Question for electron localization problem using DFT+U

[Eastar]

Dear vasp user:
Now i'm trying to investigate the electronic structure of the defective CeO2 using DFT+U, the Hubbard-U takes the formalism formulated by Dudarev, i.e. LDAUTYPE=2, and the value of U-J is set as 5 eV. As the literature indicated (PR B,71,041102). the habbard-U correction results in a broken-symmetry solution where the two electrons localize on two (rather than four) Ce atoms neighboring the vacancy, however, when i take an oxygen away form CeO2 bulk, i can't get the correct electron distribution, i.e. the two electrons are even distributed to the neighboring four Ce cations.
My question is how to obtain the right electron structure for defective CeO2 even using DFT+U? any special skills for that? Any suggestion is appreciated.

[aron]

If the four oxygens are equivalent to begin with, you will need to move them slightly off their ideal sites or turn off symmetry to get the broken solution.