Quantum-accurate SNAP Carbon Potential for MD Simulations of Carbon at Extreme Conditions

Highly-accurate interatomic potentials are urgently sought for realistic MD simulations of high strain-rate experiments of carbon materials response to high temperatures and pressures. With that goal in mind, we have developed a novel Spectral Neighbor Analysis Potential (SNAP) machine-learning potential to describe properties of carbon at extreme pressures (up to 5 TPa) and temperatures (up to 10,000 K). SNAP is formulated in terms of the bispectrum components, which play a role of descriptors that characterize the local neighborhood of each atom. Machine learning approaches are used to train the SNAP potential on a large dataset of first-principles training data. SNAP development includes (1) generation of the training database comprising the consistent and meaningful set of first-principles DFT calculations; (2) the robust and physically guided fit of the SNAP parameters; and (3) the validation of the SNAP potential in MD simulations of carbon at extreme conditions. SNAP potential is applied to investigate shock response of diamond at extreme conditions.