Research on Cu-Sn machine learning interatomic potential with active learning strategy
Description
The uploaded files include the initial training set of the CuSn potential function (in the init_data folder) as well as the data added to the training set during the training process (in the training_data folder). The file frozen_model.pb is the potential function obtained from training, used for molecular dynamics simulations with the DeepMD version.
Files
Steps to reproduce
First, an initial iteration is conducted based on the energy and force information from the initial dataset, using the scheduling software DPGEN, which includes three steps (training, exploration, labeling). First, a potential function is iteratively generated using DeepMD-kit based on the initial dataset. Once the iteration is complete, DPGEN automatically initiates the second step, which involves phase space exploration. During the exploration, some data points are labeled, specifically those where the maximum force deviation falls within the predefined upper and lower limits of maximum force deviation. These labeled data points will be used for the next step of single-point energy calculations. After completing the three steps, the initial iteration is finalized.