Hierarchical Embedded Sphere Model: An Interpretable ML-Guided Multiscale Descriptor Engineering Decodes OER Activity...

Area of Science:

• Materials Science
• Computational Chemistry
• Catalysis

Background:

• Predicting catalytic activity in transition-metal oxides is complex due to intertwined geometric and electronic factors.
• Developing generalizable descriptors for oxide electrocatalysis is crucial for catalyst design.

Purpose of the Study:

• To introduce a Hierarchical Embedded Sphere Model (HESM) integrating density functional theory (DFT) and interpretable machine learning (IML).
• To establish a generalizable descriptor framework for oxide electrocatalysis.
• To disentangle catalytic activity into hierarchical contributions.

Main Methods:

• Density Functional Theory (DFT) calculations.
• Interpretable Machine Learning (IML) models.
• Hierarchical Embedded Sphere Model (HESM) for analyzing transition metal-doped anatase MO2 (101) surfaces.

Main Results:

• HESM decomposes catalytic activity into Global electronic structure (G-class), atomic-site intrinsic properties (A-class), and local coordination (L-class).
• Identified two activation paradigms: dopant-induced electronic modulation (Rh@MO2) and host-site coordination tuning (Fe@ZrO2).
• SHapley Additive exPlanation (SHAP) analysis revealed G-class as a key predictive feature.

Conclusions:

• HESM provides a generalizable methodology for descriptor discovery and catalyst design in complex oxide systems.
• The model successfully bridges adsorption energetics with multiscale geometric-electronic couplings.
• This framework reconciles activity trends and explains site-dependent deviations in catalysis.