Bader code paused during an analysis

[jianmin]

Dear all and especially Prof. Graeme,

I tried to analyze the Bader charges of a Ce16O31 cluster in a 20x20x20 cubic box. In this job, only the gamma point was used for k-point sampling.

After I issued the command "bader -b weight CHGCAR -ref CHGCAR_sum", the code paused at a point and never went further, as shown in the following:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~
GRID BASED BADER ANALYSIS (Version 0.95 11/17/15)

OPEN ... CHGCAR
VASP5-STYLE INPUT FILE
DENSITY-GRID: 196 x 196 x 196
CLOSE ... CHGCAR
RUN TIME: 3.12 SECONDS

OPEN ... CHGCAR_sum
VASP5-STYLE INPUT FILE
DENSITY-GRID: 196 x 196 x 196
CLOSE ... CHGCAR_sum
RUN TIME: 2.99 SECONDS

SORTING CHARGE VALUES ... DONE
CALCULATING FLUX ... DONE

INTEGRATING CHARGES
0 10 25 50 75 100
PERCENT DONE: **
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I issued the command on the login node of a HPC center. One reason I can think of is that the box is so large that its needs for memory exceed what the login node can provide. Could this be the reason? If not, what could be the reason?

I have attached the four input files of the job (when you use them, simply delete the ".txt" part of the file names). It is greatly appreciated if someone could run and check it for me. Thanks in advance.

[graeme]

Please attach the CHGCAR, AECCAR0 and AECCAR2 files.

Also, make sure the the electronic structure is converged. If it requires more than 60 iterations, increase NELM.

One more thing: load up the CHGCAR_sum file in a visualization program like vesta and see if it looks physically reasonable. For some calculations there are problems with the charge density files which can make the analysis impossibly slow (and wrong, when the charge density is unphysical). You can see examples of some such files that I have posted here.

[jianmin]

Dear Prof. Graeme,

Attached please find the CHGCAR, AECCAR0 and AECCAR2 files, whose sizes are huge by the way.

It is an SCF converged single-point energy calculation, after 769 SCF iterations though.

The command "bader -b weight CHGCAR -ref CHGCAR_sum" could not make it, but "bader -vac off CHGCAR -ref CHGCAR_sum" works. Can we trust the results from the latter command? What does the option "-vac off" do to the analysis?

Thanks a lot.

[jianmin]

The AECCAR2 file for this time.

[jianmin]

Finally the CHGCAR file.

[graeme]

It's a problem with the core charges written in the AECCAR files. In the snapshot, you can see lines of oscillating charge density leading from the atomic cores along the grid directions. I've seen this before, but I don't know the reason. Not only does this make the Bader analysis take a very long time, the results will also be meaningless.

Screen Shot 2018-07-02 at 9.40.53 PM.png (785.86 KiB) Viewed 202849 times

[jianmin]

Dear Prof. Graeme,

Thanks a lot for your reply.

1. So how do we solve the problem? By using the command "bader -vac off CHGCAR -ref CHGCAR_sum"?

2. Will the problem vanish if we put the cluster in the non-cubic periodic box like 19x20x21?

3. Can we trust the spin charge density from the "chgsplit.pl CHGCAR" and then "bader CHGCAR_mag -ref CHGCAR_sum"?

Thanks.

[graeme]

There needs to be reasonable charge density files from vasp for this analysis to work. But again, I don't know what's going wrong with vasp or how to fix it - I've just seen this problem before (you can find similar posts on this forum). This problem needs to be addressed by the VASP folks - you can clearly see that the charge density is unphysical.

[jianmin]

Dear Prof. Graeme, Thanks for your time on this. So to sum up, it is a pending problem to be solved by the VASP guys. See you around.