Developing Cheap but Useful Machine Learning-Based Models for Investigating High-Entropy Alloy Catalysts.

Area of Science:

• Materials Science
• Computational Chemistry
• Catalysis

Background:

• Developing interpretable machine learning (ML) models often requires extensive computational resources.
• High-entropy alloy (HEA) catalysts offer unique properties but their application can be limited by computational cost.

Purpose of the Study:

• To present a cost-effective workflow for developing ML-based models for catalyst design.
• To predict high-quality adsorption energies for various adsorbates on CoMoFeNiCu HEA catalysts.
• To demonstrate resource-efficient methods applicable to the broader surface catalysis community.

Main Methods:

• Synergistic combination of descriptor-based approaches, ML-based force fields (ML-FF), and low-cost density functional theory (DFT) calculations.
• Implementation of three specific modifications to typical DFT workflows: sequential optimization, a new geometry-based descriptor, and repurposing DFT trajectories for ML-FF development.
• Prediction of adsorption energies for H, N, and NHx (x=1, 2, 3) adsorbates.

Main Results:

• Achieved high-quality adsorption energy predictions using a cost-effective workflow.
• Demonstrated the efficacy of combining descriptor-based methods with ML-FFs trained on low-cost DFT data.
• Successfully predicted adsorption energies for key adsorbates on CoMoFeNiCu HEA catalysts.

Conclusions:

• Cost-effective DFT calculations and well-designed descriptors can yield accurate predictive models for adsorption energies.
• The developed workflow significantly reduces computational costs for ML model development in catalysis.
• This resource-efficient philosophy is broadly relevant for advancing surface catalysis research.