UT theoretical chemistry code forum

I have analysed the CHGCAR obtened with vasp with the pseudopotential PAW type. but the main.F is not changed as suggested in http://theory.cm.utexas.edu/henkelman/r ... r/vasp.php. So my question is how many this can influence the analysis.
thanks for your help

The change in main.F will add some core electrons for some of the elements.

For elements on the left, use the _pv or _sv versions of the pseudopotentials to get the inner electrons. Elements on the right should already have enough valence electrons for the Bader analysis. In these cases, the main.F change should not affect the analysis.

For the problematic elements in the middle, the partial core can be enough to reproduce the charge density maximum at the atomic center. In this case, the change can save the analysis.

Really, the analysis is well defined for a full charge density. If the surface of minimum charge between atoms (in the bonding region) is accurately reproduced, the analysis will be correct. To the extent that it's not - you're on your own.

In VASP 5.0 we'll be able to get the core charges, and this will no longer be a problem.

Hi, I'm trying to use your program for analysis of densities from SIESTA. I've exported to CUBE and the analysis of H2O gives the expected for pseudopotential charges 0e and 2e on H and O respectively.
So, now I'm thinking about implementation of core charges.

You said that partial core can help. Does it mean that the actual shape of the core doesn't matter, it only should create a maximum on atom?

Can I use for example Gaussian shape normalized to give needed amount of electrons in the core?
Would the amount of electrons in the core or radius of such pseudo-core affect the results of Bader analysis?

I've just found in another thread that you propose to use PAW core, not all-electron core.

I also found that Abinit also has Bader analysis and as far as I understand, they use the original AIM program. And they propose to use the core from all electron calculations:
http://www.abinit.org/Infos/aim_help.html
[quote]About the core density files:
LDA core density files have been generated for the whole periodic table, and are available on the ABINIT web site. Since the densities are weakly dependent on the choice of the XC functional, and moreover, the charge density analysis is mostly a qualitative tool, these files can be used for other functionals. Still, if really accurate values for the Bader charge analysis are needed, one should generate core density files with the same XC functional as for the valence density.[/quote]

I'm not familiar with VASP so could you explain the differences between PAW and all-electron core.

This will depend upon the pseudopotential and in particular the number of valance electrons as compared to the number of bonding electrons. In the worst case, when all the valence electrons are forming bonds (e.g. Si) the shape of the core charge is important because the core charge extends to the surface of the Bader volumes. It might be possible to use a frozen core from an all electron calculation of each atom type, and add this to the valance charge. Ideally, this core should be consistent with the pseudopotentials used, but it might not matter too much. I could generate cores using vasp if you want to try this.

On a positive note, my student Wenjie Tang has done several tests with and without core charge, and found very little different for many materials. In particular, metals elements on the right of the periodic table (e.g. O) have sufficient valance electrons to reproduce the core, and vasp has pseudopotentials with explicit inner electrons for elements on the right. In our experience, if the Bader analysis fails using a valance charge density, the shape of the core will be important to get the analysis right.

The PAW framework has a frozen core. Valance electron wavefunctions are constructed to be orthogonal to the core wavefunctions. In a sense, this is similar to an all-electron calculation except that the core is not allowed to respond to the SCF calculation.

We are now able to generate all electron charge densities within the PAW framework, but this is a recent development, so previous threads may be been referring to the use of harder PAW potentials as compared to US, or just comparing different potentials to make sure the analysis is insensitive to the choice of pseudopotential.

I expect that using a calculated core from LDA calculations will be much better than using an arbitrary Gaussian function. I think their comment about this makes sense.