Modeling the thermal conductivity of black phosphorene using deep learning

With the help of a convolutional neural network with continuous filtering SchNet, trained on the simulation data by the Car-Parrinello quantum molecular dynamics method, the potential of the black phosphorene force field is constructed, applicable for use in the framework of modeling by the classical molecular dynamics method. The parameters of the neural network and the ways of its training are revealed, which allow us to build the most realistic representation of the force field. Using a force field calculated by a neural network, the thermal conductivity of a sample of black phosphorene in a LAMMPS package was simulated. The calculated values of thermal conductivity are consistent with the data obtained by other groups experimentally and within the framework of calculations.