A potential function of MoS2 based on machine learning

MoS2 has been used as a non-toxic and economical thermoelectric material, which attract the interest of researchers. In terms of computational simulation, system energy is the most basic and important property of microscopic materials. Although the first principles based on density functional theory (DFT) can calculate the energy accurately, the calculation is limited by time and scale. The potential function based on existing data can make up for this shortage. This study uses the Beller-Parrinello (BP) method to construct the machine-learning potential of the Mo-S system. The potential function is trained by the atomic energy network (aenet) software package, and is able to be used in bulk structure and two-dimensional structure. We predicted the energy of structures under different strains and conduct molecular dynamics (MD) simulation to predict the energy change during the simulation process. In addition, we calculated the defect formation energy of MoS2. The results of ANN were very close to the calculations of DFT. This potential function with high accuracy can greatly shorten the period of exploring the defect structure of MoS2 and it provides a powerful tool for exploring the large-scale defective MoS2.