Why calculated energies of grain boundary segregation are unreliable when segregant solubility is low

Density functional theory based calculations of segregation energies at interfaces are often unreliable when the bulk solid solubility of the segregant is lower than that corresponding to one solute atom per computational repeat cell. In this case, the calculated energy of a solute in the bulk cannot be used when evaluating segregation energies. We document this problem by analyzing the measured and calculated grain boundary segregation energies in α-iron available in the literature. Presumably, this idea can be extended to other systems.