Distilling universal activity descriptors for perovskite catalysts from multiple data sources via multi-task symbolic...

Area of Science:

• Materials Science
• Electrochemistry
• Machine Learning

Background:

• Developing efficient electrocatalysts is crucial for energy applications.
• Existing data for electrocatalyst activity is often incomparable due to varied experimental conditions.
• Machine learning (ML) offers a path to accelerate materials discovery.

Purpose of the Study:

• To develop a universal, interpretable ML-based activity descriptor for oxide perovskites.
• To enable accurate prediction of oxygen evolution reaction (OER) performance.
• To overcome data incomparability issues from diverse literature sources.

Main Methods:

• Utilized multi-task symbolic regression, an interpretable ML approach.
• Learned from heterogeneous datasets across multiple experiments.
• Integrated Bayesian optimization with the developed descriptor for parameter tuning.

Main Results:

• Constructed a highly accurate and generalizable universal activity descriptor for OER.
• Predicted promising double perovskites with excellent OER activity.
• Successfully synthesized and experimentally validated two ML-predicted nickel-based perovskites.

Conclusions:

• The developed descriptor accurately predicts OER performance of oxide perovskites without experiments or calculations.
• This ML approach effectively utilizes multiple data sources for materials design.
• Opens new avenues for accelerating the discovery of advanced electrocatalysts.