Distilling Accurate Descriptors from Multi-Source Experimental Data for Discovering Highly Active Perovskite OER...

Area of Science:

• Materials Science
• Catalysis
• Computational Chemistry

Background:

• Perovskite oxides show promise as catalysts for the oxygen evolution reaction (OER).
• Exploring the vast chemical space of perovskites for OER catalysts is challenging due to a lack of effective discovery methods.
• Inconsistent data from multiple experimental sources hinders accurate descriptor development.

Purpose of the Study:

• To develop a novel computational approach for accelerated discovery of perovskite oxide catalysts for OER.
• To address the challenge of data inconsistency from multi-source experimental datasets.
• To identify new, highly active perovskite catalysts for OER.

Main Methods:

• Implemented a sign-constrained multi-task learning method within the sure independence screening and sparsifying operator framework.
• Integrated 13 experimental datasets from diverse publications to derive accurate descriptors.
• Developed a new 2D descriptor (d_B, n_B) for predicting catalytic activity.

Main Results:

• The new 2D descriptor (d_B, n_B) demonstrates high universality, predictive accuracy, and bulk-surface correspondence.
• Hundreds of unreported perovskite candidates with predicted OER activity exceeding the benchmark (Ba0.5Sr0.5Co0.8Fe0.2O3) were identified.
• Experimental validation confirmed three highly active perovskite catalysts: SrCo0.6Ni0.4O3, Rb0.1Sr0.9Co0.7Fe0.3O3, and Cs0.1Sr0.9Co0.4Fe0.6O3.

Conclusions:

• The developed method effectively handles inconsistent multi-source data for accelerated catalyst discovery.
• The new 2D descriptor provides a powerful tool for identifying high-performance perovskite OER catalysts.
• This data-driven approach significantly advances the field of catalysis and offers potential for broader applications.