Development of neural network potential for MD simulation and its application to TiN

T Miyagawa and K Mori and N Kato and A Yonezu, COMPUTATIONAL MATERIALS SCIENCE, 206, 111303 (2022).

DOI: 10.1016/j.commatsci.2022.111303

Titanium nitride (TiN) is used in various applications because of its excellent wear and corrosion resistance. TiN with a rock-salt-type crystal structure has been investigated extensively, but the existence of non-rock-salt-type phases (e.g., Ti2N) has only been reported recently from first-principles calculations. Molecular dynamics (MD) simulation is a powerful tool to predict mechanical properties, but it generally requires interatomic potentials. The conventional many-body interatomic potentials (e.g., modified embedded-atom method (MEAM) potential) for rock-salt TiN are not applicable to these other phases. Hence, in this study, a neural network (NN)-based method to create interatomic potentials is developed, which are referred to as neural network potentials (NNPs); hence, MD simulations can be conducted for TiN and other phases. First, ab initio molecular dynamics (AIMD) simulations are conducted to obtain the relationships between atomic configurations and the corresponding total potential energies and forces on each atom. Subsequently, these relationships are used for the NN to generate NNP. NNP can accurately reproduce the energies and forces calculated by AIMD simulations. The mechanical properties of TiN are computed using the NNP and MD simulation and verified with the MEAM potential and its experiment. Finally, MD simulation using the developed NNP is conducted for the other phase (Ti2N) to investigate its mechanical properties.

Return to Publications page