Design of MnO₂-based catalysts with activity approaching Pt/C via machine learning

Area of Science:

• Materials Science
• Electrochemistry
• Computational Chemistry

Background:

• Manganese dioxide (MnO₂) is a key oxygen reduction reaction (ORR) catalyst for metal-air batteries, but its activity is limited compared to platinum.
• Developing highly active and cost-effective ORR catalysts is crucial for advancing battery technology.

Purpose of the Study:

• To employ machine learning to design a highly active bimetallic Ce/Ni co-doped δ-MnO₂ ORR catalyst.
• To achieve ORR performance comparable to platinum-based catalysts.

Main Methods:

• Constructed a multi-output regression model using ensemble learning algorithms.
• Identified Gradient Boosting as the optimal model after training with 64 experimental datasets.
• Predicted the optimal composition and ORR performance of Ce/Ni co-doped δ-MnO₂.

Main Results:

• The optimal composition predicted was 7.5Ce/15Ni-MnO₂, with predicted onset potential (E_onset) of 0.893 V, half-wave potential (E_1/2) of 0.830 V, and limiting current density (J_L) of 4.504 mA cm^-2.
• Experimental validation confirmed these values, with measured E_onset at 0.890 V, E_1/2 at 0.828 V, and J_L at 4.56 mA cm^-2.
• The experimental results closely matched the machine learning predictions.

Conclusions:

• Machine learning effectively guided the design of an advanced MnO₂-based ORR catalyst.
• The developed Ce/Ni co-doped δ-MnO₂ catalyst exhibits high activity, approaching that of Pt/C.
• This study demonstrates a viable machine learning-assisted strategy for developing next-generation ORR catalysts.