Generative artificial intelligence (AI) paints a beautiful blueprint for on-demand design in chemistry research. However, since most chemical descriptors are mathematically discrete or discontinuously tunable, a few generations of successful chemical descriptors can only achieve some special property values. Based on this,Professor Jiang Jun, Professor Wang Song, Associate Researcher Huang Yan (co-corresponding author) of University of Science and Technology of China, etcUse spectral descriptors and machine learning (ML) to establish quantitative spectral structure-property relationships of adsorbed molecules on metal single-atom catalysts. In addition to the catalytic properties such as adsorption energy and charge transfer, the complete spatial relative coordinates of the adsorbed molecules were successfully inverted.
Taking a single-atom catalyst (SACS) with metal atoms dispersed on a metal oxide support as an example, the quantitative relationship between the adsorption state (properties and structure) and the spectral characteristics of CO, a key intermediate molecule in the carbon dioxide reduction reaction (CO2RR), was studied. In this paper, a ML model (ML-1) was established to study the properties of adsorption energy (EADS) and charge transfer (δq), and another ML model (ML-2) was established to study the structure of adsorbed molecule CO. In this study, six structural parameters, including bond length, bond angle and dihedral angle, were identified using the infrared spectroscopic signal of CO, and the position of the small molecule relative to SACs can be accurately determined, and then the spatial relative coordinates of the CO molecule can be determined, so as to achieve structural inversion. This is a new method of obtaining structural information from spectra without relying on theoretical calculations or experimental assumptions.
In addition, based on the above two ML models, an AI generation workflow for catalytic structure design was developed. First, a large number of spectra are randomly generated to rapidly adsorption energy, and the spectral generation process is iterated to find the spectrum corresponding to the desired property by comparing it with the desired adsorption energy. Then, based on this spectrum, the structure of the CO molecule was inverted, and its properties were verified by DFT calculation. So far, it has been possible to continuously design catalytic structures with the desired performance using spectroscopy as an indicator.
catalytic structure design by ai generating with spectroscopic descriptors. j. am. chem. soc.,, doi: 10.1021/jacs.3c09299.
ht-tps: