Modeling of phosphorene with classical molecular dynamics using deep learning

In this article, an attempt was made to construct a model of black phosphorene interatomic interaction potentials using deep learning. A feedforward neural network architecture provided by the DeePMD package was selected for this purpose. The training data were collected from the results of ab initio molecular dynamics simulations. The constructed force field model was subsequently utilized in the classical molecular dynamics simulations. The density of phosphorene in the resulting computer model is approximately equal to 2.72 g/cm$^3$, and its thermal conductivity coefficients were found to be 1.685 W/(m$\cdot$K) and 2.552 W/(m$\cdot$K) along the "armchair" and "zigzag" directions, respectively. The determined physical properties of black phosphorene are adequately consistent with the corresponding real values.