Machine Learning Interatomic Potential for Molten TiZrHfNb

High-entropy alloys (HEAs) are relatively new class of materials with promising functional and mechanical properties. These alloys contain multiple elements with equi- or almost equiatomic concentrations and should represent random solid solution. Therefore, in HEAs, several different chemical elements coexist in one phase. Interaction between multiple species in one phase is of interest, since understanding of features of this interactions can provide understanding of thermodynamics stability of such systems. As far as properties of solid alloy are connected with properties of its melt, it is reasonable to start the investigation of particular multi-component system from liquid state. However, the problem of describing of potential energy surface (PES) for metals is especially vexing. For solving this problem here we applied machine learning technique, namely DEEPMD approach, for developing neural-network potential (NNP) for molten TiZrHfNb as an example of multi-component system. Training set was generated using oh initio molecular dynamics (AIMD) trajectories. Validation of the potential was performed by comparing of partial radial distribution functions (PRDFs) obtained by AIMD and DEEPMD methods. Analysis of PRDFs allowed to conclude that TiZrHfNb system is very likely to form single-phase random solid solution.