The difference between PBE potential after VASP5

[ian800726]

Hi everyone, I'm a beginner user of vasp. Recently, I would like to do some calculation about gold surface. At first, I use PBE potential before VASP5, which doesn't contain the meta GGA functions. However, while I am trying to use the new PBE potential, which including metaGGA functions, I encountered the problems that there are somewhat large difference between new and old PBE potential while I using PBE potential as a base and calculating the Bulk total energy and lattice constant with vdW functionals.
My INCAR is

System = PBE.lattice
ISTART = 0
ICHARG = 2
NELM = 60
NELMIN = 5
EDIFF = 1E-6
PREC = Accurate
ENCUT = 400.00 eV
LREAL = Auto
NSIM = 4
AMIN = 0.005
ISIF = 3
IBRION = 2
NSW = 200
EDIFFG = -1.00e-10
EDIFF = 1.00e-10
ISMEAR = -5
SIGMA = 0.02
NPAR = 1
LPLANE = .TRUE.

And I using kpoints 21*21*21 with Monkhorst mesh grid method

the consequences are showing below

lattice constant(angstrom)
using new PBE 4.15598
using old PBE 4.17260

total Energy(open optB86b functional)(eV)
using new PBE -0.78689272
using old PBE -1.6147351

Does anyone have idea about the problem? Thanks for your advice~

[kelum]

Hi,
for ISIF=3 you shouldn't use ISMEAR=-5 but ISMEAR=1 or 2.
When using ISIF=3 you need to repeat the calculation few times to get
a converged lattice and delete WAVECAR between the calculations.
For ISIF=3 you should use higher ENCUT than normal since otherwise you
get underestimated lattice parameters.
EDIFFG is way too strict, -1e-3 should be OK.
(More details in the VASP manual.)

The "Total energy" written out by VASP has little meaning for the vdW-DF
functionals and it depends on the POTCAR used. To get cohesive/lattice energy
you need to subtract the energy of an atom. You should get similar values
afterwards.

The lattice constants with different potentials differ usually a bit, potentials with more
valence electrons should be more accurate. The differences between your data are on
the edge but they are a bit higher than the numbers I got.
For the new PBE potentials I got with optB86b-vdW
Au 4.121
Au_GW 4.115
Au_pv_GW 4.123
Au_sv_GW 4.120
I did 7 calculations around the minimum and fitted to
the Birch-Murnaghan EOS, that's faster than ISIF=3 for systems
with high symmetry.
So I get about 0.2% difference between the results with different potentials,
that's actually less than I expected. (I used 15^3 kpoints, ENCUT=800, prec=high)
(The GW potentials are not necessary for vdW-DF, I used them since they have more
electrons in the valence.)

So these values are a bit lower than yours, but I'd try changing your INCAR and maybe
then they'll get closer.
jik