Rapid Algorithm for Creation of Machine Learning Potential for Viscosity Calculation of Liquid Ni30Fe70

The use of machine learning methods in materials modeling is a powerful tool for reproducing the properties of a medium with an accuracy close to ab initio calculations. Describing the properties of even binary disordered systems is a nontrivial task due to the complexity of finding the global minimum of potential energy to describe the medium. This problem can be solved using machine learning methods. Using the deep learning method implemented in the DeePMD package, we demonstrate that a machine-learned potential can be obtained for calculating the viscosity of Ni30Fe70 melt using classical molecular dynamics. Moreover, the accuracy of viscosity calculation will be close to the experimental one. This result opens up prospects for developing procedures to rapidly obtain potentials for modeling the dynamic properties of binary alloys.