Exploration of lithium-ion diffusion of LixCoO₂ system using machine learning interatomic potential

Lily Maysari Angraini^1,2,3*, Cheng-Rong Hsing⁴, Ching-Ming Wei³

¹Molecular Science and Technology, Taiwan International Graduate Program, Academia Sinica, Taipei, Taiwan
²International Graduate Program of Molecular Science Technology, National Taiwan University, Taipei, Taiwan
³Institute of Atomic and Molecular Sciences, Academia Sinica, Taipei, Taiwan
⁴Division of Natural Science, Center for General Education, Chang Gung University, Taoyuan City, Taiwan

* Presenter:Lily Maysari Angraini, email:lilyangraini@unram.ac.id

Lithium-ion diffusion is critical in determining the reaction velocity in lithium cobalt oxide’s (LiCoO₂) charge and discharge process. While ab initio molecular dynamics (AIMD) and classical molecular dynamics (MD) have been extensively used to investigate this material, AIMD can only deal with small system sizes and short-time scale simulations; meanwhile, classical MD has low accuracy issues. To address this shortcoming, we proposed using a machine learning interatomic potential (MLIP) method by combining the accuracy of AIMD and the computational efficiency of classical MD. Here, we applied moment tensor potential (MTP) to explore the properties of lithium diffusion (activation energy and diffusion coefficient (DLi)) in layered systems LixCoO₂ of different lithium-ion concentrations (x=0.400, 0.533, and 0.667). We obtain the well-transferability of MLIPs. The DLi estimated from MLIP was determined to be ~1/10⁸ in temperatures ranging from 300K to 360K and ~1/10⁷ cm²/s in temperatures ranging from 400K to 500K in all lithium concentrations. MLIP results have a similar trend with muon spectroscopy measurement in high lithium concentration at room temperature. Diffusivity increases with increasing lithium concentration at room temperature.

Keywords: LixCoO₂ system, diffusion coefficient, machine learning interatomic potential