very strange behavior in Tkatchenko and Scheffler VDW

[vbellini]

dears,

I found very strange behaviour when making calculation considering TS VDW.
Two calcuations on a magnetic systems, with fixed geometry, started with
two very slightly different kick in the MAGMOM.
Looking at the output file the two calculations seems to
converge electronically to the same state.
These are the last lines where the total energy (WITHOUT the VDW contribution)
is written for the two calculations.

RMM: 63 -0.433131760023E+02 -0.67114E-05 -0.81126E-08 4359 0.246E-03

and

RMM: 73 -0.433130665144E+02 -0.64008E-05 -0.71200E-07 5491 0.472E-03

but then when the VDW contribution is included the energies become

1 F= -.83169122E+02 E0= -.83168916E+02 d E =-.413175E-03 mag= 0.5306

and


1 F= -.72484617E+02 E0= -.72484436E+02 d E =-.361506E-03 mag= 0.5428

respectively.

This, to me, sounds like a very serious bug.
I am using the following VASP version

version vasp.5.3.3 18Dez12 (build Nov 28 2013 18:24:43

Did some of you notice something similar.
Is this a know bug which has been solved in later version of VASP ?

Than you for the help.

[admin]

No bug is reported so far.
Attach POSCAR and INCAR for the two calculations.

[vbellini]

I tried but I could not upload attachment.
Anyway.. here they are!

INCAR

-----
SYSTEM = test0
ISTART = 1 ;
ICHARG = 1
ENCUT = 500.000
EAUG = 703.443
ISMEAR = 0 ; SIGMA = 0.01
LORBIT=11
NPAR = 16
! KPAR = 4
ISYM = 2
NELM = 300
LREAL=.TRUE.
ADDGRID = .TRUE.
ALGO = FAST
spin:
ISPIN=2
MAGMOM = 4*0.7 2*0.0 -2.0
dynamic:
EDIFF = 0.1E-04
EDIFFG = -0.005
! IBRION = 2
! NSW = 40
! POTIM = 1.0
! LVDW = .TRUE.
IVDW=2
! LVDWSCS=.TRUE.
dipole & el. field:
IDIPOL = 3
LDIPOL = .TRUE.
EFIELD = 0.0
dft+U
LDAU = .TRUE.
LDAUTYPE = 1
LDAUL = 0 0 2
LDAUU = 0 0 3
LDAUJ = 0 0 1
LASPH = .TRUE.
LMAXMIX = 4

MAXMIX = 45

! LPARD = .TRUE.
! EINT = -0.5
! NBMOD = -3

! NEDOS = 3000
! NELMIN = 6

-----

and POSCAR

-----
cella grafene su Nichel
1.00000000000000
2.4680000000000000 0.0000000000000000 0.0000000000000000
-1.2340000000000000 2.1373506970000000 0.0000000000000000
0.0000000000000000 0.0000000000000000 32.0000000000000000
Ni C Co
4 2 1
Selective dynamics
Direct
0.0000000000000000 0.0000000000000000 0.0000000000000000 F F F
0.6666666123158933 0.3333332246317866 0.0629721875000016 F F F
0.3333332246317866 0.6666664492635661 0.1259446875000023 F F F
0.0000000000000000 0.0000000000000000 0.1881893327686850 F F T
0.6666666123158933 0.3333332246317866 0.2544175613907952 F F T
0.0000000000000000 0.0000000000000000 0.2542689799489222 F F T
0.0000000000000000 0.0000000000000000 0.3454219324481018 F F T

0.00000000E+00 0.00000000E+00 0.00000000E+00
0.00000000E+00 0.00000000E+00 0.00000000E+00
0.00000000E+00 0.00000000E+00 0.00000000E+00
0.00000000E+00 0.00000000E+00 0.00000000E+00
0.00000000E+00 0.00000000E+00 0.00000000E+00
0.00000000E+00 0.00000000E+00 0.00000000E+00
0.00000000E+00 0.00000000E+00 0.00000000E+00

-----

I am using the latest PAW PBE (otherwise TS is not working, right?)

PAW Ni_GW 31Mar2010
TITEL = PAW Ni_GW 31Mar2010
LPAW = T paw PP
PAW radial sets
PAW C_GW_new 19Mar2012
TITEL = PAW C_GW_new 19Mar2012
LPAW = T paw PP
PAW radial sets
PAW Co_GW 31Mar2010
TITEL = PAW Co_GW 31Mar2010
LPAW = T paw PP
PAW radial sets

Thank you for the help.
Valerio

[vbellini]

Forgot to mention..
one is obtained with this kick in the MAGMOM

MAGMOM = 4*0.7 2*0.0 -2.0

the other with the following

MAGMOM = 4*0.8 2*0.0 -2.5

Besides.. Things seem to be ever more strange.

Later I started two times the same calculation (removing CHGCAR and WAVECAR,
so both starts from scratch) for the same MAGMOM kick, i.e. MAGMOM = 4*0.8 2*0.0 -2.0.
So I mean, I did twice the SAME calculation!!
The results are:

RMM: 81 -0.432691695661E+02 -0.68156E-05 -0.36573E-06 6388 0.102E-02
1 F= -.91760146E+02 E0= -.91759984E+02 d E =-.324772E-03 mag= 0.5811

and

RMM: 80 -0.432691967476E+02 0.65658E-05 -0.17688E-06 6174 0.717E-03
1 F= -.86162603E+02 E0= -.86162440E+02 d E =-.324583E-03 mag= 0.5813

looks to me it is 'fishing' in the memory.. maybe wrong dimensioning?
note that if I use Grimme instead of TS, I do not have such issues.
V.

[admin]

1/ Values defined by MAGMOM are just initial magnetic moments entering the scf procedure.
When only one stable electronic configuration exists, scf procedure finds it, no matter which initial
magnetic moments are defined.
2/ The author of VdW says: "The VdW calculation is not compatible with the parameter ADDGRID."
Remove ADDGRID and you will get correct VdW energies.
Originally this was mentioned in the description of the method. On the way from the author
to the manual, however, the warning was lost. It will be inserted it in the manual.