Adsorption and modification behavior of single atoms on the surface of single vacancy graphene: Machine learning accelerated first principle computations

Abstract

During the preparation or processing of graphene, the emergence of structural flaws is an unavoidable circumstance, and these structural defects present a detrimental impact on the mechanical properties of graphene and its corresponding composite materials. In this paper, we use a first-principles computation to compute single-atom adsorption behavior on single vacancy graphene, from which adsorption energy and distance values are derived to create a dataset for machine learning. The comparative analysis of different machine learning models reveals that the BPNN model is best suited for the dataset. The BPNN model is enhanced by adjusting its node parameters using genetic algorithm, increasing its coefficient of determination to 0.9874 and 0.9608 for adsorption energy and distance models, respectively. The enhanced GA-BPNN model is utilized to predict the adsorption behavior of atoms across the entire periodic table onto single vacancy graphene’s surface. The accuracy of the machine learning model predictions is validated through the application of elemental modifications to the graphene. Employing machine learning to expedite first-principles calculations broadens the spectrum of available research approaches while accelerating the atomic modification process of single vacancy graphene. Our results provide valuable insights into the adsorption behavior of atoms on graphene surfaces and demonstrate the potential of machine learning for accelerating first principle computations in material science.