Publications

Material scientists and condensed matter physicists have long been divided on the issue of choosing the conceptual framework for explaining why open-shell transition-metal oxides tend to be insulators, whereas otherwise successful theories such as DFT often predict them to be (false) metals. Strong correlation becomes the recommended medicine. We point out that strong correlation can be mitigated by allowing DFT to lower the energy by breaking structural, magnetic or dipolar symmetries. Such local motifs are observed experimentally by local probes beyond the 'average structure' determined by X-Ray diffraction. Observed broken symmetries can arise from slow fluctuations that persist over the observation time or longer. The surprising fact is that when symmetry breaking motifs are used as input to electronic structure calculations, false metals are converted into real insulators without the recommended medicine of strong correlation. Consistently, DFT calculations that show energy lowering symmetry breaking correct most cases where DFT, even with advanced exchange-correlation functionals, previously missed the correct metal vs insulator designation. Total energy calculations distinguish systems that support energy-lowering symmetry breaking from those that do not. This approach distinguishes between paramagnetic insulating and metallic phases and shows mass enhancement in Mott metals. The reason is that symmetry breaking removes many of the degeneracies that exist in a symmetry-unbroken system, reducing significantly the need for strong correlation. If one chooses to ignore symmetry breaking, the persistent degeneracies often call for strong correlation treatment. Thus, symmetry breaking transforms strong to normal correlation and false metals to true insulators. This view sheds light on the historic controversy between Mott and Slater that still reverberates today.

Transition metal oxide materials are of great utility, with a diversity of topical applications ranging from catalysis to electronic devices. Because of their widespread importance in materials science, there is increasing interest in developing computational tools capable of reliable prediction of transition metal oxide phase behavior and properties. The workhorse of materials theory is density functional theory (DFT). Accordingly, we have investigated the impact of various correlation and exchange approximations on their ability to predict the properties of NiO using DFT. We have chosen NiO as a particularly challenging representative of transition metal oxides in general. In so doing, we have provided validation for the use of the r2SCAN density functional for predicting the materials properties of oxides. r2SCAN yields accurate structural properties of NiO and a local spin moment that notably persists under pressure, consistent with experiment. The outcome of our study is a pragmatic scheme for providing electronic structure data to enable the parameterization of interatomic potentials using state-of-the-art artificial intelligence (AI) and machine learning (ML) methodologies. The latter is essential to allow large scale molecular dynamics simulations of bulk and surface materials phase behavior and properties with ab initio accuracy.

Abstract The errors in atomization energies ( AE ) of molecules have long been used to measure the errors of wavefunction or density functional methods for electronic structure calculations. In particular, the G3 set of Pople and collaborators (for sp -bonded molecules from the first rows of the periodic table) has become a standard benchmark for such methods. But the mean absolute error of AE tends to increase with increasing number N at of atoms in a molecule. In fact, AE is an extensive variable, which diverges as N at →∞. Here, as did Savin and Johnson 2015, we define an intensive atomization energy, IAE = AE / N at or atomization energy per atom, which tends to the finite cohesive energy (per atom) of a large cluster or solid ( N at →∞). We find that the mean absolute error of the G3 molecular IAE from accurate density functionals remains close to 1 kcal/mol as the average molecular size increases. This makes it possible to estimate in advance the magnitude of the error in AE for a molecule similar to most of those in the G3 set. It also allows us identify the G3 “outlying molecules”, and to more directly compare the accuracy of a given functional for different kinds of molecules (such as those containing transition-metal atoms) to that for G3-type molecules, by removing the otherwise-uncontrolled size factor. Finally, we point out that the familiar concept of “chemical accuracy” needs to be qualified.

In density functional theory, traditional explicit density functionals such as the local density approximation and generalized gradient approximations cannot accurately predict the band gap of solids for a fundamental reason: They lack the exchange-correlation derivative discontinuity. By comparing Kohn-Sham and generalized Kohn-Sham calculations, we here show that the nonempirical meta-generalized-gradient-approximation (meta-GGA) TASK from Aschebrock and K\"ummel [Phys. Rev. Res. 1, 033082 (2019)] predicts the right gaps for the right reason, i.e., as a combination of a proper Kohn-Sham gap and a substantial derivative discontinuity contribution. For many materials from small-gap semiconductors to large-gap insulators, the proper band gap is thus obtained. We further study a group of metal-halide perovskites for which the band gap is notoriously hard to predict. For these materials, TASK yields band gaps very similar to the nonlocal screened hybrid Heyd-Scuseria-Ernzerhof functional, yet at a fraction of the hybrid functional's computational cost. We discuss the influence of correlation functionals, and open questions in the comparison of calculated band gaps with experimental ones.

In the Kohn–Sham density functional theory of ground-state electronic structure, only the exchange–correlation energy Exc must be approximated. Although Exc is not typically a large component of the total energy, it is the principal ingredient of the glue that binds atoms together to form molecules and solids. To illustrate this fact, we present self-consistent results for atomization energies of molecules and for surface energies and work functions of jellium, calculated within the “Hartree” approximation, which neglects Exc. The Hartree world displays weak bonding between atoms, low or negative surface energies, and work functions that are close to zero. Other aspects of the Hartree world can be deduced from known size–effect relationships. The mechanism behind the glue role of exchange and correlation is the suppression of Hartree charge fluctuations. © 2000 John Wiley & Sons, Inc. Int J Quant Chem 77: 814–818, 2000

The local spin-density (LSD) functional and Perdew–Wang 91 (PW91) generalized gradient approximations to atomization energies of molecules are investigated. We discuss the coupling-constant dependence of the atomization energy and why exchange errors of the functionals are greater than exchange–correlation errors. This fact helps to justify hybrid schemes which mix some exact exchange with density functional approximations for exchange and correlation. It is shown that the biggest errors in the atomization energies occur when there is a strong interaction between different electron pairs, which vanishes upon atomization. We argue that the amount of exchange character of a molecular property, such as the atomization energy, depends on the property itself. We define an exact mixing coefficient b, which measures this exchange character, and show that both LSD and PW91 typically overestimate this quantity. Thus, nonempirical hybrid schemes which approximate this quantity by its LSD or PW91 value typically do not improve the exchange–correlation energy. © 1997 John Wiley & Sons, Inc. Int J Quant Chem 64: 285–295, 1997