Can AI Learn What Makes Excellent Catalyst?

Researchers have long sought ways to reduce the amount of platinum needed while maintaining high performance. Promising candidates, such as platinum-nickel (Pt-Ni) alloys, have already been identified. Yet determining which elements should be combined-and in what proportions-to create the best catalyst remains a difficult challenge. The number of possible material combinations is enormous, making it impractical to evaluate them one by one.

Another challenge is that a useful catalyst must achieve two important goals at the same time: high activity, meaning it can accelerate reactions efficiently, and high stability, meaning it can maintain its performance over long periods. A catalyst that performs well initially but quickly degrades cannot be used in real applications. Balancing activity and stability has proven difficult, even in AI-assisted materials discovery.

To address this challenge, a research team led by Associate Professor Atsushi Ishikawa and doctoral student Taishiro Wakamiya of Institute of Science Tokyo (Science Tokyo) set out to develop a method that enables AI to propose promising catalyst materials with both high activity and durability."

As the team began their work, they encountered a major obstacle: there were simply too many possible catalyst materials to examine. Performing high-accuracy calculations for every candidate would require an enormous amount of time and computational resources. To overcome this problem, the researchers developed a workflow in which AI proposes promising candidates and then learns from the evaluation results. This approach allowed the team to search efficiently through vast numbers of possibilities while keeping computational costs manageable.

Something even more interesting happened during the process. The researchers never told the AI what kind of atomic arrangement would lead to a high-performance catalyst. Nevertheless, through repeated cycles of learning and exploration, the AI began to identify patterns on its own. It discovered features associated with high-performance catalysts and proposed structures that closely resembled those that researchers had previously identified as promising.

In other words, the AI was doing more than simply screening large numbers of candidates. It was uncovering clues about what kinds of atomic structures are likely to make effective catalysts. Researchers can use these insights to guide materials development more efficiently than before.

The method developed in this study could be applied far beyond fuel cell catalysts. It represents an important step toward a concept known as inverse design, in which researchers specify the properties they want and AI helps identify material structures that can achieve them.

In the future, this approach could accelerate the discovery of battery materials, chemical catalysts, and advanced materials used in aircraft and automobiles. It may also help reduce the use of scarce resources such as platinum by identifying more efficient material designs. As AI becomes an increasingly powerful research partner, it may enable scientists to discover materials that would otherwise remain hidden among countless possibilities.

When searching through an enormous number of possible materials, human experience alone inevitably has its limits. AI is not a magical tool that automatically gives us the answers, but it can offer perspectives that humans might overlook. By connecting catalysis, simulation, and AI, I hope to continue exploring new possibilities in materials research.

(Taishiro Wakamiya, Doctoral Student, School of Environment and Society, Institute of Science Tokyo)

Taishiro joined our laboratory after entering the doctoral program and quickly demonstrated strong skills in coding and computational research. I hope this project will contribute to his growth as a researcher. AI is not a replacement for researchers; rather, it is a new tool that can expand our creativity and ways of thinking. I hope this study serves as one example of that potential.

(Atsushi Ishikawa, Associate Professor, School of Environment and Society, Institute of Science Tokyo)

Science for All showcases cutting-edge research at Science Tokyo and highlights the ideas, people, and possibilities shaping the future.