Gold-Nanoclusters

In this study, 99 gold nanocluster (Au NC) structures were collected to construct a comprehensive dataset, covering the size range from Au₄ to Au₆₀ and including 41 core structures with 15 types of ligands. Geometry optimizations of all structures were performed using the DMol³ module within the Materials Studio software suite, employing the Generalized Gradient Approximation (GGA) with the Perdew–Burke–Ernzerhof (PBE) functional. The convergence criteria for energy, gradient, and displacement were set to 2 × 10⁻⁵ Hartree, 4 × 10⁻³ Hartree/Å, and 5 × 10⁻³ Å, respectively. Based on the optimized structures, single-point energy calculations were carried out using the ADF package with the COSMO model (ethanol as solvent), applying the PBE functional and the TZP basis set. Through this workflow, the HOMO and LUMO orbital energies of all 99 clusters were obtained. The dataset was then divided into training (80%) and testing (20%) subsets. Structural features were extracted using a core–shell model and refined via bidirectional stepwise regression (BSR) to optimize the parameter set. Finally, the KAN model was trained and subsequently validated to assess its accuracy in predicting HOMO and LUMO energies, HOMO–LUMO gaps, and relative oxidation potentials.