A machine learning methodology for investigating the liquid-liquid transition of hydrogen under high-pressure

Due to its unique properties such as superconductivity and superfluidity, high-pressure properties of hydrogen attract a lot of attention. However, the Liquid-Liquid Transition (LLT) of hydrogen under high-pressure and high-temperature is of particular significance for understanding its metallization. We propose a data-driven machine learning approach based on the density functional theory data to fit the potential energy surface into a deep neural network form. This method overcomes the simulation scale limitations of first-principles approaches to investigate the dissociation behavior of hydrogen molecules under high pressure. Our findings reveal an LLT curve exhibiting a first-order continuous change along with transition zone corresponding to hydrogen molecule dissociation. This study offers valuable insights into the LLT phenomenon and the metallization of hydrogen under high pressure.