【Frontiers in HaiChuan Technology】Using AI technology to reveal the nature of the \"metal-support interaction\" in supported metal catalysts.
Thread Content
A problem that has plagued the field of catalysis for 50 years has finally been solved. 2024-12-11 Source: China Science Daily. Professor Li Weixue from the University of Science and Technology of China has a folder on her computer containing 329 different versions of the same paper. Each version is named according to the date; the earliest version was dated September 5, 2018, and the latest version was dated October 20, 2024. Recently, this paper, which solves a problem that has plagued the field of heterogeneous catalysis for nearly 50 years, was published online in Science. Li Weixue’s team used artificial intelligence (AI) technology to uncover the nature of the \"metal-support interaction\" in supported metal catalysts. Li Yadong, an academician of the Chinese Academy of Sciences and professor at Tsinghua University, believes that this achievement \"solves a major fundamental scientific problem in multiphase catalysis research and holds great value as a guide for the rational design of highly efficient supported catalysts.\" Solving difficult problems in a truly “elegant” manner: Over 90% of chemicals are synthesized with the help of catalysts, and oxide-supported metal catalysts are among the most widely used industrial catalysts. For a long time, scientists have been striving to develop catalysts with high activity, high selectivity, and high stability. “There are two fundamental aspects in supported catalyst systems: the relationship between the catalyst and the reactants, and the relationship between the catalyst and the support. The former is key to improving catalytic activity and selectivity, while the latter acts as a ‘stabilizing factor’. ”said Li Weixue, the corresponding author of the paper. Early studies mostly focused on the interaction between metal catalysts and reactants, ignoring the interaction between metal catalysts and supports. As research progresses, it has been found that oxide carriers not only serve to stabilize metal catalysts, but also influence charge transfer, morphological changes, the formation of new interface sites, chemical composition, and interface coating within the catalyst system, thereby significantly affecting the catalyst’s activity and selectivity. In fact, as early as 1978, scientists discovered that oxide carriers exhibited the phenomenon of encapsulating metal catalysts under high-temperature reduction conditions. This phenomenon is attributed to strong metal-support interactions. Later, this concept was used to explain all experimental phenomena showing significant interfacial effects. “There are too many and too complex factors that affect the interaction between metal and support, including the components, structure, size, and morphology of both the catalyst and the support. This effect is in turn sensitive to factors such as the catalyst preparation process and reaction conditions. ”Li Weixue said that currently, nearly 7,000 papers are published each year on the concept of metal-carrier interactions, and this number is on a steady upward trend. However, research on these interactions has not yet been quantified, especially due to the lack of a clear structure-activity relationship. To this end, starting in 2017, he led his students to use AI technology to tackle this problem, and after 8 years they finally solved this challenge that had plagued the field of heterogeneous catalysis for nearly 50 years. “To our surprise, among the many complex factors, it was the metal-metal bonds formed between the metal catalyst and the metals in the carrier that had a decisive impact, rather than the metal-oxygen bonds as previously believed. ”Li Weixue said that this provides a new perspective for understanding metal-carrier interactions. Reviewers in Science highly praised this work: “This research is very important for improving industrial catalysts; I congratulate the authors on addressing this issue from an atomic level in a ‘beautiful’ manner!” The research findings are highly creative, insightful, and profound. ”
Developing a universal and concise formula from 30 billion expressions: In this study, researchers first collected 178 experimental data sets for 27 oxides of 25 metals. They then used an explainable AI algorithm to create a candidate space comprising 30 billion expressions, and on this basis developed a machine learning formula with clear physical meaning. “Explainable AI algorithms start from the basic properties of materials, and through iterative mathematical combinations of the known physicochemical parameters of those materials, generate 30 billion expressions. ”Li Weixue said that they further utilized the principle of compressed sensing to derive an equation from 30 billion expressions that could reproduce all the experimental data. “It is relatively simple to derive a machine learning formula; the difficulty lies in how to adapt it to specific problems so that the formula becomes highly interpretable, and in extracting physical meanings and scientific principles from it. ”Li Weixue said that it depends on the researcher’s understanding and judgment of the issue. Ultimately, by combining domain knowledge with theoretical derivations, they developed a concise equation that is physically meaningful and numerically accurate, revealing for the first time the two key physical quantities that influence metal-carrier interactions, namely the \"metal-oxygen interaction\" and the \"metal-metal interaction\". After deriving the formula, they conducted repeated verifications and analyses on 675 metal-oxide systems, finding that the \"metal-oxygen interaction\" was the main factor contributing to the formation of interfaces, while the \"metal-metal interaction\" was the key factor in distinguishing the effects of different carriers. “Simply put, all oxide carriers contain oxygen; the difference lies in the metals they contain. Therefore, it is obvious that it is the ‘metal-metal interaction’ that has a decisive influence on the carrier effect. ”Li Weixue lamented that it took almost 50 years for people to break through this layer of pretense. It is worth noting that they also proposed a principled criterion for \"strong metal-metal interaction\", namely that when the interaction between two metals is stronger than the interaction between the metals within the oxide itself, the oxide carrier will coat the metal catalyst. For example, the metal catalyst is platinum, and the carrier is titanium oxide. When the platinum-titanium bond is stronger than the titanium-titanium bond, titanium oxide coating of the platinum catalyst occurs. This concise criterion effectively explains almost all the coating phenomena observed in such systems to date.
