Comparison of computational and experimental inorganic crystal structures

Abstract

The chemical and physical properties of inorganic compounds predicted from density functional theory (DFT) calculations depend on their crystal structures, with the underlying assumption being made that the computed lattice parameters are accurate. This assumption is evaluated by comparing the lattice parameters computed in the Materials Project using PBE-GGA with the experimental values reported for 10,033 compounds in Pearson's Crystal Data. On average, the computed lattice parameters are overestimated relative to the experimental parameters by 1–2% for the cell lengths and 4% for the cell volumes. The discrepancy can largely be traced to the neglect of corrections due to London dispersion forces in many DFT calculations, and is especially severe for layered crystal structures in trigonal systems. Within the experimental crystal data alone, the uncertainty in unit cell volumes for multiple entries of the same compound can range between 0.1 and 1%, which is at least an order of magnitude greater than the stated uncertainties for individual entries. The stability of 36,857 compounds computed in the Materials Project was also assessed by comparing to experimental structures reported in Pearson's Crystal Data. This analysis suggests that predicted compounds having a formation energy of >200 meV atom-1 above the convex Hull should not necessarily be ruled out as synthetically inaccessible.