Abstract
The nucleation and growth of graphite in Fe-C alloys (cast iron) can both be related to the 2-D nucleation of a single atomic layer of graphene. Here, we use molecular dynamics (MD) simulations to study the nucleation of graphite from molten iron. A neural-network potential is developed for the Fe-C system using a pragmatic approach to treat the paramagnetic nature of liquid iron, and its thermodynamic behavior is validated against data from thermochemical software. The critical nucleus size and free-energy barrier for nucleation are computed using MD simulations in combination with enhanced sampling methods, and the results are compared with continuum models based on adaptations of classical nucleation theory. 2-D nucleation is found to be a feasible mechanism for graphite nucleation and growth at a carbon supersaturation of 5 at.% or higher. At low supersaturation the nucleation rate declines, and other mechanisms such as interaction with substrates other than graphite, solute elements, or defects are necessary to facilitate the formation of new layers.