Machine learning on frontier orbital energy of atomically precise gold nanoclusters

Area of Science:

• Computational chemistry and materials science
• Nanotechnology and catalysis

Background:

• Atomically precise gold nanoclusters (Au NCs) show great potential in catalysis.
• Accurate prediction of their electronic states is crucial but computationally expensive.
• Existing methods face challenges due to high experimental and computational costs.

Purpose of the Study:

• To develop a cost-efficient machine learning (ML) model for predicting electronic states of Au NCs.
• To identify key structural descriptors that govern the electronic properties of Au NCs.
• To enable accurate prediction of Highest Occupied Molecular Orbital (HOMO), Lowest Unoccupied Molecular Orbital (LUMO), HOMO-LUMO gap, and oxidation potential (OP).

Main Methods:

• Developed an ML model utilizing interpretable automated feature engineering.
• Screened 227 candidate parameters using bi-directional stepwise regression and a Kolmogorov-Arnold network model.
• Trained the model on 79 data points and validated on 20 data points, using only 4 key descriptors.

Main Results:

• Achieved mean average errors (MAE) of 0.17 eV (HOMO), 0.27 eV (LUMO), and 0.16 eV (HOMO-LUMO gap) on the testing set.
• Predicted oxidation potential (OP) with an MAE of 0.20 V.
• Identified the number of cluster charges (NC) and average Au-Au coordination numbers (CNAu-Au) as critical descriptors.

Conclusions:

• A small set of key structural descriptors enables accurate and cost-efficient prediction of Au NC electronic structures.
• The developed ML model provides a valuable tool for accelerating the design and application of Au NCs in catalysis.
• This approach significantly reduces the computational and experimental burden associated with characterizing Au NCs.