Quantum-accurate SNAP Potential For Large-Scale Molecular Dynamics Simulations of Carbon at Extreme Conditions

Highly accurate interatomic potentials are of critical importance for trustworthy MD simulations of materials at extreme conditions of high pressures, temperatures, and high strain-rates. A new quantum accurate Spectral Neighbor Analysis Potential (SNAP) for carbon has been developed to describe the behavior of carbon at multi-megabar pressures and 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 is used to train the quadratic SNAP on a large set of first-principles training data. The SNAP development involves (1) the generation of the training database comprising a consistent and meaningful set of first-principles DFT data; (2) the robust and physically guided fit of SNAP parameters; and (3) the validation of the SNAP potential in simulations of carbon at extreme conditions. In this presentation, several applications of SNAP to study carbon at extreme conditions are described.