Predicting Segregation Energy in Single Atom Alloys Using Physics and Machine Learning.

Area of Science:

• Materials Science and Engineering
• Computational Chemistry
• Catalysis

Background:

• Single atom alloys (SAAs) offer tunable properties for enhanced catalytic activity and selectivity.
• Accurate prediction of heterometal dopant stability on SAA surfaces is crucial for catalyst design.
• Traditional methods like Density Functional Theory (DFT) for calculating surface segregation energy are computationally expensive.

Purpose of the Study:

• To develop an accelerated machine learning framework for predicting surface segregation energy in SAAs.
• To enable rapid screening of metal segregation for new SAA catalyst combinations.
• To provide a general tool for predicting SAA segregation energies across different structures and compositions.

Main Methods:

• Developed a machine learning model using elemental descriptors and bond-centric features to predict surface segregation energy.
• Utilized second-order polynomial kernel ridge regression for the final five-feature model.
• Trained and validated the model on periodic slabs and applied it to nanoparticles.

Main Results:

• The machine learning model accurately predicts surface segregation energy for various FCC-based SAAs.
• The five-feature model demonstrates high accuracy due to physically motivated features.
• The model successfully predicts DFT segregation energies for SAA nanoparticles, even when trained on periodic slabs.

Conclusions:

• The developed machine learning model serves as a rapid and accurate tool for predicting SAA segregation energies.
• This framework accelerates SAA catalyst design by enabling efficient screening of metal segregation.
• The study highlights the physicochemical properties driving the thermodynamic stabilization of SAAs.