Tokyo Institute Of Technology: Empirical Workflow Using Machine Learning Reveals Design Guidelines for Ternary Metal Sulfides

One of the most promising avenues for actively reducing CO₂ levels in the atmosphere is recycling it into valuable chemicals via electrocatalytic CO₂ reduction reactions. With a suitable electrocatalyst, this can be achieved under mild conditions and at a low energy cost. Many types of electrocatalysts are being actively investigated, but most suffer from either low electrocatalytic activity, poor selectivity, or low stability.

Metal sulfides might hold the huge potential solution to this puzzle. By combining ionic and covalent characteristics, this unique family of materials offers good catalytic activity and energy efficiency. The ternary metal system is expected to be a better solution since, according to recent studies, simple metal sulfides can still only yield a few simple carbon compounds in CO₂ reduction reactions, thus lacking versatility. However, there is still a very low number of publications that discuss the functionality of ternary metal sulfide as a CO₂ reduction electrocatalyst.

Against this backdrop, a research team led by Assistant Professor Akira Yamaguchi from Tokyo Institute of Technology, Japan, gave their effort to study the trend of ternary metal sulfide that has not been reported elsewhere . In their latest study, which was published in Materials Science & Engineering R: Reports, they combined experimental data analysis and machine learning to gain insights into this uncharted territory in materials science.

“Ternary metal sulfides may offer synergistic bi-metal effects that enhance CO₂ reduction performance. However, these materials possess complex electronic structures, and it is difficult to employ their adsorption energy for intermediate compounds to analyze electrocatalyst performance trends of different metals and alloys,” explains Yamaguchi.

To overcome these challenges, the researchers developed a novel screening methodology. Unlike previous screening methods, which often involve computationally expensive calculations of electrocatalyst adsorption energies, the researchers focused on analyzing more easily measurable and computable material properties derived from experiments and data analysis.

Using experimental data obtained from various measurements of their own synthesized metal sulfide samples, the researchers calculated a set of material properties representing structural, bulk, and surface parameters. They also measured the electrochemical CO₂ reduction activity of the materials. Additionally, they used four different high-dimensional regression algorithms within the machine learning models to unveil possible relationships between material properties and electrocatalytic performance.

In this way, the researchers devised a streamlined workflow that can identify important parameters to explain the origin of high activity in electrocatalytic materials. One of the main findings of this study was that focusing on the crystal structure of ternary metal sulfides leads to better results than focusing solely on their elemental composition. “Our approach is less burdensome than other screening techniques and does not require high-throughput experimental tools. In addition, it is generalizable and applicable to many materials, making it particularly beneficial given the limited availability of material activity data for CO₂ reduction reactions,” explains Yamaguchi.

The research team hopes their efforts will lead to effective design guidelines for the development of CO₂ conversion catalysts using materials that are ubiquitous in nature, and also the possible application of their guidelines to the other research field.