ability to identify different bonding types:
- Bulk Aluminum (fcc Al): To evaluate metallic bonding.
- Bulk Silicon (Diamond Si): To evaluate strong covalent ($sp^3$) bonding.
- Graphene on Nickel (Ni): To test the functional's performance at complex interfaces involving physical and chemical adsorption.
Validation was performed using the Elk all-electron code, comparing the new approximation against the true orbital-based ELF.
Key Results and Discussion
The primary finding is that for bond visualization, topology is more critical than absolute numerical accuracy. While the functional may not reproduce the exact numerical value of the ELF, it successfully places peaks and valleys in the correct spatial locations to describe the chemistry accurately.
In Silicon, the functional clearly identified $sp^3$ hybridized bonds, matching the visual profile of the true ELF. In Aluminum, it correctly depicted the spread-out localization characteristic of metals. However, a limitation was noted in "void" regions of crystals where electron density is extremely low; here, the approximation tends to drop to zero while the true ELF remains moderate, as the functional struggles to map regions of minimal electron confinement.
Conclusion
This semi-local parameterization provides a practical tool for "seeing" bonds in systems where orbital information is missing. Its success in transferring a 1D harmonic model to 3D crystals suggests that the physics of electron confinement is remarkably universal. Future work will aim to improve quantitative accuracy and extend the model to handle more complex material interfaces and negative density gradient regions.
Summary updated