Graph Neural Network Machine Learning Aided Design of Anti-MXene-Based Single-Atom Catalysts for Hydrogen Evolution Reaction
Ivan Zhang
Camberwell Grammar School, Melbourne, Victoria, 3126, Australia
Publication date: January 21, 2025
Camberwell Grammar School, Melbourne, Victoria, 3126, Australia
Publication date: January 21, 2025
DOI:http://doi.org/10.34614/JIYRC202435
ABSTRACT
The potential use of catalysts to efficiently produce green hydrogen by means of electrochemical hydrogen evolution reactions has sparked interest. Of the explored groups of catalysts, a fairly recent branch, anti-MXenes, has provided great promise. This research aims to establish a machine learning-based model to accelerate the design and optimisation of anti-MXene-based single-atom catalysts. A pre-trained material 3-body graph network (M3GNet) was used to predict the adsorption energy for hydrogen of 667 single-atom catalysts. Subsequently, the machine learning model was employed to screen 132 single-atom catalysts whose energy was in the range of -0.35 to -0.05 eV. Consequently, four catalysts, CoS-Zr, RhAs-Ti, IrB-Ru, and CrAs-Re that displayed optimal activity were finalised and further validated by density functional theory calculations. This work demonstrates the strong capability of machine learning in accelerating the design and optimisation of catalysts for the production of green hydrogen, one of the most promising and clean energy sources for the future Net Zero Emission era.
The potential use of catalysts to efficiently produce green hydrogen by means of electrochemical hydrogen evolution reactions has sparked interest. Of the explored groups of catalysts, a fairly recent branch, anti-MXenes, has provided great promise. This research aims to establish a machine learning-based model to accelerate the design and optimisation of anti-MXene-based single-atom catalysts. A pre-trained material 3-body graph network (M3GNet) was used to predict the adsorption energy for hydrogen of 667 single-atom catalysts. Subsequently, the machine learning model was employed to screen 132 single-atom catalysts whose energy was in the range of -0.35 to -0.05 eV. Consequently, four catalysts, CoS-Zr, RhAs-Ti, IrB-Ru, and CrAs-Re that displayed optimal activity were finalised and further validated by density functional theory calculations. This work demonstrates the strong capability of machine learning in accelerating the design and optimisation of catalysts for the production of green hydrogen, one of the most promising and clean energy sources for the future Net Zero Emission era.
