Does the difference in results come from molecular dynamics programm? I use LAMMPS and author used MOLDY.

By results do you mean the binding energies for He clusters along the GBs or just  the GB energies of pure Fe themselves?  

For now I try to calculate grain boundary energies of pure Fe.

It maybe worth looking into Tschopp et. al old publications on Cu and Al GBs (2007), detailing the methodology. In addition you can compare from this link the input scripts. In answer to your follow up question there shouldn't be  a large difference in using MOLDY vs LAMMPS for the generation of GBs. If the aim of your project is simply looking at the GB energies, I would work with the LAMMPS platform that way you can compare against a vast majority of atomistic work.

Hope that helps.

https://icme.hpc.msstate.edu/mediawiki/index.php/LAMMPS_Input_Deck_for_Grain_boundary_generation

Are your energies always higher than those published? It is critical when constructing GBs to allow for full relaxation, e.g. try constructing the same GB but using an offset in the lattice command for LAMMPS for one of the bicrystal layers.

Thank you both very much.

Now I understand a lot more thanks to your suggestions and my my energies are consistent with those in literature.