A prediction model for CO₂/CO adsorption performance on binary alloys based on machine learning.

Area of Science:

• Computational materials science
• Catalysis
• Machine learning applications

Background:

• Predicting catalytic material performance from atomic structure is challenging.
• Quantum mechanics methods (like DFT) are accurate but computationally expensive.
• Machine learning (ML) offers a faster alternative for screening catalytic materials.

Purpose of the Study:

• Develop an ML model to predict CO2 and CO adsorption affinity on single-atom doped binary alloys.
• Utilize thermochemical properties of component metals for predictions.
• Enhance the understanding of structure-property relationships in alloy catalysts.

Main Methods:

• A greedy algorithm was used to select the optimal features.
• An ML model was trained and validated using a dataset of 78 alloys.
• Adsorption energy values were calculated using Density Functional Theory (DFT).
• Extreme Gradient Boosting (XGBoost) algorithm was employed.

Main Results:

• The XGBoost model demonstrated excellent generalization performance.
• High R-squared values were achieved: 0.96 for CO2 and 0.91 for CO adsorption energy.
• Low prediction errors: 0.138 eV for CO2 and 0.075 eV for CO.
• Accurate prediction of adsorption affinity based on alloy composition.

Conclusions:

• The developed ML model accurately predicts CO2 and CO adsorption on doped binary alloys.
• This approach significantly accelerates the screening of potential alloy catalysts.
• The model advances fundamental understanding of structure-property relationships in catalysis.