AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-19
Tersoff-NN: Hybrid Machine-Learned Bond-Order Extension of the Tersoff Potential
Machine-learning potentials are accurate but costly and can be unreliable outside their training domain. Tersoff-NN, a hybrid potential for silicon, keeps the analytic radial terms and cutoff structure of the Tersoff potential and represents only the environment-dependent bond-order factor with compact neural networks. On 32,768-atom benchmarks it runs 4 to 12 times faster than the Behler-Parrinello-type NN potentials tested, and it improves the short-range structure of liquid Si over the conventional Tersoff potential.