Searching for an Optimal Multi-Metallic Alloy Catalyst by Active Learning Combined with Experiments

Area of Science:

• Catalysis research
• Materials science
• Nanotechnology

Background:

• Optimizing multi-metallic alloy catalysts is crucial but challenging due to extensive data needs and high experimental costs.
• Current methods often combine density functional theory with machine learning (ML).

Purpose of the Study:

• To overcome limitations in catalyst discovery by integrating experiment and active learning.
• To develop an efficient method for identifying optimal catalyst components and compositions.

Main Methods:

• An active learning model was iteratively updated by evaluating the electrocatalytic performance of fabricated solid-solution nanoparticles.
• The model used precursor mixture composition as input data for searching optimal catalysts.

Main Results:

• An optimal metal precursor composition of Pt$_{0.65}$Ru$_{0.30}$Ni$_{0.05}$ was identified.
• This catalyst exhibited a hydrogen evolution reaction (HER) overpotential of 54.2 mV, outperforming pure Pt.
• The model successfully improved overpotential using only precursor composition data.

Conclusions:

• The active learning approach effectively searches for optimal multi-metallic alloy catalysts.
• This method is widely applicable for determining optimal electrocatalyst components and compositions without significant restrictions.
• The study demonstrates a significant step towards efficient and cost-effective catalyst design.