The Reconstruction of Pt(001) Surface and the Shell-Like Reconstruction of the Vicinal Pt(001) Surfaces Revealed by Neural Network Potential

Abstract

In this work, a highly accurate neural network potential (NNP) is presented, named PtNNP, and the exploration of the reconstruction of the Pt(001) surface and its vicinal surfaces with it. Contrary to the most accepted understanding of the Pt(001) surface reconstruction, the study reveals that the main driving force behind Pt(001) quasi-hexagonal reconstruction is not the surface stress relaxation but the increased coordination number of the surface atoms resulting in stronger intralayer binding in the reconstructed surface layer. In agreement with experimental observations, the optimized supercell size of the reconstructed Pt(001) surface contains (5 x 20) unit cells. Surprisingly, the reconstruction of the vicinal Pt(001) surfaces leads to a smooth shell-like surface layer covering the whole surface and diminishing sharp step edges. This study provides novel insights into the Pt(001) quasi-hexagonal surface reconstruction. It is found that this reconstruction is driven by surface energy reduction rather than surface stress relaxation. Furthermore, the vicinal Pt(001) surfaces are reconstructed into a shell-like structure in sharp contrast to the traditional terrace-step-kink models. This may provide new insights into the catalytic activity on these surfaces. image