We develop an atomic cluster expansion (ACE) interatomic potential for lithium that accurately models both the solid and liquid phase and the corresponding melting point. The training data is obtained from 0 K density functional theory (DFT) and finite temperature ab initio molecular dynamics simulations of both solid and liquid Li. The ACE predicted properties for both phases obtained from molecular dynamics simulations are in close agreement with DFT and experimental data from the literature. The potential is able to capture the energy differences of the different competing phases of the solid at 0 K and finite temperature properties of the experimentally observed bcc phase. The potential also accurately predicts the temperature dependence of liquid density, viscosity, and the diffusion coefficient. The melting point is calculated using the two-phase coexistence method and is remarkably close to the experimental value. The potential is used to predict stress-induced phase transformations in solid Li and pressure-volume isotherms in liquid Li. We underline the necessity for a complete training set that includes both solid and liquid configurations in order to obtain a potential that precisely models both phases. By using the ACE formalism, we also systematically investigate the contributions of interactions involving $N$ bodies and the number of radial parameters needed to separately represent both phases since they have a direct consequence on the computational cost of the potential. We shed light on the complexity of the ACE potential needed to model solid and liquid lithium efficiently.

Close

PrevNext

View figure in article

• Interatomic & molecular potentials
• Density functional theory
• Machine learning
• Molecular dynamics