Modeling Phase Transformation in Complex Oxides with Charge-Informed Interatomic Potential

Disordered rocksalt materials represent the most promising earth-abundant cathode materials for Li-ion batteries, potentially enabling the scaling of Li-ion energy storage to numerous TWh/year production. These advanced battery materials often comprise multiple elements and exhibit significant site disorder and structure complexity. Mn-rich DRX cathodes have observed a phase transformation from disorder to partial spinel-like order during charge/discharge cycling. To understand this phenomenon and reveal the underlying physics, we employed atomistic modeling with charge information derived from ab-initio calculations.<br/><br/>CHGNet is a novel machine-learning interatomic potential (MLIP) with atomic charge inference from magnetic moments. The explicit incorporation of magnetic moments allows CHGNet to learn and accurately represent the orbital occupancy of electrons, thereby enhancing its ability to describe both atomic and electronic degrees of freedom. We fine-tuned the pre-trained universal CHGNet within the Li-Mn-Ti-O-F chemical space using high-fidelity DFT calculations. The fine-tuned CHGNet was then applied to Li_1.1-xMn_0.8Ti_0.1O_1.9F_0.1 DRX system, using charge-informed molecular dynamics to investigate the structural ordering change, charge distribution, and electrochemical properties of the transformed DRX compounds.