The accurate estimation of the third virial coefficients for helium using three‐body neural network potentials. (1st February 2022)

• Record Type: Journal Article
• Authors: Kwon, Taejin; Song, Han Wook; Woo, Sam Yong; Kim, Jong‐Ho; Sung, Bong June
• Is Part Of: Bulletin of the Korean Chemical Society. Volume 43:Number 5(2022)
• Page Start: 612
• Page End: 619
• Publication Date: 2022-02-01
• Subjects: deep learning method -- neural network potentials -- third virial coefficients -- three‐body interactions; Chemistry -- Periodicals; 540.5
• DOI: 10.1002/bkcs.12497
• Languages: English
• ISSNs: 0253-2964

Abstract

Abstract: The description of many‐body interactions is one of challenging problems in molecular dynamics simulations. Recently, neural network potentials have been spotlighted as an approach to describe many‐body interactions. In this study, we obtain the neural network potentials for three‐body interactions of helium using a deep learning method. We perform quantum calculations to obtain single point energies for helium trimers and obtain the neural network potentials for three‐body interactions by performing a deep learning method. In order to test the validity of the neural network three‐body interactions, we perform Mayer‐sampling Monte Carlo simulations and calculate third virial coefficients for helium. We show that the third virial coefficients obtained from three‐body neural network potentials are more accurate than those obtained from two‐body neural network potentials. The deep learning method in our study would be extended to obtain the high‐order virial coefficients for complex molecules. Abstract : We obtain the three‐body neural network potentials for helium using a deep learning method. Our results show that the third virial coefficients of helium obtained from the neural network potentials are quite accurate when compared with previous results. Our neural network potentials capture nonadditive three‐body interactions of helium successfully.