Combining artificial intelligence and physics-based modeling to directly assess atomic site stabilities: from...

Area of Science:

• Materials Science
• Computational Chemistry
• Nanotechnology

Background:

• Material performance relies on atomic site properties.
• Site stability is key for catalyst design, traditionally requiring intensive computation.
• Predicting stability in complex nano-materials is challenging.

Purpose of the Study:

• To present a machine learning (ML) approach for rapid computation of atomic site stability in nano-materials.
• To enable accurate prediction of site stability in sub-nanometer metal clusters.
• To extract physical insights and reveal structure-property relationships using a genetic algorithm (GA).

Main Methods:

• Developed a widely applicable machine learning (ML) model for instantaneous computation of atomic site stability.
• Applied a genetic algorithm (GA) for feature selection to interpret ML models.
• Combined ML with physics-based models for real-time stability predictions across various material scales.

Main Results:

• Predicted atomic site stability in 3-55 atom metal clusters with mean absolute errors of 0.11-0.14 eV.
• Identified key structural and chemical properties governing site stability via GA.
• Achieved real-time prediction for structures from sub-nanometer clusters to extended surfaces.

Conclusions:

• The ML approach provides a rapid and accurate method for predicting atomic site stability.
• The GA enhances model interpretability and reveals crucial structure-property relationships.
• The combined framework enables the design of stable and active nanocatalysts.