A relay race across time and space: Li Weixue conducted postdoctoral research at the Fritz Haber Institute of the Max Planck Society in Germany, and upon returning to her home country, she established a Sino-German Max Planck partnership research group with the institute’s director, Matthias Scheffler. In 2016, Scheffler was invited to attend an academic conference at the University of Science and Technology of China. When meeting Li Weixue, Scheffler praised his postdoc, Ouyang Runhai, who was previously Li Weixue’s doctoral student. He developed an explainable AI algorithm called SISSO, which holds great promise for research in the field of materials. This inspired Li Weixue. In 2017, he assigned Wang Tairan, who had just joined the research group as a graduate student, to use SISSO to study the metal-support interaction problem. Wang Tairan gathered a large amount of high-quality interface interaction data from hundreds of papers, and used SISSO to establish the corresponding formulas. However, the physical meaning and significance behind the formula still need to be further explored. In 2021, this paper was already in its initial form, but Li Weixue was not eager to publish it; firstly, the physical meaning of the equations obtained at that time was not very clear, and secondly, the importance of the new metal-metal interaction parameters in the catalytic process was still uncertain. Addressing these issues, Li Weixue led her team in extensive research, using AI-based equation predictions to calculate various possible physical quantities and attempting to relate them to important catalysis problems; the articles were rewritten repeatedly, yet the results remained unsatisfactory. A turning point came in 2023. PhD student Hu Jianyu used molecular dynamics simulations based on neural network potential functions to successfully reproduce theoretically the encapsulation phenomenon observed experimentally in 1978. This time, they found a breakthrough for the application of the theory. Through repeated revisions to the article’s innovation, importance, logic, clarity, and so on, their understanding of the physical meaning of the formulas became increasingly clear, profound, and comprehensive. It took only 85 days from submission to acceptance. On July 30, 2024, the research team officially submitted their work to Science. This is version 320 of the paper. On September 7, they received the review comments. To Li Weixue’s surprise, the editor had already revised the paper thoroughly. “This means the paper has been basically accepted. ” In accordance with the reviewer’s comments, the research team completed the necessary work and submitted a revised version on October 2. Soon, on October 22, the paper was officially accepted. The research took 8 years, with the paper undergoing 329 revisions; it only took 85 days from submission to acceptance for publication... Behind this slow and fast pace lay Li Weixue’s dedication to scientific research, as well as the efforts of 3 students who worked together across time and space. As soon as this theoretical work was published, it immediately attracted the attention of researchers in the same field. “Currently, two research groups have used the theory proposed this time to synthesize a series of new coated catalytic material systems, and research on new catalytic reactions is also advancing rapidly. ”Li Weixue said. “So far, we have only described the interaction at the metal-oxide interface. Next, we will continue to study the interactions between metals and various metal compound carriers. ”Li Weixue said that their long-term goal is to establish a general theory for describing the interactions at material interfaces, thereby accelerating the discovery of new catalytic materials and new catalytic reactions and contributing to the green transformation and sustainable development in the fields of energy, environment, and materials. Li Weixue said that this scientific breakthrough also demonstrates the great potential of explainable AI algorithms, offering new perspectives and solutions for solving major scientific problems. Related paper information: https://doi.org/10.1126/science.adp6034
Thank you for sharing; AI is indeed becoming more and more powerful