ion jumps possibly spoil a dimer run?

[perez]

hello

during my dimer-method runs for atomic displacement/defect creation in bulk germanium, I have observed that it is not uncommon to encounter, depending on the case at hand (the specific direction selected for atomic displacement, the size of the unit cell), with the following situation: your initial state geometry SCF iteration converges fine, and once the first ionic move has been taken according to the dimer algorithm, before the SCF cycle starts, you get the famous warning "ions are too close" and the ensuing SCF energy is much higher. For a few subsequent ion moves/dimer iterations, the "ions are too close" warning continues appearing, with correspondingly high SCF energies. Eventually the dimer moves get you into a geometry for which the "ions are too close" warning disappears and the SCF energies are closer to that of the initial state, eventually converging into a saddle point (forces are zero)

will you please let me know the possible effect of such "atomic jumps" (as a consequence of PAW potentials with too large radii for the atoms that get close to each other, I guess) on the dimer run? Has the atomic displacement directional search lost its direction as a result of the jump? Can I trust the resulting saddle point as being the closest static transition state in that specific direction? My questions are related to my getting exactly the same final state energy when running the dimer search from the same initial state with direction search along different directions (in one case there have been "jumps", in the other not)

I could send all data for a specific case, although this happens often during dimer searches, and I have seen it in the germanium case and also in CuCrO2

regards

Manuel

PS the most recent case is a 16 atom bulk germanium cell (2x2x2 wrt the primitive cell), displacing an atom along direction [111], final state energy -68.380068 eV;
when I run displacement along the [100] direction there is no jump/"ions too close" warning, and the final-state energy is -68.380085 eV

[perez]

hi,

just rephrasing my enquiry: will you please let me know whether I must discard/throw away at once any dimer calculation in which the "ions are too close" warning (and subsequent jump of the atom(s) getting out of such a trap) appears?

thanks

Manuel

[perez]

hi again

just to make sure the warning "ions are too close" is not a mistake on my side, the direction vector that you enter in your input file (MODECAR) is that in cartesian coordinates or else, in lattice vector fractional coordinates?
certainly the direction vector is updated at every dimer step, that's why I have not discarded as yet the dimer runs that contain the warning...

thanks

Manuel

PS my guess is that for the diamond lattice bulk germanium case, displacements along 111 in either coordinate system are equivalent (cartesian and lattice-vector fractional), whether your use a supercell based on the primitive or on the conventional cubic cell. The 100 direction would be different, I think (if we talk about lattice-vector fractional coordinates)