Deep Reinforcement learning generation and optimization of AgPd nanoclusters

Abstract The determination and prediction of atomic cluster structures are crucial for nanocluster and materials research, as the molecular structures directly influence the properties of nanoclusters. Traditional quantum chemical calculations are often time-consuming, and structural prediction approaches tend to be computationally expensive. Determining the most stable configurations of nanoclusters is particularly challenging due to their complex energy landscapes and the existence of numerous metastable states. In this study, we employed a novel Deep Reinforcement Learning (DRL) method to investigate the potential energy surfaces (PES) and identify the global minima (GM) structures of Ag-Pd nanoclusters. The DRL agent, through an iterative process of trial and error, and reward feedback, dynamically adapts to the increasing complexity of the energy landscapes. By exploring a wide range of possible configurations, the DRL agent successfully identifies global minima even in the face of high-dimensional and complex PES landscapes. The model’s ability to balance the exploration of new configurations and the exploitation of known low-energy states allows it to efficiently navigate the intricate energy landscapes of nanoclusters. The DRL agent identifies stable configurations faster than traditional optimization algorithms by refining its strategy through continuous learning. To validate the results, density functional theory (DFT) calculations were performed on all nanocluster configurations identified as global minima. The results indicate that all Ag-Pd nanoclusters have negative binding energies, suggesting their stability. Furthermore, the inclusion of Pd has a significant impact on the electronic characteristics of the nanoclusters, as demonstrated by the analysis of the electronic density of states (DOS). Charge analysis, specifically through Bader charge calculations, shows a clear redistribution of charge from Ag to Pd, confirming charge transfer and polarization effects, especially as the Pd content increases. Additionally, optical absorption spectra reveal that Ag-Pd nanoalloys with varying Pd contents exhibit tunable absorption properties across the UV–visible spectrum. This study highlights the critical role of metal composition in determining the structural, electronic, and optical properties of nanoclusters, providing a robust DRL framework for designing stable and versatile nanomaterials. The integration of DRL and DFT provides an efficient methodology for optimizing nanocluster properties, and advancing the development of functional metallic nanostructures.