Machine learning-assisted precision inverse design research of ternary cathode materials: A new paradigm for material design

The Li+ diffusion rate directly affects the cathode rate performance, and it is inefficient to precision design cathode materials with excellent rate performance using the Edison approach method. Here, a new paradigm for the precision design of ternary cathode materials is exploited. The data of Ni-Co-Mn ternary (NCM) cathode materials doped with Li sites and transition metal (TM) sites, respectively, were extracted from publications, and the model Gradient Boosted Regression (GBR), which can accurately reveal the relationship between physical characterization variables and Li+ diffusion rate, was trained. Subsequently, the inverse design of the synthetic experimental parameters was carried out based on the desired target Li+ diffusion rate with the GBR model and particle swarm optimization (PSO) algorithm. A global search of the crystal structure is then performed using the Universal Structure Predictor: Evolutionary Xtallography (USPEX) code based on the parameters of the reverse design. Finally, first-principle calculations are performed to verify Li+ diffusion rate of the searched structures. The theoretical calculations show that the Li+ diffusion rates of the designed materials Ce-NCM and Li/Ni@CeNCM are 8.66 x 10 9 cm2/s, and 9.67 x 10 9 cm2/s, respectively, which are better than the target values (1.23 x 10 10 cm2/s). The density functional theory (DFT) calculations of charge transfer density indicate that moderate Li/Ni mixing induces a built-in electric field, which facilitates Li + diffusion in the NCM cathode materials. This work demonstrates the potential of accurate inverse design of ternary cathode materials, advances the research process of ternary cathode materials, and provides a reference for the design of cathode materials and its counterparts. This work will open new avenues for designing cathode materials and counterparts, potentially revolutionizing traditional trial-and-error experiments.