posted on 2023-11-29, 16:20authored byTongtong Yang, Donglai Zhou, Sheng Ye, Xiyu Li, Huirong Li, Yi Feng, Zifan Jiang, Li Yang, Ke Ye, Yixi Shen, Shuang Jiang, Shuo Feng, Guozhen Zhang, Yan Huang, Song Wang, Jun Jiang
Generative artificial intelligence has depicted a beautiful
blueprint
for on-demand design in chemical research. However, the few successful
chemical generations have only been able to implement a few special
property values because most chemical descriptors are mathematically
discrete or discontinuously adjustable. Herein, we use spectroscopic
descriptors with machine learning to establish a quantitative spectral
structure–property relationship for adsorbed molecules on metal
monatomic catalysts. Besides catalytic properties such as adsorption
energy and charge transfer, the complete spatial relative coordinates
of the adsorbed molecule were successfully inverted. The spectroscopic
descriptors and prediction models are generalized, allowing them to
be transferred to several different systems. Due to the continuous
tunability of the spectroscopic descriptors, the design of catalytic
structures with continuous adsorption states generated by AI in the
catalytic process has been achieved. This work paves the way for using
spectroscopy to enable real-time monitoring of the catalytic process
and continuous customization of catalytic performance, which will
lead to profound changes in catalytic research.