PotentialMind: Graph Convolutional Machine Learning Potential for Sb–Te Binary Compounds of Multiple Stoichiometries

Machine learning potential (MLP) has emerged as a powerful tool in materials research and design. However, most MLP methods rely only on a single descriptor generated by mathematical functions instead of mapping the three-dimensional space of the materials structure, and thus this type of potential is typically limited to specific compositions. In this research, we present graph convolutional machine learning potential (GCMLP) software, termed PotentialMind, which can transform three-dimensional atomic structures into vectors comprising nodes, edges, and weights based on multiple descriptors. Using Sb–Te phase change materials as examples, a model named GCMLP-ST suitable for 12 stoichiometries of Sb–Te compounds has been constructed, whose root-mean-square errors for energy and forces are, respectively, 4.51 and 73.13 meV/Å for training data sets and are, respectively, 4.97 and 76.25 meV/Å for unfamiliar testing data sets. Moreover, for the energy-volume curves and radius distribution function by molecular dynamics, the GCMLP-ST model with 10,000 atoms exhibits good agreement with the ab initio molecular dynamics (AIMD) results across crystalline, liquid, and amorphous phases for the six representative Sb–Te material systems, which also exhibit 50 times the computational efficiency of AIMD. With this framework, the architecture of the machine learning model can be customized by deep and transfer learning, extending to other material systems. In addition, benefiting from the high efficiency of PotentialMind molecular dynamics (PMMD), it can be used for real devices, spanning tens of nanoseconds and comprising millions of atoms under different programming conditions that are impossible with AIMD simulations.