Transferable, Living Data Sets for Predicting Global Minimum Energy Nanocluster Geometries

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Nanocluster geometry is crucial for understanding catalytic activity and active sites.
• Heterogeneous catalysis involves metal-support interactions that alter cluster properties.
• Computational costs for modeling diverse nanocluster shapes are often prohibitive.

Purpose of the Study:

• To develop data-driven methodologies for efficient nanocluster geometry modeling.
• To reduce the computational expense of exploring structural configurations in catalysis.
• To leverage transferable structural information across different metal systems.

Main Methods:

• Construction of a 'living' data set using overlapping cluster shapes from various systems.
• Iterative refinement of the data set for initializing new cluster studies.
• Application of data-driven approaches to reduce model construction costs.

Main Results:

• Utilization of transferable structural information reduced model construction costs by over 90%.
• The methodology significantly benefits the study of alloy systems with large configurational spaces.
• A low-cost pathway for initializing nanocluster studies was established.

Conclusions:

• Data-driven methods and transferable structural data offer a highly efficient approach to nanocluster modeling.
• This strategy substantially decreases computational overhead in catalysis research.
• The findings are particularly impactful for complex alloy catalyst development.