Predicting the Carbon Melting Line Under 100 GPa with the ChIMES Machine-Learned Interatomic Potential

Investigating the carbon melt line is important in diverse research areas, including planetary science, detonation science, and high-throughput high T/P production of nanocarbon materials. Precise experimental determination of the carbon melt line is difficult since platforms capable of creating these conditions can only maintain them for very short timescales (e.g., < 1 μs), making characterization of ensuing temperature exceedingly difficult. As experiments are confronted with these challenges, simulations could help fill the gap. First principles methods have been used to predict the carbon melt line, but the high computational expense precludes their application to the relevant spatiotemporal. Others have attempted to overcome this limitation by using classical interatomic potentials. However, their underlying functional forms preclude accurate description of the intricate behavior of complex molten carbon phases. Recently, machine-learned interatomic potentials have emerged as an effective solution for closing the gap between the accuracy of first principles simulations and the efficiency of classical interatomic potentials. In this work, we have applied ChIMES, a physics-informed machine-learned interatomic potential, to revisit the prediction of the carbon melt line. We will present our melt line predictions and discuss our results within the context of available experimental data and prior simulation predictions. We will also discuss the evolving structures at the interface between solid and liquid phases of carbon.