UT theoretical chemistry code forum

I'm looking for any hints on what am I doing wrong. I have calculated total charge density distribution from SIESTA/Denchar programs. And was trying to get charges with Bader from the Denchar's output cube file. However Bader's atomic charges computed for alanine amino acid are far from correct. Number of reported electrons is 4 times more than should be.

I tried to make analysis for crystlline structure and structure in water, but in both case resulting defined charges are far from reality. For instance 0.000 on hydrogens (must be about 0.37) and over 7 on N (should be about -1.1). Visualisation of cube file with VMD shows that grid is correct.

The question is: are there any obstacles for getting correct atomic charges with Bader from charge density calculated from density-functional program which uses numerical basis set and pseudopotentials (say like SIESTA)? are there any prerequisities for grid density? 0.05 eA^-3 is enough?

Hi,
we need to add core electrons to our RHO, i.e. make all-electron density (the maximum of density should be on the atom), otherwise Bader analysis doesn't work correctly.

I'm trying to implement this right now for SIESTA, i.e. just add the core the CUBE file after all calculations done by SIESTA.

I can see that even less than 300Ry mesh cutoff is enough for accurate analysis.

Hi,
I've just realized that the problem of wrong charges in water (and in other compounds with O-H, and probably with N-H bonds) in pseudopotential approach comes not from the absence of the core on the O as it is claimed in many places. Even without core one should get something on H, but we get 0.

I've looked closer at it and see that in my calculations there is no maximum on H! thus adding core wouldn't help since H has no core, that's why we get 0 charge.

I thought that Bader analysis is quite unambiguous and reliable but now I'm thinking what's more wrong - pseudopotential approach or Bader analysis.
Even if there is no charge maximum on H, other properties are reproduced quite well, moreover it is hard to call such a result unphysical, since the molecule doesn't dissociate.

I'm trying to investigate what's exactly wrong with pseudopotential approach. I've taken very short cutoff radii for pseudopotentials, i.e. making them as transferrable as possible.
I'm gonna try to use more electrons on O for pseudopotential generation to better reproduce it's actual charge state in water, but I'm not sure it will help.

Any other suggestions are welcomed.

In any case this makes Bader analysis less reliable tool, since for all other elements I don't see such straightforward tests of the correctness of pseudopotentials.

Pseudopotentials are used both to treat core charge implicitly and to smooth the wavefunctions (and charge density) in the core region. The H pseudopotentials do not replace core charge, but they do result in an artificially smooth charge density at the H nucleus.

We have done a Bader analysis of H2O using charge densities generated both with Gaussian and VASP (with the proper H core). In both cases there is a charge density maximum at the H atoms, and a charge of (roughly) 0.5 e is transferred from each H to the O atom.

I'm not trying to suggest that the Bader analysis is somehow intrinsically correct or necessarily meaningful. Its strength (in my view) is that it provides a well defined volume around each atomic center - and one that is independent of the wavefunctions used. The Bader charges can even be found experimentally, since the charge density is an observable.