Page 1 of 1

Error when using CHGCAR_sum

Posted: Sat Nov 12, 2016 6:32 am
by tp.joe
Hi Prof.,

I have read through other thread regarding having an error "ERROR: should be no new maxima in edge refinement", I tried checking that the cause of this error is not as you mentioned earlier. My setup is a slab model with spin polarization, van der Waals interaction (optB88-vdW). I have checked that the AECCAR0 doesn't have values order of 10^+-90. This occurs only when CHGCAR_sum is used, if only CHGCAR is used, everything seems to work out just fine.


I could not upload the file on this forum due to size issues, please download it from my google drive using the link below, thank you for your help in advance.


https://drive.google.com/open?id=0ByHLt ... HVhX19KSTQ

Re: Error when using CHGCAR_sum

Posted: Sat Nov 12, 2016 9:16 am
by tp.joe
I think I know now, it's likely to be because negative values in AECCAR0 correct?, if so do you have any suggestion of what may be the cause, my INCAR is,

SYSTEM = Ni_slab_phenol
PREC = Accurate
ENCUT = 500
IBRION = 2
NSW = 0
ISIF = 2
NELMIN = 6
EDIFF = 1.0e-04
EDIFFG = -0.02
VOSKOWN = 1
NELM = 250
ALGO = Fast
ISPIN = 2
LAECHG = .TRUE.
LCHARG = .TRUE.

LDIPOL = .TRUE.
IDIPOL = 3
ISMEAR = 2
SIGMA = 0.15
ADDGRID = .TRUE.

#FOR VAN DER WAALS
GGA = BO
PARAM1 = 0.1833333333
PARAM2 = 0.2200000000
LUSE_VDW = .TRUE.
AGGAC = 0.0000

#REAL SPACE PROJECTION
LREAL = Auto

I have tried it with high ENCUT (1200), doubled fine FFT grids, without dipole correction, and without van der Waals correction, however none of these seems to be the cause. Any suggestion or insight would be very much appreciated.

Re: Error when using CHGCAR_sum

Posted: Sun Nov 13, 2016 9:16 am
by tp.joe
I found a way to "alleviate" the issue. For future new users that may encounter the same issue, you should increase NG(X,Y,Z)F by MANY MANY folds, at least in my case. Increasing this should reduce the negative values in CHGCAR_sum significantly, and hence eradicate oscillation.

In my case, I did my best by increasing the fine grid to its maximum (around 0.01285 mesh density, denser than this the program won't run, and it's not a memory issue). However, there are still some negative values left in the CHGCAR_sum file, although extremely infrequent (have to squint an eye to locate ones), and hence it still shows an error message " ERROR: should be no new maxima in edge refinement ", but only in very limited times.


It would be nice if Prof.Henkelman could address:
1. if this would affect the calculation significantly?
2. if there is any other mean to completely eradicate negative values in CHGCAR_sum apart from increasing NG(X,Y,Z)F?
3. if there is any way to check, at least back-of-an-envelope calculation, that the values obtained make sense?

Re: Error when using CHGCAR_sum

Posted: Sun Nov 13, 2016 11:16 am
by tp.joe
Just an extra information regarding question no. 3 in the previous reply, I did an integration of no. of core electrons from AECCAR0 of a simpler system (just phenol), which is C6H6O1,


therefore principally the core electrons should be 14 electrons



using the maximum accuracy, the core electrons are 13.7195
while if I use lower accuracy (basically, PREC = med, without added grid or NG(X,Y,Z)F specification) the core electrons are 15.



So essentially, I can get 72% more accuracy, but still not a perfect core electron charges.

Any comment on this, like is this within typical error margin, would also be appreciated.


Thank you sir.

Re: Error when using CHGCAR_sum

Posted: Sun Nov 13, 2016 1:58 pm
by graeme
It is very hard to resolve the core charge with sufficient precision to integrate it accurately. The density cusps at the atomic centers are the real problem. The total charge does, however, have the proper shape for the Bader partitioning. The scheme that we recommend is to partition space according to the total charge but then integrate the valence charge in those Bader volumes. The valence charge has the correct total number of electrons, by construction.