All atoms drift during dimer calculation

[arosen]

Occasionally when using the dimer method, all the atoms in the unit cell tend to drift in a certain direction despite the MODECAR only have one non-zero (x,y,z) entry. I don't think this impacts the rate of TS convergence since all the atoms seem to be translated by the same amount, but it's a bit puzzling to me why this would occur. Any suggestions? Example files are attached. An animation showing the drifting can be found here (https://imgur.com/a/9SXsjTH).

EDIT: Turns out that my POSCAR/MODECAR were downright terrible. I'll keep this post up, but it's not a major issue since I don't get this behavior with more reasonable starting files.

[graeme]

You can see that the curvature in this calculation is almost zero (grep for "Dimer: CN" in the OUTCAR file, or look in the DIMCAR file). This corresponds to a translational mode, along which the dimer is pushing. The dimer method is really only working when the curvature is negative. If you displace a single atom, or have a POSCAR which is near to a saddle, the dimer will likely find a negative mode and there will be no significant translation. You can also free a single atom in the cell to avoid translation.

[arosen]

Thanks for the helpful reply. When I noticed the curvature was pretty much 0, it was clear something was definitely off, and it makes sense that the translating atoms would only occur in these cases. Fortunately, when you set things up the right way, things tend to behave as you'd expect...