Machine‐learning‐based interatomic potentials for advanced manufacturing

Abstract

This paper summarizes the progress of machine-learning-based interatomic potentials and their applications in advanced manufacturing. Interatomic potential is essential for classical molecular dynamics. The advancements made in machine learning (ML) have enabled the development of fast interatomic potential with ab initio accuracy. The accelerated atomic simulation can greatly transform the design principle of manufacturing technology. The most widely used supervised and unsupervised ML methods are summarized and compared. Then, the emerging interatomic models based on ML are discussed: Gaussian approximation potential, spectral neighbor analysis potential, deep potential molecular dynamics, SCHNET, hierarchically interacting particle neural network, and fast learning of atomistic rare events.

Revealing the improved stability of amorphous boron-nitride upon carbon doping

“…From the theoretical and computational perspective, atomistic calculations represent a suitable tool to describe complex structures of concern, giving access to details at the atomic and molecular level. However, the extremely disordered nature of amorphous materials requires a computational approach able to capture the interatomic potentials in arbitrary complex local environments, a challenge that can only be tackled with machine learning based methods 23,24 . Specifically, classical molecular dynamics with the employment of force-fields derived using machinelearning and ab-initio techniques constitute a powerful methodology to describe sufficiently large disordered materials samples while keeping first-principles accuracy.…”

Advances in data‐assisted high‐throughput computations for material design

“…ML models can achieve significant acceleration of electronic structure calculations by accurately reproducing pseudopotentials of atomic or ion electronic structures through learning from high‐precision reference data. One of the practical approaches is to develop machine learning interatomic potentials (MLIP) to describe a very large system at the ab initio accuracy [137] . Algorithms to fit MLIP have been reported, for example, Gaussian approximation potential (GAP), [138] spectral neighbor analysis potential (SNAP), [139] high dimensional neural network potential (HDNNP), [140] deep potential molecular dynamics (DPMD), [141] SchNet, [142] hierarchically interacting particle neural network (HIP‐NN), [143] and fast learning of atomistic rare events (FLARE) [144] …”