I find this strange, since according to the source code (VASP 5.3.5), if you run with SOC (i.e. the non-collinear compiled binary) the code is supposed to just stop with the error message:

> vdW-DF calculation: either NGZhalf or NGXhalf needs to be used for compilation

Further analysis of the source code seems to indicate that this really shouldn't work, unless some pretty major rewriting of the source code is done. A few possibilities come to mind:

* Someone has tinkered with the source code, removing this stop (really bad idea).

* Someone tinkered so that the code quietly turns off the vdW-DF when run with spin-orbit (also a really bad idea).

* No-one has been tinkering, but you have run into a VASP bug that causes one of the above.

* No-one has been tinkering, but something went wrong in the compilation to cause one of the above problems.

* Someone re-wrote the source code to actually work also in the non-collinear case (that would be great of course, but perhaps not the most likely case here).

* ...plus all the possibilities that I'm missing...

Is the vdW-DF contribution to the energy printed to the OUTCAR file? If so, how does the non-local energy compare between your +vdW and +vdw+SOC cases? (By the way, the proper name here should probably be "optB88+vdW+SOC" etc. since you are not using the PBE functional).

In case you have somehow accidentally run without the vdW-DF non-local correlation contribution, the expanded lattice constant is to be expected. The reason for this is that the optB88 exchange is fitted to work well with the extra attractive interaction supplied by vdW-DF, so on its own it should be too repulsive (for more of my views on vdW-DF's, see J. Chem. Phys. 141, 074708 (2014)).

Dear Torbjörn,

Thank you very much for your answer!

I just found that my non-collinear version of VASP can run the optB88+vdW+SOC calculations smoothly with or without the vdW kernel! In either case, the LUSE_VDW was both switched on, but the kernel was not loaded. So I was misled (and more my carelessness, because the vdW correction energy, 'Total vdW correction in eV', was not printed the standard output in either case) by the successful ending of the calculations, without knowing the fact that vdW was in fact not included!

Then, I think my calculations are actually optB88+SOC, without the non-local correlation from vdW-DF. Therefore the optB88 exchange functional alone causes the  over-expansion of the lattice constants.

Now, it seems that I need to turn to other vdW correction schemes if I want to do vdW+SOC calculations. Can I still use other vdW-DFs, can they be combined with SOC? Or I have to use the semi-empirical DFT-D2 or DFT-D3 methods? 

And in the end, do you know why optB88+vdW-DF is not working with SOC?

Thank you in advance for your help!

Kind regards,

Wun-Fan

Hi Wei,

I think the Z-direction should also be relaxed so that we can get the right bond length. Isn't it?

Hi,

I am using vasp 5.4.4 and i am doing band structure calculations.

I am getting this error.

>vdW-DF calculation: either NGZhalf or NGXhalf needs to be used for compilation

Here is my incar:

# Parallelization

#NPAR = 4

# ElectronicMinimization

ALGO = Normal

PREC = Accurate

ENCUT = 500 eV

EDIFF = 1E-5

ISTART = 0

ICHARG = 1

LREAL = Auto

ISYM = 0

ISMEAR = 0

SIGMA = 0.1

NEDOS = 3001

EMIN = -20

EMAX = 20

LORBIT = 10

# OutputOptions

LWAVE = F

LCHARG = T

LVHAR = F

# VanDerWaals

#IVDW = 12

GGA = BO

PARAM1 = 0.1833333333

PARAM2 = 0.22

AGGAC = 0.0

LUSE_VDW = .TRUE.

LASPH = .TRUE.

NGX=140

NGY=140

NGZ=140

Better to use INCAR with minimum Tags, or use DFT-D3 method