MOLECULAR DYNAMICS SIMULATION OF STRATIFICATION IN Bi–Ga MELTS

Abstract

In present work, the process of stratification in melts of the Bi–Ga system was simulated using molecular dynamics method. The interaction between atoms was specified using a neural network potential parameterized on ab initio data (DeePMD model). The parameterization of the DeePMD potential was performed using an active machine learning algorithm. During molecular dynamics simulation, melts with the compositions GaxBi100 – x where x = 0, 10, …, 90, 100 were cooled from 800 to 300 K. The phase separation was registered by changes in the temperature behavior of the partial radial distribution function for the Ga–Bi pair. It has been established that the DeePMD potential, in the initial training set of which no configurations corresponding to the phase separated state were introduced, is still able to reproduce the stratification in the Bi-Ga system. The concentration range of separation determined by molecular dynamics modeling with the DeePMD potential coincides with the experiment. It was also possible to correctly determine the shift of the maximum of the stratification dome towards melts rich in gallium. However, the stratification dome maximum was incorrectly defined as Ga80Bi20 instead of the experimental Ga70Bi30. In addition, a certain temperature range of the delamination dome is wider than in the experiment. Despite this, the use of neural network potentials in atomistic simulations, as shown in present work, can be effectively used to predict delamination in binary metallic systems.