Efficient preparation of TiO2 nanoparticle models using interatomic potentials

Abstract

Computational modeling has proven to be extremely useful for understanding how the morphology, size, and structure of TiO2 nanoparticles (NPs) affect their electronic properties and their usage in targeted applications (e.g., photocatalysis). Density functional theory (DFT) based calculations of NPs (on the order of hundreds to thousands of atoms) are, however, computationally highly demanding. Herein, we show that interatomic potentials (IPs) can provide a highly computationally efficient means to prepare NP structures which are sufficiently accurate to significantly reduce the computational cost of subsequent DFT calculations. We first compare the direct DFT optimization of faceted NPs directly cut from the anatase bulk crystal with the same calculation where the NP is preoptimized using four different IPs. We then establish the subsequent computational time saving for the respective complete DFT optimizations. We show that IP-based preoptimizing can greatly speed up DFT convergence, with speed-ups of 3×-10× for single point DFT energy evaluations. Moreover, as IP preoptimized NP structures can be closer to those of DFT energy minima, further speed-ups of 2× for DFT structure optimizations can be achieved. Finally, taking NPs derived from anatase spherical cuts, we show that IP-based molecular dynamics annealing gives rise to significant structural reconstruction with an associated high energetic stabilization, as confirmed by DFT calculations. Although similar results can be achieved using DFT tight binding methods, IP-based methods are 3-4 orders of magnitude faster and thus provide a particularly highly computationally efficient route to the preparation and design of large and diverse NP sets.