A Gaussian Process Based Δ-Machine Learning Approach to Reactive Potential Energy Surfaces.

Yang Liu,Hua Guo
DOI: https://doi.org/10.1021/acs.jpca.3c05318
2023-01-01
The Journal of Physical Chemistry A
Abstract:The Gaussian process (GP) is an efficient nonparametric machine learning (ML) method. A distinct advantage of the GP is its ability to provide an estimate of statistical uncertainties. This is particularly useful in constructing high-dimensional potential energy surfaces (PESs) from ab initio data as it offers an optimal way to add new geometries to reduce the overall error. In this work, GP is employed in the context of Delta-machine learning (Delta-ML), in which a correction PES to an inaccurate low-level PES is constructed using a small number of high-level ab initio calculations. This new method is tested in three prototypical reactive systems, namely, the H + H-2 -> H + H-2, OH + H-2 -> H2O + H, and H + CH4 -> H-2 + CH3 reactions. The results show that the GP-based Delta-ML approach is more efficient than its direct application in constructing high-level PESs. We also compare the new method to a previously proposed neural-network-based Delta-ML approach [Liu and Li J. Phys. Chem. Lett. 2022, 13, 4729-4738]. The results indicate that the two Delta-ML methods have comparable efficiencies in constructing accurate PESs.
What problem does this paper attempt to address?