Exploring compressed lithium through machine learning-assisted metadynamics simulations

Lithium exhibits remarkable structural diversity, transforming from high-symmetry cubic to complex low-symmetry crystalline phases across its pressure-temperature phase diagram. Here we explore the 100-300 GPa regime at elevated temperatures using a novel crystal structure prediction approach. Our method combines metadynamics simulations of crystallization from the liquid phase with machine-learning interatomic potentials trained on near-melting-point configurations at various pressures. This physically motivated approach provides insights into the potential crystal structures of Lithium under extreme pressures at finite temperatures and demonstrate the effectiveness of combining machine learning with metadynamics for crystal structure prediction.