Machine learning force field study of carboxylate ligands on the surface of zinc-blende CdSe quantum dots

Abstract

In colloidal quantum dots (QDs), the geometries of surface ligands may play significant roles in tuning the electronic structure, optical spectra and exciton dynamics. We here propose an effective approach to build a diverse dataset of small QDs, based on which the machine learning force field (MLFF) can be obtained based on the DeePMD framework and the energy of each atom is expressed based on the local atomic structure. Using the obtained QD force field (QDFF), molecular dynamics simulation of large zinc-blende CdSe QDs passivated by carboxylate ligands is successfully carried out, and the complex surface structure is extensively studied. We find that bridging, tilted, chelating and claw geometries are the major geometries of carboxylate ligands in CdSe QDs, and the alkyl chain length of ligands plays a significant role. The Markov state model is utilized to reveal the detailed geometry transformation channels. Due to the high performance of QDFF, the present approach is promising for systematic studies of large QDs with different kinds of ligands that can be synthesized in experiment.