Strengths and Pitfalls of classical interatomic potentials for the modelling of hydrogen embrittlement in BCC-Fe: A benchmarking analysis

Abstract

The rational design of cost-effective, hydrogen-resistant structural materials is essential for establishing hydrogen as a competitive alternative to other emission-free storage technologies. To this end, atomistic models based on empirical interatomic potentials (IPs) provide valuable insights on the interplay between H diffusion and micromechanics at a fraction of the cost of electronic calculations. For the BCC-Fe – H system, several such IPs have been proposed and deployed under a wide variety of conditions. However, IP validation has largely been conducted in the infinite dilution limit and on the basis of thermodynamic metrics, leaving doubts on their accuracy under realistic hydrogen loads in dynamic settings. To address this shortcoming, we provide a comprehensive assessment of seven widely used IPs for the BCC-Fe–H system, encompassing the popular embedded atom (EAM), Modified EAM (MEAM) and bond-order (BOP) potential models. Our analysis incorporates critical metrics, including mechanical behavior under volumetric and uniaxial deformation, hydrogen distribution and kinetics, and grain boundary segregation at both moderate and high hydrogen concentrations. Our findings reveal significant discrepancies in predictive accuracy, along with system-size and simulation-length artifacts that are easily overlooked in the application of these IPs. Additionally, we identify an inherent failure of EAM-type IPs (the most frequently used IP type) to both prevent unrealistic H clustering and accurately estimate its transport properties. Lastly, we present a detailed ranking of the evaluated IPs and assess the overall-best performing model on a large polycrystal system, enabling researchers to make informed choices based on the specific requirements of their studies.