The physical and chemical properties of clusters undergo significant changes with changes in size. Therefore, the evolution of cluster structure with size has always been an important issue in nanoscience. However, due to the complexity of cluster potential energy surfaces and limitations in computational simulation methods, the evolution of cluster structures has long been a challenge. Here, we developed a high-accuracy machine learning interatomic potential (MLIP) for palladium systems, including clusters, surfaces, and crystal. This MLIP enables a cross-scale (${\mathrm{Pd}}_n, n=12-21856$) study of palladium systems from clusters to nanocrystal. For ${\mathrm{Pd}}_n$ ($n=12-60$) clusters, simulated annealing revealed 32 new lower-energy structures. Small ${\mathrm{Pd}}_n$ clusters (n < 85) show core-shell structures and repeated transitions from truncated to complete octahedra near FCC magic sizes (e.g., ${\mathrm{Pd}}_{38}$ to ${\mathrm{Pd}}_{44}, {\mathrm{Pd}}_{79}$ to ${\mathrm{Pd}}_{84}$). When the size exceeds 200, the Wulff construction becomes increasingly stable. There is competition among multiple structural modes (icosahedron, octahedron, and Wulff construction) within the range of 200–460. Beyond 460 atoms, Wulff constructions dominate and the system transfers from cluster to nanocrystal. Our work reveals continuous transition from Pd clusters to nanocrystal and the critical size, and paves the way for studies on other clusters.

Close

PrevNext

View figure in article

• Clusters
• Nanoclusters
• Machine learning