Unveiling fullerene formation and interconversion through molecular dynamics simulations with deep neural network potentials

Abstract

Utilizing deep neural network potentials within molecular dynamics simulations, this research uncovers insights into fullerene formation and interconversion, particularly during the cooling stage of the annealing process. Our deep learning-enhanced approach effectively models the generation of fullerenes from C₂ units in carbon vapor, highlighting the critical role of carbon density in structuring outcomes in a primary iron–carbon system. This study provides differences in molecular dynamics simulations for fullerene generation and also demonstrates the potential of deep learning in investigating complex carbon structures, paving the way for further investigations into the fullerene family.