From material design and synthesis to catalyst innovation and carbon neutrality, Tsinghua Wang Xiaonan’s team explores the frontier and implementation of “AI+ materials”

Author | Wang Xiaonan, Tsinghua University

Editor | Kaixia

In today's era of rapid technological development, the research and development of new materials has become a driving force A key force in scientific progress and the industrial revolution. From energy storage to information technology to biomedicine, the design, synthesis and functional characterization of innovative materials are the cornerstones of breakthroughs in these fields. The research and development of new materials has shown a trend of breakthroughs in many fields. In terms of energy storage, researchers are working to develop more efficient and safer battery materials to meet the storage needs of renewable energy. At the same time, the advancement of information technology has also prompted materials scientists to

With the continuous advancement of artificial intelligence (AI) technology, its application in new materials research has opened a new research paradigm and become a new research paradigm that transcends the traditional R&D model. qualitative productivity. Especially in the design, synthesis and characterization of materials, the assistance of AI has greatly improved research efficiency and accuracy.

##"Goed to Tsinghua University at the age of 17, became a doctoral supervisor at the age of 27, returned to Tsinghua University at the age of 30, a post-90s scientific research goddess, and was selected into the 2023 Global Scholars Lifetime Academic Influence List..." This is The legendary resume of teacher Wang Xiaonan from Tsinghua University.

The team she leads is committed to cross-disciplinary research on AI-accelerated material development and application, catalyst design, new energy, and low-carbon technology.

In recent years, research on cutting-edge new energy and low-carbon technologies and systems has been carried out around interdisciplinary subjects such as AI, energy, chemical industry, and environmental new materials, to improve the overall efficiency and effectiveness of energy and resource systems from the perspective of multi-scale system integration. Economical, supporting dual carbon goals.

In terms of AI accelerated materials research and application, there are a series of highly cited papers and algorithm software output in fields such as new energy systems and chemical intelligent models.

In the AI ​​era with the explosion of large models, the "Chemical Materials GPT" is underway. Wang Xiaonan said that basic model research is a long-term matter, "let the large models be implemented, go deep into the fields of science and engineering, and find The era of appropriate application objects, implementation scenarios, and the integration of large and small models has arrived.”

The prompt project embedding domain knowledge promotes the application of LLM in the scientific field

Wang Xiaonan’s team. We have long valued generative AI, especially the vertical application of large language models (LLM) in the scientific field.

Currently, large language models have proven their great potential in processing and analyzing large-scale data sets in multiple general fields. However, these models often require more fine-tuning to achieve optimal results when faced with the complexity of specific vertical domains.

Prompt engineering refers to optimizing and guiding the output of large language models through carefully designed prompts or guidance statements, so that they can better adapt to and handle problems in specific fields.

Recently, Wang Xiaonan’s team developed a hint engineering method to enhance its performance in the scientific field by integrating chemical domain knowledge in large language models.

Illustration: Flowchart of prompt engineering algorithm embedding knowledge in the chemical domain.

The study first created a benchmark cue engineering test dataset that includes complex physicochemical properties of small molecules, drug availability, and functional properties of enzymes and crystalline materials to highlight their potential in biology and chemistry. Relevance and application. At the same time, a domain-knowledge embedded prompt engineering method is proposed by combining several prompt engineering heuristics (heuristics). This method is superior to traditional prompt engineering strategies in multiple metrics.

Additionally, the team demonstrated the effectiveness of the approach through case studies on complex materials such as MacMillan catalysts, paclitaxel, and lithium cobalt oxide, highlighting the role of large language models equipped with domain-specific hint engineering as Potential as a powerful tool for scientific discovery and innovation.

The related research was titled "Integrating Chemistry Knowledge in Large Language Models via Prompt Engineering" and was published on the preprint platform arXiv on April 22, 2024.

Paper link: https://arxiv.org/pdf/2404.14467

Intelligent atom robot probe technology, Quantum materials can be efficiently manufactured with atomic precision

Recently, the intelligent atomic robot probe technology jointly developed by Wang Xiaonan’s team and Associate Professors Lu Jiong and Chun Zhang of the National University of Singapore provides a typical demonstration of this paradigm change. By combining AI with probe chemistry technology, atomically precise synthesis of carbon-based quantum materials was achieved.

Illustration: Atomic robotic probe that can accurately create organic quantum substances at the single-molecule level.

This work proposes a conceptual system of the Chemist's Intuitive Atomic Robot Probe (CARP) to prepare and characterize open-shell magnetic nanographene at the single-molecule level to achieve its π Precise construction of electronic topology and spin configurations.

CARP is driven by a series of deep neural networks trained by the experience and knowledge of surface chemists. It can realize the independent synthesis of molecular materials and effectively obtain valuable hidden information from the experimental training database, providing a comprehensive understanding of exploration. Provide important support for theoretical simulations of chemical reaction mechanisms.

The related research was titled "Intelligent synthesis of magnetic nanographenes via chemist-intuited atomic robotic probe" and was published in "Nature Synthesis" on February 29, 2024.

