Page 1 of 1

pseudo-potential dependent charges

Posted: Thu May 26, 2005 8:05 pm
by DJC
Dear Bader develop team,

I tried to use this Bader charge analysis program and found it is very helpful. As a test, I run the program on CHGCAR file calculated using different psedopotentials on MgO. I found that if I use psedopotentials (US or PAW) that have no semi-core electrons on Mg the final charge analysis gave results of close to 100% ionicity of the cations, i.e. +2.0 for Mg. But if I use psedopotentials with semicore electrons (PAW _pv) Bader analysis gave +1.75 for Mg, close to the value calculated in your paper. So is the psedopotential so important in the charge analysis? what if some species (such as Si) that we could not find psedopotential that have semicore electron?

Another observation is that the charge files calculated with PAW potentials gives much fewer maxima than the one calculated using US potentials.

In the website you mentioned that when reading the results we should pay attention to the minimum distance to the surface (I guess it refers to the surface of zero charge flux). But when I checked the core radius in the POTCAR file the radius is generally quite large, for example, for Si is 2.48 and for O is 1.55 (I am not sure this has the unit of Angstrom of Bohr) while the minimum distance from Bader analysis is 0.13 and 1.39 ( in unit of Angstrom? for US potential/no core electron case). So are these results reliable?

Thanks,
DJC

Posted: Thu May 26, 2005 11:05 pm
by graeme
The issue with Mg, and all elements on the left side of the periodic table is that the minimal number of valence electrons does not provide enough charge to capture the charge density maximum which should be at the Mg atomic center. For these elements you must use the _pv or _sv potentials to get any meaningful numbers from the Bader analysis. No element should ever have a charge of zero in this analysis.

To compare the effects of different pseudopotentials, you could only compare US vs PAW using either the _pv or _sv versions. I would be surprised if you didn't find one Bader maxima per atom using any of these four potentials (for bulk MgO).

The elements on the right side of the table typically have enough valence electrons to capture the cores. The elements in the middle, such as Si (and H) are tricky. For these, you need to use the modification to main.F described on out Bader analysis webpage.

The radii described in the POTCAR files are in Bohr. What you found with the US potential with 2 electrons is that the Bader volume for O extended to the Mg nucleus -- this is why the charge on Mg was zero. This is a clear indication that you need the _pv or _sv potential to get a meaningful result.

Posted: Thu May 26, 2005 11:14 pm
by graeme
Another quick note about pseudopotentials -- this is clearly an area of concern for using the Bader analysis with vasp CHGCAR files. The vasp folk have said they will provide a way of getting proper core charges. The speed at which this happens is likely related to the interest of their users, so it would be excellent if people happen to mention this issue to them.

Posted: Fri Oct 14, 2005 12:16 pm
by egillsk
Hi Graeme and Andri

I have calculated the H2O molecule in the ab initio code dacapo. Here are the results of the bader analysis, ACF.dat:

# X Y Z VORONOI BADER % MIN DIST
------------------------------------------------------------------------------------------
1 14.2841 15.7456 15.8223 1.2366 0.0011 0.0134 3.6395
2 14.2814 14.2838 14.6904 5.4305 7.9976 99.9714 4.4779
3 14.2841 12.8222 15.8226 1.3328 0.0012 0.0151 3.7670

As you can see, the H atoms (no. 1 & 3) have only about 0.001 electrons or about 0.1% of the total number of electrons in this system, which is rather off the value that you calculated in your Bader paper, or 0.42 el per H. The pseudopotential for H is ch_e9g4.pseudo and for O it is co_gef_e13_gga.pseudo. In these calculations are 1 el per H and 6 el for Oxygen, so 8 el in total. These are the only pseudopotentials offered for O and H in dacapo (see http://oldwww.fysik.dtu.dk/CAMPOS/Docum ... short.html).

Have you any idea what I can do? The ultimate goel is to make a bader analysis on dissolved protons in water (currently all the H in the water have almost 0 charge so we don't know anything about how much our protons are dissolved)

Best regards,

egillsk

Posted: Fri Oct 14, 2005 8:21 pm
by graeme
H-O bonds are definately a problem with pseudopotential codes, and possibly C-H. The results in our paper were generated using an all electron calculation with Gaussian.

Even though H has 1 electron, the electrostatic repulsion is smoothed out with the pseudopotential, and as you have found, it can remove the charge density maximum at the H atom.

In order to properly resolve these issues, we really need a way of reconstructing the core potential from the PAW formalism. I don't know how to do this, but it is clearly essential for some types of bonding.