Strategic sampling with stochastic surface walking for machine learning force fields in iron's bcc-hcp phase transitions

This study developed a machine learning-based force field for simulating the bcc-hcp phase transitions of iron. By employing traditional molecular dynamics sampling methods and stochastic surface walking sampling methods, combined with Bayesian inference, we construct an efficient machine learning potential for iron. By using SOAP descriptors to map structural data, we find that the machine learning force field exhibits good coverage in the phase transition space. Accuracy evaluation shows that the machine learning force field has small errors compared to DFT calculations in terms of energy, force, and stress evaluations, indicating excellent reproducibility. Additionally, the machine learning force field accurately predicts the stable crystal structure parameters, elastic constants, and bulk modulus of bcc and hcp phases of iron, and demonstrates good performance in predicting higher-order derivatives and phase transition processes, as evidenced by comparisons with DFT calculations and existing experimental data. Therefore, our study provides an effective tool for investigating the phase transitions of iron using machine learning methods, offering new insights and approaches for materials science and solid-state physics research. A machine learning force field for predicting the bcc-hcp phase transitions of iron, demonstrates good performance with DFT calculations, offering new insights and approaches for materials science and solid-state physics research.