Single molecule total energy variation

[Eduard]

Dear All,

I am computing the Ti(OH)4 molecule in a big box ( 15 A x 15 A x 15 A). For this calculation I am using the pseudopotentials: PAW_PBE O 08Apr2002, PAW_PBE Ti_sv 07Sep2000, PAW_PBE H 15Jun2001. The vasp version is vasp.5.2.2 15Apr09 complex. My INCAR is:

ISTART = 0
ICHARG = 2
ENCUT = 500

PREC = High
ISIF = 2
ISYM = 0
IDIPOL = 4

EDIFF = 5e-7
NELM = 300
NELMIN = 5
IALGO = 48
NBANDS = 36

EDIFFG = -0.025711
NSW = 200
POTIM = 0.0052918
IBRION = 2


ISMEAR = 0
SIGMA = 0.11


LCHARG = TRUE
LWAVE = TRUE
LREAL = AUTO

LPLANE = TRUE


I have used initially only 1 k-point ( Gamma), and the total energy I have obtained it as -54.687 eV. The energy and structure remain the same if the tag LREAL is set to FALSE
(instead of AUTO).

Just out of curiosity I have increased the number of k-points to 2x2x2. The total energy changes to -24.866 eV . From the structural point of view the only noticed modification is the lowering of the Ti-O-H angles ( from ~128 degree to 124 degree). If the LREAL tag is set to FALSE, the total energy lowers to -39.776 eV. The Ti-O-H angles changes to 131 degree.

For a k-points mesh of 3x3x3 no change in structure or energy is found by VASP when comparing with the 1x1x1 structure. Therefore, for the 3x3x3 mesh the total energy is identical with that for the 1x1x1 mesh ( -54.687 eV). The setting of LREAL = FALSE does not influence the result.

For the 4x4x4 mesh, the total energy is -47.222 eV.
For the 5x5x5 mesh, the total energy is -39.374 eV.

Similar results are obtained if I used the MP mesh instead of a Gamma centered mesh.

I would have expected my system to be properly described by a 1x1x1 k-point mesh ( is a
single molecule in a big cell). Nevertheless, with the increase of the k-point mesh I do not see any convergence. I notice however changes in both energy and structure. I would like to ask for your advise: is something wrong with my settings ? Which k-point mesh should I use for an accurate description of my system ?

With all my best regards,
Edi

[Eduard]

I would really appreciate any advice. If anyone could suggest how to interpret these results it would be really great.

[Eduard]

After careful tests, the setting IALGO=48 was found to be the one to lead to the strange total energy values. Changing to ALGO=Normal resulted in correct total energies for all higher k-point meshes.
The failure of IALGO=48 came, however, a bit as a surprise for me. Till now the RMM-DIIS scheme has always performed well for me (surfaces and bulk computations).

With best regards,
Eduard