New approach to understanding the experimental Cs NMR chemical shift of clay minerals via machine learning and DFT-GIPAW calculations

Okubo, Takahiro*; Takei, Akihiro*; Tachi, Yukio  ; Fukatsu, Yuta  ; Deguchi, Kenzo*; Oki, Shinobu*; Shimizu, Tadashi*

The identification of adsorption sites of Cs on clay minerals has been studied in the fields of environmental chemistry. The nuclear magnetic resonance (NMR) experiments allow direct observations of the local structures of adsorbed Cs. The NMR parameters of Cs, derived from solid-state NMR experiments, are sensitive to the local neighboring structures of adsorbed Cs. However, determining the Cs positions from NMR data alone is difficult. This paper describes an approach for identifying the expected atomic positions of Cs adsorbed on clay minerals by combining machine learning (ML) with experimentally observed chemical shifts. A linear ridge regression model for ML is constructed from the smooth overlap of atomic positions descriptor and gauge-including projector augmented wave (GIPAW) ab initio data. The Cs chemical shifts can be instantaneously calculated from the Cs positions on any clay layers using ML. The inverse analysis from the ML model can derive the atomic positions from experimentally observed chemical shifts.