Accelerating\nGlobal Search of Large-Sized Silver Clusters\nUsing Cluster Graph Attention Network

Great efforts have been devoted to understanding the\nstability\nand reactivity of silver clusters, which usually depend on geometric\nstructures, electronic configuration, and cluster size. Despite the\nfact that the jellium model and Wulff construction rule have successfully\nrationalized the stable clusters with “magic number”\nbehavior, some experiments imply that silver clusters with 48 valence\nelectrons also possess puzzling enhanced stability. In this work,\nusing a recently developed deep learning technology, i.e., cluster\ngraph attention network (CGANet), combined with a homemade comprehensive\ngenetic algorithm (CGA) program, we searched the global minimum (GM)\nstructures of Ag<sub><i>n</i></sub> (<i>n</i> =\n30–60) clusters with graphics processing unit acceleration,\nwhose efficiency is about 2 orders of magnitude higher than that of\nthe conventional density functional theory (DFT) calculations. GM\nstructures and some representative isomers are reported at each size,\nrevealing the competitive structural patterns based on truncated octahedra\nand icosahedra as well as the icosahedra-based layer-by-layer growth\nmode of large-sized Ag clusters. Most importantly, the size-dependent\nevolution behavior of structural and electronic properties of Ag<sub><i>n</i></sub> (<i>n</i> = 30–60) clusters\ncan successfully explain the observed stability at Ag<sub>48</sub>. Therefore, CGANet provides a powerful tool for rapidly exploring\nthe potential energy surface of atoms with an accuracy comparable\nto that of DFT.