Phase diagrams of elemental metals with DFT and machine Learning

Prediction of materials phase diagrams is an important challenge in computational materials science. This becomes even more difficult at extreme pressures and temperatures where various approximations that work at low pressure regimes might fail. An additional obstacle is the lack of full experimental data in such conditions. In this work we show our analysis for elemental metals, with examples of aluminum and cerium. We demonstrate the use of the AFLOW package with the VASP Density Functional Theory (DFT) code to predict the cold curve of aluminum and cerium. We show that while some pseudopotentials have poor performance, with most pseudopotentials we manage to get results which are very close to all electrons calculations. We then combine VASP on the fly machine learning and deep learning (DL) codes such as NequIP to predict the temperature dependence of the materials with methods such as heat until it melts (HUM) and the two-phase approach (TPA).