Hi,

Jefferson Maul has run a few tests and indeed we were able to reproduce the odd behavior you've been experiencing. Those negative values for the electron population of the vacancy are indeed very strange and may be suggestive of a basis set unbalance. It seems that whenever basis functions are left on the vacant site (either with your initial ATOMSUBS approach or with the more "canonical" GHOSTS approach) those functions are much "needed" by surrounding atoms.

We have found a possible way forward for your system through the use of the ATOMREMO option. With this option, you create the vacancy by removing the selected atom alongside its basis functions. For instance, we have used this approach to study Oxygen vacancies in CaSnO3 here.

We have used this approach on your system (on a smaller supercell to be able to run the tests more efficiently) and the geometry optimization went well. See the attached output file as a guide.

Hope this helps,