About the convergence problems in the course of energy correction for Ni, etc.

[guest]

I did the energy correction for Ni and Nb, etc, following the setting shown in 10.5 in VASP Manual.

To reduce the error (T*S) to less than 1 meV per atom, the value of SIGMA have to reduce to about 0.01 or even 0.005. Then the convergence was stuck. I had tried to modify the mixing setting follow the rules shown in manual, but it seemed nothing better.

What is going wrong? Did I make any mistake?

Some correlative setting is followed:
INCAR:
ISYM = 0 ! no symmetry
ISPIN = 2 ! allow for spin polarisation
VOSKOWN = 1 ! this is important, in particular for GGA
ISMEAR = 0 ! Gaussian smearing, otherwise negative occupancies
SIGMA = 0.005 ! intermid. smearing width
AMIX = 0.2 ! mixing set manually
BMIX = 0.0001
# NELM = 20 ! 20 electronic steps
ICHARG = 1
-------------------
POSCAR:
Ni-ec
20.0
1.0 0.0 0.0
0.0 1.0 0.0
0.0 0.0 1.0
1
direct
0.0 0.0 0.0
---------------------
KPOINTS:
Monkhorst-Pack method 1x1x1
0
Monkhorst
1 1 1
0 0 0
---------------------
POTCAR: PAW_GGA Ni
---------------------

Thanks for your help.

[admin]

For free atoms, there is no T-smearing of the levels, the atomic energy levels are discrete.
Hence the T*S term has no physical meaning