Performance of different traditional and machine learning-based atomistic potential functions in the simulation of mechanical behavior of Fe nanowires

Characterization of nanoscale material properties using experimental methods is difficult due to their high costs and complexities. Hence, many researchers apply alternative techniques, such as atomic/molecular simulations, to overcome the limitations of experimental methods. A well-established method to study material behavior at nanoscales is molecular dynamics simulation. However, the accuracy of its results is highly dependent on the selection of appropriate potential functions/force fields modeling atomic interactions. This work investigates the effects of considering different potential functions on the mechanical characteristics of BCC Fe nanostructures using molecular dynamics simulations. Laboratory data are used to verify the calculated elastic properties of bulk structures. Furthermore, systematic simulations are performed to determine the elastic properties of nanowires under tensile loading, various potential functions, and different temperatures. According to the results, Ackland's potential function and experimental data are in excellent agreement. On the contrary, Asadi's potential function shows an opposite trend comparing other functions and experimental mechanical data. In addition, the potential function formulated based on artificial neural networks is capable to capture tension-induced necking phenomena in Fe nanowires.