I am looking at the total energies for two systems. One has the ideal cubic perovskite geometry and the other has the atoms displaced from their ideal positions. When I run VASP I get the energy difference between the two structures to be 4 eV which I know is ridiculous. I expect the difference to be around 0.5 eV. There are no other warnings in the run. I guess I am converging to a local minima. How can I get around the problem? I tried changing the number of KPOINTS, but that did not help. I attach the INCAR file used.

ISMEAR = 0

SIGMA=0.001

PREC=ACC

ISTART=0

EDIFF=0.1E-5

NBANDS=120

NSW=20

IBRION=2