I am trying to simulate a ABO4 material with vacancies while the models always experienced collapse of atomic layers with the expansion of volume. The results of running a supercell without defects look good and so I am sure that the FIELD is a suitable one. The CONTROL file is posted as follows:
I think the tauP should be greater e.g. 500 or 1000. Perovskites with vacancies can result in a very fragile structure to equilibrate. If your box is free to expand a lot you can easily loose the the structure. Instead with a slow deformation the structure is better preserved.
Try to run your structure without shell model and keep the resulting box as a starting point.
then modify your config file, to incorporate newly the shell model
(the code atomsk does the job in a very efficient way).
Then run your simulation with the shell model. If it fails, it certainly comes from a problem with the shell. If it goes too far/close, you may get unstable system. Some function exists to avoid shell going far away or too close to another or a small weight on the shell avois too big unrealistic movements.