The Rise of Neural Networks for Materials and Chemical Dynamics

Abstract

Machine learning (ML) is quickly becoming a premier tool for modeling chemical processes and materials. ML-based force fields, trained on large data sets of high-quality electron structure calculations, are particularly attractive due their unique combination of computational efficiency and physical accuracy. This Perspective summarizes some recent advances in the development of neural network-based interatomic potentials. Designing high-quality training data sets is crucial to overall model accuracy. One strategy is active learning, in which new data are automatically collected for atomic configurations that produce large ML uncertainties. Another strategy is to use the highest levels of quantum theory possible. Transfer learning allows training to a data set of mixed fidelity. A model initially trained to a large data set of density functional theory calculations can be significantly improved by retraining to a relatively small data set of expensive coupled cluster theory calculations. These advances are exemplified by applications to molecules and materials.

Machine learning potential era of zeolite simulation

“…In the past decade, with the rapid development of articial intelligence methods, the machine learning (ML) technique has been utilized to provide both accurate and efficient description of the PES of materials. [27][28][29] The ML potential can be considered as an advanced version of classic force elds, which contains signicantly more tting parameters without the explicit correspondence between the physical interaction term and the tting functional form. At present, there are many different ML potentials developed to date using different ML techniques, such as neural network (NN) potentials, [30][31][32][33][34][35][36] Gaussian approximation potentials [37][38][39] and moment tensor potentials.…”

Auto3D: Automatic Generation of the Low-energy 3D Structures with ANI Neural Network Potentials

“…As expected, the ANI geometries are very consistent with the ωB97X/6-31G* during the geometry optimization process, although there is a possible space for improvements. The ANI-2xt model showed very similar performance to ANI-2x in terms of geometry optimization, which has been widely adopted in different applications and showed reliable performance [51][52][53] . Therefore, ANI-2xt could be also used as a reliable optimizing potential in Auto3D.…”

Auto3D: Automatic Generation of the Low-energy 3D Structures with ANI Neural Network Potentials

“…As expected, the ANI geometries are very consistent with the ωB97X/6-31G* during the geometry optimization process, although there is a possible space for improvements. The ANI-2xt model gave very similar performance to ANI-2x in terms of geometry optimization, which has been widely adopted in different applications and has shown reliable performance [50][51][52] . Therefore, ANI-2xt could also be used as a reliable optimizing potential in Auto3D.…”