Prediction of Li intercalation voltages in rechargeable battery cathode materials: Effects of exchange-correlation functional, van der Waals interactions, and Hubbard U

Quantitative predictions of the Li intercalation voltage and of the electronic properties of rechargeable battery cathode materials are a substantial challenge for first-principles theory due to the possibility of (1) strong correlations associated with localized transition metal d electrons and (2) significant van der Waals (vdW) interactions in layered systems, both of which are not accurately captured by standard approximations to density functional theory (DFT). Here, we perform a systematic benchmark of electronic structure methods based on the widely used generalized-gradient approximation of Perdew, Burke, and Ernzerhof (PBE) and the new strongly constrained and appropriately normed (SCAN) meta-generalized-gradient approximation for battery cathode materials. Studying layered LixTiS2, LixNiO2, and LixCoO2, olivine LixFePO4, and spinel LixMn2O4, we compute the voltage, crystal structure, and electronic structure with and without extensions to incorporate onsite Hubbard interactions and vdW interactions. Within pure DFT (i.e., without corrections for onsite Hubbard interactions), SCAN is a significant improvement over PBE for describing cathode materials, decreasing the mean absolute voltage error by more than 50%. Although explicit vdW interactions are not critical and in cases even detrimental when applied in conjunction with SCAN, Hubbard-U corrections are still in general necessary to achieve reasonable agreement with experiment. We show that no single method considered here can accurately describe the voltage and overall structural, electronic, and magnetic properties (i.e., errors no more than 5% for voltage, volume, band gap, and magnetic moments) of battery cathode materials, motivating a strong need for improved electronic structure approaches for such systems.