Locating Hydrides in Ligand-Protected Copper Nanoclusters by Deep Learning

Area of Science:

• Nanomaterials Science
• Computational Chemistry
• Artificial Intelligence in Chemistry

Background:

• Hydrides are crucial components in metal nanoclusters and nanoparticles, occupying interstitial and interfacial sites.
• Determining precise hydride locations is challenging due to limitations in neutron diffraction techniques.
• Existing deep-learning methods for hydride site determination were limited to smaller nanocluster sizes.

Purpose of the Study:

• To develop and validate an improved deep-learning model for accurate hydride site determination in metal nanostructures.
• To assess the model's applicability to significantly larger and recently discovered nanoclusters.
• To investigate the hydride configurations in specific copper nanoclusters where neutron diffraction data is unavailable.

Main Methods:

• Development of an improved deep-learning model based on convolutional neural networks (CNNs).
• Application of the CNN model to two copper nanoclusters: [Cu32(PET)24H8Cl2]2- and [Cu81(PhS)46(tBuNH2)10H32]3+.
• Comparison of CNN model predictions with existing density functional theory (DFT) calculations.

Main Results:

• The improved CNN model demonstrated high accuracy and robustness in predicting hydride locations.
• For [Cu32(PET)24H8Cl2]2-, the CNN model confirmed the hydride sites proposed by DFT calculations.
• For [Cu81(PhS)46(tBuNH2)10H32]3+, the CNN model predicted a more stable structure with distinct hydride site occupancies compared to DFT.

Conclusions:

• The developed deep-learning model is effective for determining hydride locations in large copper nanoclusters.
• This AI-driven approach overcomes limitations of traditional experimental methods for hydride site identification.
• The findings contribute to a more precise understanding and design of complex metal nanostructures.