average electrostatic potential at the core does not converge

[fzj]

Dear vasp users,
I'm doing some calculations on system with impurity. For example: doping GaN with Mg. To simulate the defect system, I employed the zinc-blende GaN, and enlarged it by 2x2x2, with an impurity atom Mg in the center of the supercell. After total energy calculation, I examined the average potential at the core in the OUTCAR file, and found that the average potential at the core for atom Ga or N does not converge as the host atom (Ga or N ) is apart from the impurity Mg. In my opinion, the potential ought to be converged. What is the possible reason for my odd results?
This is not only the case for GaN:Mg. I have tested ZnO:N, and got nearly the same odd results. The ZnO:N is also simulated by a 2x2x2 64-atom supercell.
Thank you very much.

[admin]

in principle you are correct, in the limit of infinite distance from an impurity, the average potential at the core (of an electronically fully converged calculation) has to converge (it is in fact constant for chemically equivalent atoms, eg at different layers of a surface). The fact that it is not in your case, unfortunately implies that your supercell still is too small. You might check for other parameters, like the decay of the forces on the atoms with increasing distance (or, alternatively, their displacements from the respective equilibrium positions)