Interatomic potentials (IP) based on neural network (NN) have gained significant

Mingjie Liu, Somayeh Faraji

Department of Chemistry, University of Florida

Interatomic potentials (IP) based on neural networks (NN) have gained significant attention in materials science for their ability to model complex atomic interactions. In this study, we report an NN-based IP specifically designed for C-H systems. The potential is trained on cluster data and effectively describes the potential energy surface (PES) for C-C bond dissociation in carbon dimer and hydrocarbons, as well as the adsorption of molecules on 2D materials. First, we investigate the PES generated by NN potential for carbon dimer (C2), ethyne (C2H2), ethene (C2H4), and ethane (C2H6), demonstrating that it accurately captures the general trend of energy versus C-C bond length. Next, we examine the adsorption of H atom and CHx (x=1-4) molecules on 2D graphene, highlighting the potential’s reliability in modeling complex interactions. Finally, we explore the hydrogenation of periodic and non-periodic carbon chains, emphasizing the potential’s ability to capture adsorbed geometry and bond alternation.