UT theoretical chemistry code forum

Hello.

I have a question regarding the addgrid command in the vasp incar file.

It seems that in some cases the use of addgrid affects the resulting charge values. For example, I have run calculations on MgO, with and without the use of addgrid, while varying the number of grid points. The resulting ionic charges are as follows:

NGX/Y/Z.......addgrid=f.........addgrid=t
-------------------------------------------------------
38.................1.148..............1.389
76.................1.161..............1.386
152................1.090..............1.339

The vasp manual explains addgrid as an additional support grid with eight times as many points as the grid defined by NGX/Y/Z. So I was expecting the values obtained without the use of this support grid to approach those obtained with it, when increasing NGX/Y/Z. This is obviously not the case in the example above. In contrast, the effect of addgrid on the resulting charge values of a couple of other compounds I have been investigating was negligible.

So my question is how the findings above can be explained and whether you would recommend using addgrid or not. I have attached one of the incar files below; please let me know if any other information might be of interest. I appreciate any help or comments.


PAW_GGA, Monkhorst 14x14x14

ISMEAR = -5
IBRION = -1
POTIM = 0.1
EDIFF = 1E-05
EDIFFG = 1E-04
NSW = 0
PREC = high
NWRITE = 2
ISYM = 1
LREAL = F
NGX = 152
NGY = 152
NGZ = 152
ISTART = 0
IALGO = 48
LORBIT = 2
ADDGRID = T
LELF = T
LAECHG = T
ENCUT = 700

Convergence of the Bader charges should be verified with respect to all of the calculation parameters. I am pretty surprised by the sensitivity to the addgrid tag, since the extra-fine grid is only used for the augmentation charges near the core. Please make sure that you have included core charges as described here: http://theory.cm.utexas.edu/vasp/bader/vasp.php

Yes, I had followed the instructions you have linked to. I am not sure if this a meaningful indicator here, but in each case the total number of (valence) electrons equals 32, as it should.

With regard to convergence, I am also a bit surprised by the change when going from 76 to 152 points on each axis. Does this mean they have not fully converged or is this within the typical error range of Bader analysis? Could you give me a hint what other calculation parameters I should investigate? The k-grid seems rather dense already and the cut off energy quite high.

Hi,

I have the same problem with addgrid. I am investigating adsorption of N2H4 on Cu surfaces. I use the NGF (168,288,480), I also doubled them and got the same core charge (AECCAR0). With or without using addgrid I got different bader charges for N2H4. You can see the ACF.dat for the molecule with or without addgrid.



with ADDGRID=.true. , LELF= .true.
...
121 9.5433 5.6440 10.7942 4.9661 0.5160 5.7966
122 8.2271 6.1324 10.4136 5.2933 0.4852 7.0499
123 8.3788 6.8969 9.7296 1.0773 0.1229 18.5483
124 7.7130 6.5108 11.2174 0.8615 0.1577 352.6616
125 9.4087 4.9166 11.5005 1.0025 0.2132 435.2063
126 10.1237 6.3919 11.1885 0.9697 0.1462 274.8997


with ADDGRID =.false. , LELF= .false.
...
121 9.5433 5.6440 10.7942 5.7737 0.6562 47.6623
122 8.2271 6.1324 10.4136 5.7680 0.6468 25.4096
123 8.3788 6.8969 9.7296 0.5726 0.1556 5.3323
124 7.7130 6.5108 11.2174 0.5999 0.1810 262.0719
125 9.4087 4.9166 11.5005 0.5687 0.1612 322.7024
126 10.1237 6.3919 11.1885 0.5812 0.1462 191.3092

Any help would be appreciated.
Saeedeh

Unfortunately, I don't have experience with the addgrid setting, or how it effects the total charge density. As I mentioned before, my impression was that this should largely effect the core region, so I am very surprised that you see such differences in the Bader charges. Clearly this needs to be investigated further. I would suggest starting by comparing the total charge densities with and without the fine support grid. Subtracting the two charge densities seems like a good way to start figuring out how the charge density changes, and if there are significant differences away from the cores in the valence region.