Your problem is quite common when preparing initial replicas for NEB calculations. In fact, they are usually generated by making a linear interpolation between initial and final position for each atom. If, for some reasons, one atom is "shifted" to the other side of the cell due to periodic boundary conditions, the linear interpolation yields spurious results.

The only solutions I know are to replace by hand the "shifted" atoms BEFORE the interpolation, or use a simple script that will check the distance between the initial and final position for each atom, and if the distance is larger than say, half the supercell, apply PBC to shift it back.

Hope this helps

Best regards

L.

This problem occurs quite commonly when preparing the initial and final images for NEB calculations. One solution is to move the atoms by hand before you start doing the calculations as suggested above. Other option that worked for me is to just shift simulation box boundaries little bit and not the atoms such that no atom is at the boundary. For example, in your first figure (initial.jpg) you can shift just the simulation cell boundary to the right. As a result, the white atoms on the left boundary of the simulation cell will need to be shifted to the right side to maintain the periodicity of the cell but now they will be inside the simulation cell instead of just at the boundary.

Hope this helps.

Why don't you fix the boundary atoms? If your substrate's structure is optimized then the structure can be supposed to be fixed during the adsorption.