Leveraging Machine-Learned Interatomic Models for Quantum Accurate Carbon Melt Line Prediction

The carbon melt line is of significant research interest due its relevance across a broad range of domains, including inertial confinement fusion science, detonation science, and nanocarbon manufacturing, among others. However, accurate experimental determination remains a challenge due to limitations of drivers capable of imposing such conditions for a sufficient duration and probes capable of temperature measurement on short timescales. First principles approaches have been previously deployed to estimate this phase boundary, but the associated computational costs preclude reaching necessary spatiotemporal scales. Efforts have been made to use classical interatomic models to overcome this challenge, but the underlying functional forms preclude accurate description of complex molten carbon phases. Recently, machine-learned interatomic models have proven to be a practical way to bridge the accuracy/efficiency gap between first principles and classical interatomic models. Hence, we have deployed ChIMES, a physics-informed machine-learned interatomic model to revisit this problem. Our melt line predictions will be presented alongside available experimental and simulation data, and analysis of emerging interfacial structures will be discussed.