problems with convergence (PAW_PBE Al_h potential)

[chrbrandl]

I tried to determine the convergency behavior of a primitive Al fcc cell. I have chosen the PAW_PBE Al_h potential and did some test within I varied the parameter in the INCAR file:
SYSTEM = Al fcc unit cell
ISTART = 0
ENCUT = 350.0
IALGO = 38
PREC = high
LREAL = A
ISMEAR = 1
SIGMA = 0.2
NBANDS = 15
------------------------
By increasing the ENCUT the total energy showed suddenly a jump instead of an asymtotic behaviour:

SIM_NO KMESH ENCUT IALGO NBANDS F E0
1 6 300 48 10 -.14987492E+02 -.14987228E+02
2 8 300 48 10 -.15034429E+02 -.15033879E+02
3 10 300 48 10 -.15028306E+02 -.15028087E+02
4 12 300 48 10 -.15011341E+02 -.15010816E+02
5 14 300 48 10 -.15008594E+02 -.15008136E+02
6 16 300 48 10 -.15015669E+02 -.15015594E+02
7 12 350 48 10 -.15028270E+02 -.15027743E+02
8a 12 400 38 10 -.15184279E+02 -.15188389E+02 -> warning of occupied highest band
8b 12 400 48 15 -.15056896E+02 -.15056361E+02
9a 12 450 38 25 -.15941366E+02 -.15976946E+02
9b 12 450 38 40 -.15941362E+02 -.15976935E+02


So to my question, is this result related to a wrong procedure or to the potential.

Thanks

[Veronika]

try again with increased NBANDS and probably more k-points (whats your grid?)
<span class='smallblacktext'>[ Edited Tue May 16 2006, 04:13PM ]</span>

[admin]

the way you summarize your results is a little confusing, I must confess.
1) by 'primitive fcc cell' do you mean 1 atom in an fcc Bravais lattice, am I right? ten bands should be enough for 3 valence electrons.
2) were your results equally well converged in energy or did they stop at the NELM-limit (none of the convergence parameters is set, therefore I ask)
3) by kmesh=N, do you mean NxNxN k-points or was N the number of k-points in the IBZ (if this was the case, the mesh is certainly too small)
4) apart from runs 1-6, there are not enough consistet runs to check any convergence.
1-6 may be influenced by the quality of the k-mesh (see 3)
for all other data, there are only 2 sets which are comparable in the sense that only one parameter of the set (k-points, ENCUT, NBANDS, ALGO) is changed consistenly , which certainly will not give any 'trends'.
please
1) FIRST choose ALGO, and keep it the same for the whole series
2) choose an appropriate number of bands (NBANDS)
3) converge the number of k-points
4) THEN increase ENCUT.

the warning must of 8a must have some different reason, at least if you really have a primitive fcc cell wit 3 valence electrons all in all