Understanding Phase Transformations in Carbon through Simulations using the ChIMES Machine-Learned Interatomic Potential

Incomplete understanding of phase transformations in carbon has been a bottleneck in planetary, synthesis, and inertial confinement fusion sciences, among other related research areas. Basic properties like phase boundary loci in the high-pressure regime are not well-defined, and these uncertainties are amplified when considering finite-scale and non-equilibrium effects. For example, experimental determination of the carbon melt line at high pressure is challenging, as platforms capable of generating such conditions can sustain them for only very short timescales (e.g., <1 μs), making precise temperature characterization difficult. Simulations offer a way to bridge these gaps. First principles methods have been used to predict the carbon melt line, but the high computational expense precludes their application to the relevant spatiotemporal scales. Classical interatomic potentials have been used to overcome this limitation. However, their underlying functional forms preclude accurate description of the intricate behavior of disordered (e.g., molten) carbon phases. In this study, we address these limitations using ChIMES, a physics-informed machine-learned interatomic potential, to revisit predictions of the carbon phase diagram. We will present our findings in the context of existing experimental data and prior simulation results. Our work highlights the potential of machine-learned interatomic potentials to enhance the understanding of high-pressure carbon phase transformation and its broader implications for fundamental and applied science.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL-ABS-2002308.