A Portable Data Set for Borophene Growth Modeling with Reactive Neural Network Potentials

Abstract

In this study, we develop and validate machine learned interaction potentials (MLIPs) for reactive simulation of borophene on metal substrates. A versatile training data set is constructed to accurately represent both extended and reactive borophene structures. It should be portable to train any MLIP architecture. Indeed, three generations of MLIPs, namely n2p2, DeePMD and NNMP, are trained and validated against density functional theory (DFT) calculations. Our results demonstrate the capability of the MLIPs to accurately reproduce DFT-calculated structures, energies, and forces. We finally show that it is possible to use these MLIPs to simulate the growth of borophene on a silver substrate.