Atomic Cluster Expansion for Quantum-Accurate Large-Scale Simulations of Carbon

We present an atomic cluster expansion (ACE) for carbonthat improvesover available classical and machine learning potentials. The ACEis parametrized from an exhaustive set of important carbon structuresover extended volume and energy ranges, computed using density functionaltheory (DFT). Rigorous validation reveals that ACE accurately predictsa broad range of properties of both crystalline and amorphous carbonphases while being several orders of magnitude more computationallyefficient than available machine learning models. We demonstrate thepredictive power of ACE on three distinct applications: brittle crackpropagation in diamond, the evolution of amorphous carbon structuresat different densities and quench rates, and the nucleation and growthof fullerene clusters under high-pressure and high-temperature conditions.