Atomistic modeling of lithium materials from deep learning potential with ab initio accuracy

Abstract

Lithium has been paid great attention in recent years thanks to its significant applications for battery and lightweight alloy. Developing a potential model with high accuracy and efficiency is important for theoretical simulation of lithium materials. Here, we build a deep learning potential (DP) for elemental lithium based on a concurrent-learning scheme and DP representation of the density-functional theory (DFT) potential energy surface (PES), the DP model enables material simulations with close-to DFT accuracy but at much lower computational cost. The simulations show that basic parameters, equation of states, elasticity, defects and surface are consistent with the first principles results. More notably, the liquid radial distribution function based on our DP model is found to match well with experiment data. Our results demonstrate that the developed DP model can be used for the simulation of lithium materials.