Accelerated discovery of multi-elemental reverse water-gas shift catalysts using extrapolative machine learning...

Area of Science:

• Catalysis
• Materials Science
• Data Science

Background:

• Designing novel catalysts is crucial for addressing energy and environmental issues.
• Machine learning (ML) offers potential for accelerating catalyst development but often struggles with discovering truly novel materials due to extrapolation limitations.

Purpose of the Study:

• To demonstrate an extrapolative machine learning (ML) approach for discovering new multi-elemental reverse water-gas shift catalysts.
• To overcome the common ML limitation of failing to identify extraordinary or unprecedented materials.

Main Methods:

• Utilized a closed-loop discovery system involving 44 cycles of ML prediction and experimental testing.
• Started with an initial dataset of 45 catalysts and experimentally evaluated a total of 300 catalysts.
• Employed an ML approach designed for extrapolation to explore beyond the initial data space.

Main Results:

• Identified over 100 catalysts exhibiting superior activity compared to existing high-performance catalysts.
• Discovered an optimal catalyst with the composition Pt(3)/Rb(1)-Ba(1)-Mo(0.6)-Nb(0.2)/TiO2.
• Successfully identified a novel catalyst containing niobium (Nb), an element absent from the original dataset and unpredictable by experts.

Conclusions:

• The developed extrapolative ML approach is effective in discovering novel, high-performance multi-elemental catalysts.
• This method demonstrates the capability of ML to identify extraordinary materials beyond expert predictions and initial datasets.
• The findings pave the way for accelerated discovery of advanced catalysts for energy and environmental applications.