Convergence with respect to cut-off energy in TiO2 and MgO

[lle]

Hi!

I have a weird problem when trying to search for a proper kinetic energy cut-off ENCUT in TiO2 rutile and MgO bulk materials. When i raise ENCUT, i get higher total energies (smaller absolute values), when they usually get lower as the cut-off energy grows. The results for TiO2, for example, are as follows:

E_cut E_tot
280 -53.713087
300 -53.738972
320 -53.555159
340 (program crashed)
360 -53.293021
380 -53.255149
400 -53.200219
420 -53.172368
440 -53.151048
460 -53.144156
480 -53.130391
500 -53.127086
520 -53.126038
540 -53.126846
560 -53.113928

ENMAX values in the POTCAR files are 178.367 eV for Ti and 400 eV for O, and the above results are obtained with PAW. The MgO behaviour is qualitatively the same. Results with US PP's vary even more than one eV for TiO2. I have also tried with many different INCAR parameters, but they don't seem to have much effect. In bulk Ti, the convergence is normal. Does anyone know if this is just a property of these metal oxides or is there something wrong with the calculation? An example of the INCAR parameters I've used is given below:

SYSTEM = Tio2 Bulk
LREAL = .FALSE.
ISMEAR = -5
SIGMA = 0.5
ENCUT = $Ecut eV (determined in the job script)
NBANDS = 19
PREC = Normal
LPLANE = .TRUE.
NSIM = 10
LWAVE=.FALSE.
IBRION = 1
EDIFFG = -0.02
NSW = 0
NELM = 100
ALGO = VeryFast
LCHARG=.FALSE.

[admin]

Please also increase PREC if you increase the cutoff, to make sure that all cutoffs are re-scaled properly ewith ENCUT.
the behaviour may also be (partly) due to the automatic convergence corrections (please have a look at p4 of the 'accuracy.pdf' VASP-workshop tutorial file).