cell "explodes" during relaxation for adsorption

[awang95]

Hi,

I'm trying to study H2O adsorption onto oxygen vacancy sites in CeO2(111). I've successfully relaxed the structures of CeO2(111) with 2 vacancies on the surface, as well as the same structure with vacancies but with an H2O molecule far from the surface.

However, when I try to bring the H2O molecule close to the surface, eventually the cell starts to expand and kind of "explodes" - I think my starting structure is reasonable since it was the previously relaxed structure, but just with an extra H2O near the surface. The previously relaxed structure was obtained by removing oxygens from a relaxed CeO2(111) surface without vacancies (which was relaxed from a surface constructed from a cif file) and then relaxing it. I've tried different combinations of oxygen vacancies, tried restarting the calculations, and also turning on/off the vdw correction in the INCAR, but all lead to a similar result. Can I please get some suggestions on why this might happen, and how to best address it? I've searched the forum but couldn't find a problem similar to mine.

My files are attached (CONTCAR of the successful relaxation - CeO2 with surface vacancies, with H2O far away, CONTCAR and XDATCAR of the "exploded cell" - CeO2 with surface vacancies but with H2O close to the surface, INCAR for both calculations, and the starting POSCAR of both calculations; for both calculations I'm using the GW version of the pseudopotentials and am sampling only the gamma point)

thanks!

[merzuk.kaltak]

Dear awang95,

could you please provide the stdout as well?
I want to know when exactly the cell "explodes".
It may help to start the relaxation from a preconverged WAVECAR file?

It might also help to first perform a few steps with another DFT functional, for instance PBE+U, write the WAVECAR file and continue the relaxation from this WAVECAR with your selected DFT functional.

[awang95]

sorry for the late reply; here's a folder containing the stdout files

thanks for the suggestion; will try it