Computational catalyst discovery: Active classification through myopic multiscale sampling.

Area of Science:

• Computational chemistry
• Materials science
• Catalysis research

Background:

• Computational chemistry enables materials and catalyst discovery.
• Challenges include scaling atomic-scale calculations (e.g., DFT) to macro-scale properties and balancing optimization with exploration in early-stage discovery.

Purpose of the Study:

• To address challenges in computational catalyst discovery.
• To develop a method for estimating bulk catalyst activity using a multi-scale model.
• To reframe catalyst discovery as a classification problem: identifying catalysts worth experimental testing.

Main Methods:

• Developed a multi-scale model combining Density Functional Theory (DFT) and machine learning for adsorption energy predictions.
• Introduced myopic multiscale sampling, an active classification strategy.
• Classified catalysts as "worth investigating" or "not worth investigating" experimentally.

Main Results:

• Achieved a 7-16 times speedup in catalyst classification compared to random sampling.
• Validated the method through offline simulations on synthesized and real datasets.

Conclusions:

• Myopic multiscale sampling offers an efficient approach to computational catalyst discovery.
• The classification strategy effectively prioritizes catalysts for experimental validation, accelerating the discovery process.