Predicting and Analyzing Nitrate Adsorption on High-Entropy Alloys Based on Pair Distribution Function Using a Hybrid...

Area of Science:

• Materials Science
• Catalysis
• Computational Chemistry

Background:

• High-entropy alloys (HEAs) offer exceptional stability and electronic properties, making them promising catalysts.
• The complex design space of HEAs necessitates machine learning for optimizing catalytic performance.
• Structural features are crucial for machine learning accuracy in HEA catalyst development.

Purpose of the Study:

• To investigate the use of pair distribution function (PDF) data as input features for machine learning-based optimization of HEA catalysts.
• To develop a hybrid machine learning framework for predicting catalytic activity using PDF data.
• To assess the performance of this framework compared to conventional machine learning algorithms.

Main Methods:

• Principal Component Analysis (PCA) was employed to reduce the dimensionality of PDF data.
• A hybrid framework combining a transformer-based model and a large language model (LLM) was utilized.
• The framework predicted the Gibbs free energy of nitrate adsorption on FeCoNiCuZn HEA surfaces.

Main Results:

• The hybrid framework accurately predicted Gibbs free energy of nitrate adsorption using PCA-reduced PDF data.
• Performance significantly surpassed conventional algorithms like random forest, support vector regression, and gradient boosting.
• LLM integration further improved prediction accuracy and provided interpretable insights.

Conclusions:

• PCA-reduced PDF data can serve as effective input features for machine learning models in HEA catalyst design.
• The developed hybrid transformer-LLM framework enables accurate prediction and optimization of HEA catalysts.
• This approach facilitates the predictive design of HEA-based catalysts with enhanced activity and selectivity.