Abstract
Polyanionic sodium cathode materials exhibit promising electrochemical properties and high stability, making this chemical space worth exploring to enhance the performance of sodium-ion batteries. Given the vast chemical space, fast and efficient computational methods are needed, highlighting the need for machine learning (ML)-guided approaches. To support the ML-driven discovery, we have developed a comprehensive theoretical dataset comprising 113532 DFT-calculated structures with atomic charges and 184612 structures without atomic charges for four polyanion sodium-ion cathode materials: NaTMPO(4) (olivine), NaTMPO(4) (maricite), Na(2)TMSiO(4), and Na2.56TM1.72(SO4)3 , with transition metals TM ∈ {Fe, Mn, Co, Ni}. This dataset includes optimizations of structures, ab initio molecular dynamics simulation trajectories sampled at 1000 K, and structures generated from ML-driven molecular dynamics simulation at 1000 K using an active learning algorithm. The dataset contains both single and multiple-TM compositions, providing a diverse representation of cathode materials. Using cathode specific subsets of this dataset to train ML models, we demonstrate that we can create ML interatomic potentials that reproduces the DFT results of the polyanion sodium cathode materials.