Machine learning analyzes the active center and reaction mechanism of CuZnAl catalyzing methanol synthesis

On July 19, 2022, Professor Liu Zhipan and Shangcheng Professor Team from the Department of Chemistry of Fudan University published an article entitled “Methanol Synthesis from CO2/CO Mixture on Cu–Zn Catalysts from Microkinetics-Guided Machine Learning” at the Journal of the American Chemical Society Pathway Search” research results.

This achievement develops a microscopic kinetic-guided machine learning reaction path search method, and automatically searches the reaction paths of CO2 hydrogenation and CO hydrogenation on the surface of Cu and CuZn alloys, and confirms the kinetic optimal reaction channel. The theory points out that CO2 hydrogenation is the main channel to produce methanol, the reaction mainly occurs in the metal step, and the role of Zn is mainly related to the stabilization of Cu nanoparticles and step positions, and has little relationship with reaction kinetics.

The corresponding authors of the paper are Liu Zhipan, Shangcheng; The first author is Shi Yunfei.

Cu/ZnO/Al2O3 catalytic methanol synthesis is an important traditional industrial process that has been industrialized for nearly 60 years, and generally uses CO and CO2 mixed intake gas to react with H2 at high temperature and pressure. Since the experimental high-resolution characterization method can generally only be carried out under ultra-high vacuum or low pressure, it is difficult to achieve industrial reaction conditions, and there has been great controversy about the active state of Zn species in the catalyst and the role of CO2 in methanol synthesis under industrial catalytic conditions. The solution of exceptionally complex catalytic problems under such operating conditions requires both a deep understanding of the thermodynamics of materials and precise data on the reaction kinetics of complex multi-step reaction networks.

In the past five years, Liu Zhipan’s research group has developed a potential function method based on global artificial neural networks, created the LASP software platform (, integrated a series of potential energy surface search methods developed by the research group, and formed a set of effective efficient computing platforms for studying complex materials and reactions. Typical work of the research group based on LASP software in recent years includes the search for complex catalytic active site structures (Nature Catal. 2019, 2, 671, J. Am. Chem. Soc., 2021, 143, 11109), Automated Reaction Prediction (J. Am. Chem. Soc., 2019, 141, 20525), Si/SiO2 Semiconductor Interface Prediction (Phys. Rev. Lett. 2022, 128, 226102), theoretical guidance for PdAg catalysis experiments (J. Am. Chem. Soc., 2021, 143, 6281) et al. These theoretical works provide an important basis for further development of theoretical new methods to study Cu/ZnO/Al2O3 catalytic methanol synthesis.

Figure 1: Flowchart of the MMLPS method

Aiming at the hydrogenation process of CO/CO2 mixture at different CuZn possible sites, we propose a microscopic kinetic-guided machine learning reaction path search method (MMLPS). The MMLPS method adopts the idea of divide-and-conquer, using large-scale machine learning atomic simulations, combined with the shortest path of graph theory, microscopic kinetic solver, and on multiple surfaces such as Cu (111), Cu(211) and Zn alloyed Cu(211), the possible reaction paths on CO2 hydrogenation and CO hydrogenation reactions were sampled, and millions of structures were sampled, from which the reaction channels with the lowest kinetics were confirmed. MMLPS can achieve fully automated search, does not rely on human guesswork, and has irreplaceable advantages for complex multi-step reactions catalyzed by the surface of complex materials.

Fig. 2: a) Contour plot of the state density of the reaction pair sampled on Cu(211); b) reaction path of CO2 hydrogenation; c) Microscopic dynamics simulation.

Based on the global thermodynamic analysis of the surface structure of the material and the reaction kinetics data of MMLPS, we found that under the reaction conditions, the coverage of metal Zn on the Cu(211) step was up to 0.22 ML, and the Zn-Zn dimer site was unstable. CO2 and CO hydrogenation occur only at the step edge of the (211) step surface, and the low coverage Zn (0.11 ML) has little effect on the reaction kinetics, but the Zn with higher coverage (0.22 ML) poisons the catalyst. Microscopic kinetic simulations show that CO2 rather than CO is the main carbon source for methanol synthesis, mainly due to the rapid step intermediates of CO hydrogenation[CHO] Very unstable, kinetic simulation results are consistent with the results of previous isotope labeling experiments.

Although the direct catalytic effect of Zn at the metal step (active site) is not significant, we found that under industrial reaction conditions, the surface of cu(111) metal can grow thermodynamically stable[-Zn-OH-Zn-]The chain structure (cationic Zn) shows that the ZnO carrier can partially reduce and cover the main crystal plane of the metal Cu, which shows that the use of ZnO can effectively disperse Cu nanoparticles to prevent high temperature inactivation, while the CO presence in the mixture can effectively reduce the cation Zn on the Cu surface to expose the metal sites used for methanol synthesis, and the remaining metal Zn promotes the formation of CuZn alloys and reconstructs a rich stepped active site.

The work was completed by Shi Yunfei, a 19th-level doctor in the Department of Chemistry of Fudan University, under the guidance of teacher Liu Zhipan, and Mr. Shangcheng and Dr. Kang Peilin provided help in programming. The work was supported by the National Key Research Program Nano Special Project (2018YFA0208600) and the National Natural Science Foundation of China (22033003, 21533001, 91745201, 91945301), as well as the support of the Department of Chemistry of Fudan University in laboratory construction. (Source: Science Network)

Related paper information:

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button