Challenges for density functional theory in simulating metal-metal singlet bonding: a case study of dimerized VO2

TL;DR

This paper investigates the limitations of density functional theory in accurately modeling the complex structural, electronic, and magnetic properties of VO2's phase transition, highlighting the challenges in simulating metal-metal singlet bonds.

Contribution

The study demonstrates that common DFT functionals fail to predict V-V dimer lengths accurately and reveals the inherent difficulties in simulating VO2's phase transition using DFT methods.

Findings

DFT functionals do not accurately predict V-V dimer length

Spin-restricted DFT overestimates V-V bond strength

Spin-symmetry-breaking DFT underestimates V-V bond strength

Abstract

VO2 is renowned for its electric transition from an insulating monoclinic (M1) phase characterized by V-V dimerized structures, to a metallic rutile (R) phase above 340 Kelvin. This transition is accompanied by a magnetic change: the M1 phase exhibits a non-magnetic spin-singlet state, while the R phase exhibits a state with local magnetic moments. Simultaneous simulation of the structural, electric, and magnetic properties of this compound is of fundamental importance, but the M1 phase alone has posed a significant challenge to density functional theory (DFT). In this study, we show none of the commonly used DFT functionals, including those combined with on-site Hubbard U to better treat 3d electrons, can accurately predict the V-V dimer length. The spin-restricted method tends to overestimate the strength of the V-V bonds, resulting in a small V-V bond length. Conversely, the spin-symmetry-breaking method exhibits the opposite trends. Each bond-calculation method underscores one of the two contentious mechanisms, i.e., Peierls or Mott, involved in the metal-insulator transition in VO2. To elucidate the challenges encountered in DFT, we also employ an effective Hamiltonian that integrates one-dimensional magnetic sites, thereby revealing the inherent difficulties linked with the DFT computations.