In a collaborative effort, researchers from various institutions, including the Federal University of Minas Gerais, the University of São Paulo, and the University of California, Davis, have developed a novel approach to predict the stability of bimetallic nanoclusters. These nanoclusters, which are tiny particles of metal atoms, have potential applications in catalysis and energy conversion, making this research highly relevant to the energy industry.
The team combined density functional theory (DFT) and artificial intelligence to study 13-atom icosahedral nanoclusters. They focused on hosts made of titanium, zirconium, or hafnium, with a single transition-metal dopant spanning the 3d-5d series. The researchers performed spin-polarized DFT calculations on 240 bimetallic clusters to understand the trends in binding and formation energies, distortion penalties, effective coordination number, d-band center, and HOMO-LUMO gap. These factors help determine whether the dopant atom prefers to be in the core (in) or on the surface (out) of the nanocluster.
To predict formation energies and the preference for core-shell or surface-segregated arrangements, the researchers trained a transformer architecture, a type of artificial intelligence model. They first pretrained the model on a curated set of 2968 unary clusters from the Quantum Cluster Database and then fine-tuned it on the bimetallic data. The model achieved mean absolute errors of about 0.6-0.7 eV and provided calibrated uncertainty intervals. Notably, the model could rapidly adapt to an unseen Fe-host domain with only a few labeled examples.
The researchers also analyzed attention patterns and Shapley attributions to identify key descriptors influencing the model’s predictions. These included size mismatch, d-electron count, and coordination environment. All data, code, and workflows followed FAIR/TRUE principles, ensuring reproducibility and interpretability.
This research, published in the journal Nature Communications, offers a powerful tool for screening unexplored nanocluster chemistries. By rapidly and accurately predicting the stability and reactivity of bimetallic nanoclusters, this approach can significantly accelerate the development of new materials for catalysis and energy conversion applications. This could lead to more efficient and cost-effective processes in the energy industry, such as improved catalysts for fuel cells or more effective materials for energy storage.
This article is based on research available at arXiv.