An AI-Powered Methodology for Atomic-Scale Analysis of Heterogenized Correlated Single-Atom Catalysts

Area of Science:

• Catalysis
• Materials Science
• Nanotechnology

Background:

• Correlated single-atom catalysts are crucial for electrocatalysis, especially oxygen evolution reactions.
• Traditional characterization methods are insufficient for dispersed catalysts on carrier materials.
• Understanding atomic-scale properties requires advanced analytical techniques.

Purpose of the Study:

• To develop a methodology for detecting and quantifying metal-metal interactions in heterobinuclear complexes.
• To apply machine learning and mathematical optimization to HAADF-STEM images.
• To study the dynamics and bond interactions of gold-palladium complexes.

Main Methods:

• Combining machine learning (U-net architecture) and mathematical optimization.
• Analyzing atomically resolved HAADF-STEM images of Au(III)-Pd(II) macrocyclic complexes.
• Evaluating both supervised and unsupervised machine learning approaches.

Main Results:

• The U-net architecture excelled at distinguishing gold and palladium species.
• Mathematical optimization provided precise distance metrics for metal pairs.
• Most metal pairs remained stable under electron beam irradiation, with unchanged Au-Pd distances.

Conclusions:

• The developed methodology reliably detects and quantifies metal-metal interactions in single-atom catalysts.
• Heterobinuclear Au(III)-Pd(II) complexes exhibit robust metal-ligand interactions and stability on amorphous carbon substrates.
• This approach enables detailed study of catalyst dynamics and bonding at the atomic scale.