Benchmarking DFT surface energies with quantum Monte Carlo

Abstract

Previous work has demonstrated that for many materials, surface energies σ calculated by density functional theory (DFT) methods depend significantly on the exchange-correlation (xc) functional used. This could pose significant problems when using DFT for predicting structures of nanocrystals both in vacuum and on substrates. Here, we present initial results from a systematic study of σ, for a series of ionic materials using the VASP code and its projector augmented wave implementation of DFT. Calculations on LiH and MgO presented here were performed using four of the functionals available in VASP (LDA, PW91, PBE and RPBE) along with the recent Wu–Cohen modification of PBE. The results we present show that there is indeed a significant variation in σ due to differing functionals. Furthermore, we are able to ascertain which functionals gave the most accurate results by performing calculations of σ for LiH using diffusion quantum Monte Carlo methods that are generally accepted to be significantly more accurate, albeit more expensive, in calculating these quantities.