Build a Ruler to Measure Electron Localization Function with Machine Learning Models: S-Au bonds Quantification in Thiolate Protected Gold Nanoclusters

Xinxu Zhang,Hui Jia,Fangzhen Tian,Yandi Liu,Ullrich Carsten,Yulong Wu,Changlong Liu,Xiaodong Zhang,Yonghui Li

2023-05-13

Abstract:Complicated S-Au interaction patterns in thiolate-protected gold nanoclusters (TP-AuNCs) are important in the formation of versatility among clusters. In this study, the ELFnet, a novel convolutional neural network (CNN) model is trained and tested to bridge the electron localization function (ELF) images and the bond lengths. As believed to be a successful bond describer, a dataset of 3959 ELF images is obtained out of Density Functional Theory (DFT) simulations, followed by a series of model architecture and hyperparameter exploration. The ELFnet with its best performance shows its correct feature extraction ability to gain insights in the prediction of bond lengths. Besides, the ELFnet also possess 3 different "ELF reading modes" which inspires other possibilities in understanding ELF images. The ELFnet serves as an "ELF-ruler" to give precise bond type labels (or the predicted bond lengths). Such ELF-ruler measured bond types represent a particular bond when distorted by external factors with a quantitative indicator. With the help of such an ELF-ruler, the classification of chemical bonds in general may be extended to more precise types.

Mesoscale and Nanoscale Physics

What problem does this paper attempt to address?

The problem this paper attempts to address is: how to accurately measure the length of S-Au bonds in thiolate-protected gold nanoclusters (TP-AuNCs) using machine learning models (specifically convolutional neural networks), thereby improving the understanding and classification accuracy of these chemical bonds.

Specifically, the paper focuses on the following aspects:
1. **Complexity of S-Au Bonds**: In thiolate-protected gold nanoclusters, the interaction patterns of S-Au bonds are very complex, which is crucial for the diversity and stability of the clusters.
2. **Application of Electron Localization Function (ELF)**: The Electron Localization Function (ELF) is an effective tool for describing chemical bonds, but its graphical display method limits the interpretation of its details.
3. **Development of Machine Learning Models**: The paper proposes a new convolutional neural network model (ELFnet) to extract features from ELF images and predict the length of S-Au bonds.
4. **Dataset Construction**: A total of 3959 ELF images were generated through density functional theory (DFT) simulations, and data augmentation and preprocessing were performed to improve the training effect of the model.
5. **Model Performance Evaluation**: The performance of ELFnet was validated through various experiments, including cross-validation, feature visualization, and occlusion experiments, demonstrating the model's high accuracy and robustness.
6. **Application Examples**: ELFnet was applied to the intermediate chemical states of oxygen atom adsorption on gold nanoclusters, showcasing its practical application value in catalytic processes.

In summary, this paper aims to provide a high-precision "ELF ruler" for accurately measuring and classifying the length of S-Au bonds by developing and applying the ELFnet model, thereby offering new tools and methods for understanding the electronic structure and chemical bond properties of thiolate-protected gold nanoclusters.