Short-range order influence on the quality of interatomic potential built using machine learning technique for high entropy alloys

Lobzenko, I.   ; Tsuru, Tomohito   

Nowadays quantum mechanical approach is essential part of computer modeling of materials. Nevertheless, long time scale and large systems are still inaccessible for that method because of high demand of computational resources. That is why classical molecular dynamics is the method of choice for modelling such processes as dislocation movement or crack propagation. This work is devoted to development of high accuracy interatomic potentials for high entropy alloys. To begin with, we have built potentials for medium entropy ternary alloys MoNbTa and ZrNbTa. The technique of machine learning of neural network was used. It allows effective fitting of the data set calculated using quantum mechanical approach. The composition of the data set is discussed. To ensure high quality of the potential the structures with a short-range order were added to the data set. For that purpose, Monte Carlo simulations of special quasirandom structures were carried out.