Paper link: https://www.nature.com/articles/s44160-024-00488-7

Professor Michael Gottfried, University of Marburg, Germany At the same time, he wrote an article "Single-molecule chemistry with a smart robot", which highly praised this work as a leading example of the combination of AI and nanotechnology.

「Stands out as a pioneering example, showcasing remarkable advancements in controlling molecules at the limit of single chemical bonds.」

This study It not only overcomes the problems of poor reaction selectivity and low production efficiency in traditional surface-assisted synthesis, but also transforms complex chemical processes through deep neural networks, allowing the synthesis precision of single-molecule operations to reach an unprecedented level.

Active learning is combined with first-principles calculations for catalyst screening and design

In recent years, Wang Xiaonan’s team has established a series of chemical materials design, synthesis and characterization evaluation The machine learning framework builds a high-throughput catalyst screening model based on active learning strategies, and simultaneously optimizes process parameters to achieve multi-scale precise design optimization. To address the difficult problem of integrating complex data and knowledge from the atomic level to the macro level, multi-scale digital twins and low-carbon intelligent connected systems are established.

In addition to the above-mentioned breakthroughs in basic research, a series of important applications for the main battlefield of the national economy have also been developed.

In terms of the design of catalysts for the selective hydrogenation of low-carbon alkynes, we collaborated with the team of Professor Duan Xuezhi of East China University of Science and Technology to achieve precise control of the Ni active site structure at the atomic scale, which not only provided orientation for the target reaction path control strategies, and promote the wide application of non-precious metal catalysts in the petrochemical industry.

The research team proposed a research method that combines an active learning framework based on Bayesian optimization with DFT calculations to determine the energy barrier difference between ethylene desorption and its further hydrogenation. As selectivity descriptors, a workflow for automated catalyst high-throughput screening was constructed for predicting high-performance Ni-based intermetallic compounds for the selective hydrogenation of acetylene.

Subsequently, 15 high-performance Ni-based intermetallic compounds were quickly screened from more than 3000 candidate Ni-based intermetallic compounds as potential alkyne hydrogenation catalysts, and DFT calculations were used to further verify the predictions of the ML model. accuracy, the recommended NiIn catalyst was finally determined as the optimal candidate catalyst for further experimental verification.

Catalytic reaction performance evaluation shows that: when the acetylene and propyne conversion rate of the NiIn intermetallic compound catalyst is 100%, the ethylene and propylene selectivity is as high as 97.0%, which is significantly higher than the reference catalyst, demonstrating the power of artificial intelligence. Huge potential in catalyst design.

Relevant results were published online in the Journal of the American Chemical Society under the title "Atomic Design of Alkyne Semihydrogenation Catalysts via Active Learning". A series of catalysts discovered are also being used in industry Enlarging and transforming.

Illustration: Catalyst screening design based on Bayesian optimization active learning framework combined with DFT calculation.

Paper link: https://pubs.acs.org/doi/full/10.1021/jacs.3c14495

AI Carbon Neutrality: Accelerating High-Performance Biochar Development, Improving CO₂ Capture Capacity

Wang Xiaonan’s team has long been paying attention to the field of AI carbon neutrality. In terms of research on carbon dioxide capture with biochar, we have co-founded the Pacific Rim University Alliance Sustainable Waste Management Project with collaborators from many countries to develop low-carbon, zero-carbon, and negative-carbon technologies to mitigate climate change while promoting sustainable waste management.

In view of the challenges of traditional biochar synthesis process being time-consuming, laborious and poor in accuracy, Wang Xiaonan’s team designed a customized active learning strategy that can guide and accelerate the synthesis of biochar and improve its adsorption of carbon dioxide. Ability.

This framework learns experimental data, recommends the best synthesis parameters, verifies the learning effect through experiments, and iteratively uses experimental data for subsequent model training and re-validation, thereby establishing a complete closed loop.

The research team ultimately synthesized 16 engineered biochar samples with specific properties, nearly doubling the amount of carbon dioxide absorbed in the final round. This study demonstrates a data-driven workflow that accelerates the development of high-performance engineered biochar materials.

The relevant results were published in the authoritative environmental journal "Environmental Science & Technology" under the title "Active Learning-Based Guided Synthesis of Engineered Biochar for CO₂ Capture" and was selected as the cover paper.

Illustration: Data-driven closed-loop development process for CO₂ capture materials.

Paper link: https://pubs.acs.org/doi/full/10.1021/acs.est.3c10922

Open up new avenues for scientific exploration and provide practical applications Strong Support

This series of research work is supported by projects such as the "New Generation Artificial Intelligence National Science and Technology Major Project" for which Professor Wang Xiaonan serves as the project leader and chief scientist.

Relevant results not only open up new avenues for scientific exploration, but also provide strong support for practical applications, especially showing great potential in promoting sustainable development and responding to global issues.

With the rapid progress of AI technology, its application prospects in intelligent chemical engineering, new material development, new energy technology and other fields are very broad, and will produce more innovative results.

The above is the detailed content of From material design and synthesis to catalyst innovation and carbon neutrality, Tsinghua Wang Xiaonan’s team explores the frontier and implementation of “AI+ materials”. For more information, please follow other related articles on the PHP Chinese website!