Neural network atomic potential to investigate the dislocation dynamics in bcc iron

Abstract

To design the mechanical strength of body-centered-cubic (bcc) iron, clarifying the dislocation dynamics is very important. Using systematically constructed reference data based on density functional theory (DFT) calculations, we construct an atomic artificial neural network (ANN) potential to investigate the dislocation dynamics in bcc iron with the accuracy of DFT calculations. The bulk properties and defect formation energies predicted by the constructed ANN potential are in good agreement with the reference DFT calculations. The ${a}_{0}/2\ensuremath{\langle}111\ensuremath{\rangle}{110}$ screw dislocation core structure predicted by the ANN potential is compact and nondegenerate. The Peierls barrier predicted by the ANN potential is 35.3 meV per length of the Burgers vector. These results are consistent with the DFT results. Furthermore, not only the Peierls barrier, but also the two-dimensional energy profile of the screw dislocation core position predicted by the ANN potential are in excellent agreement with the DFT results. These results clearly demonstrate the reproducibility and transferability of the constructed ANN potential for investigating dislocation dynamics with the accuracy of the DFT. Combined with advanced atomistic techniques, the ANN potential will be highly useful for investigating the dislocation dynamics in bcc iron at finite temperatures.