UT theoretical chemistry code forum

Dear everyone,

I've been trying to get Bader charges for a Fe/MgO system. The system size is
5.42 by 5.42 by 38.61 Angstrom, the best grid I used is 300x300x2100. I used
EXCITING, which is a all-electron FP-LAPW code, to get the all-electron charge
density. The system has 192 electrons, however, Bader analysis leads to 192.053 electrons.
Probably, this error is to big, especially for charge transfer analysis.
300x300x2100 is the best grid I can afford, the memory of our cluster is 8GB.

My question is : Is it possible to do Bader analysis using all-electron density only?
Anybody has done Bader analysis with all-electron density?
Any information and advice would be greatly appreacited. Many thanks in advance.
Best wishes,

Xiaobing

This is a good question. We have never used this code, but here's a few thoughts:

What is the normalization for the charge density? In the Bader code, the total charge is calculated as an integral of the charge density over the grid points. It looks like the EXCITING code is using a different normalization which is leading to this difference in the total charge - it doesn't have anything to do with the Bader partitioning.

It would be good to make sure that there are no problems due to the boundaries. Our integration over grid points assumes that there is one voxel per grid point. In non-periodic Gaussian calculations there is one extra grid point in each dimension. I'm not suggesting that this is the source of the error, but it would just be good to make sure that there are no differences in convention in the structure of the charge density grid.

Is the error in the total charge systematic or random; i.e. if you move the atoms around, is the total charge always overestimated by the Bader code? If it is systematic, and you always get 192.053, the charge density differences may be ok.

The normalization problem is likely due to the rapidly changing charge density around the ions, where the specifics of the integration method is most important. This is actually the least important part of the charge density for the Bader analysis, where the dividing surfaces are in the bonding region between atoms. What we do with the vasp code is to use the total charge density (which does not give an accurate total charge) to calculate the Bader surfaces, but the integrate the valance charge inside these surfaces. If you can generate the valance charge density with EXCITING, and it integrates to the correct total valance charge with a reasonable grid, you could use this same trick. To use our code in this way, run:

bader valance_charge_grid -ref total_charge_grid

Good luck.

Dear Graeme,

Thanks very much for the reply.
I checked the .cube file grid points are found only on one side of the boundary, i.e. no
repeatitive grid points due to using PBC.
I noticed that all-electron density is very sharply peaked at ion positions, the error
in total charge was even worse using coarser grid mesh. You mentioned, the total
charge density from vasp also leads to not accurate total charge, probably this is
the reason of the error for my case too. I agree with you that this error has nothing
to do with the partitioning surface. Unluckily, EXCITING can only remove the
contribution of electrons in a very small core, like 1s2 for Oxygen. I believe I would
not have this problem if, as you mentioned, I could have a separate valence charge density.
I'll shift to vasp to redo the charge analysis.
Thanks again.
Best regards,

Xiaobing