Data Quality, Data Sampling and Data Fitting: A Tutorial Guide for Constructing Full-Dimensional Accurate Potential Energy Surfaces (PESs) of Molecules and Reactions

Jun Li,Yang Liu

2023-01-01

Abstract:Molecular dynamicComputational chemistry properties, including spectra, collision energy transfer, kineticsKinetic energy and dynamics, are largely determined by the system’s potential energy surface (PES)Potential energy surface (PES), whose performance is essentially determined by data quality, data samplingData sampling and data fittingData fitting. Here, we provide careful discussions and considerations on these three key factors with ample interpretative examples, centering on accuracy, efficiency, and generality. Briefly, for a molecular system, a sufficiently large number of data points are sampled and calculated at some accurate electronic structure level. Then the PES can be fitted to a specific function, such as the permutational invariant polynomial-neural network (PIP-NN)Permutational invariant polynomial-neural network (PIP-NN) method, which has been successfully applied to nearly 60 neutral or charged molecular systems with up to 8 atoms. For more and more complicated molecular systems, a NNNeural network (NN) based Δ-machine learning method is proposed to efficiently obtain high-level electronic energies of nuclear configurations in all dynamically relevant regions based on ample direct low-cost low-level calculations. Relevant perspectives and improvements are provided. Finally, a checklist is proposed to train and report a full-dimensional accurate PES, which is essential for publishers, authors, refereesCoupled cluster, readers, and users for reproducing, evaluating, and guiding.