Calculating thermal properties of phosphorene using a deep learning force field

Danila Shein,Dmitry Zav’yalov

DOI: https://doi.org/10.1140/epjp/s13360-024-05685-z

2024-10-01

The European Physical Journal Plus

Abstract:In this article, an attempt was made to calculate thermal conductivity and sound velocity of black phosphorene through a series of molecular dynamic simulations. In order to perform modeling of large-scale phosphorene sheets, we first constructed an interatomic interaction potential model using deep learning. A feedforward neural network architecture provided by the DeePMD package was selected for this purpose. Because the data for training the networks was collected from the results of ab initio molecular dynamics simulations, the predicted forces acting on atoms are nearly as precise as those calculated with ab initio methods themselves. The constructed force field model was subsequently utilized in classical molecular dynamics modeling. Tests of the phosphorene computer model built with the constructed potential show that the material structure at the minimized state is similar to the one derived with an ab initio approach. The Pearson correlation coefficient between the two radial distribution functions of particles in phosphorene in these systems is 0.858. Moreover, the density of the material is approximately equal to 2.688 g·cm −3 and remains consistent over long trajectories. These two factors indicate that it is safe to use the resulting force field model in further research. A modification of the Müller-Plathe method was proposed and used to calculate thermal conductivity coefficients. They were found to be 1.562 W·m −1 ·K −1 and 2.026 W·m −1 ·K −1 along the "armchair" and "zigzag" directions, respectively. The sound velocities determined with dynamic structure factor analysis are 3871 m·s −1 and 9340 m·s −1 in the same directions.

physics, multidisciplinary