UT theoretical chemistry code forum

Dear colleagues,
I am using Bader code to analyze cube files generated either by Gaussian or GAMESS, and recently I have encountered into problems analyzing results of calculations for molecules with Y and La atoms when the basis sets with effective-core potentials were used (at the same time, there was no problem with Lu). When dense grids were used in constructing cube files, there are simply no electrons on Y or La, and MIN DIST parameter is very small (say, 0.025 - it turns to be just a step-size in the grid). Formally, this problem can be "solved" by using rather sparse grids, but in this case accuracy of the charges is very low. I guess the problem is that when ECPs are used in GAMESS or Gaussian, resulting electron density has the volume with zero value close to the nuclei, and Bader code sees this zero-value volume as an atom. So, my question is: Is it possible to overcome this problem somehow still using ECP basis sets? And my suggestion for developers: maybe it would be nice if one could model the nuclei in the Bader analysis not as a point, but as a finite size sphere (the size could be a user-defined parameter). In such case some problems resulting from incorrect treatment of the core electron could be gone...

We have struggled with a similar problem with pseudopotentials in VASP. The code can deal with this in a fairly simple way by associating each Bader volume with the nearest atomic center. But, if the charge density minimum between atoms is missing, there will be no maximum near atomic centers. In this case, it is my view that the Bader analysis will be qualitatively wrong, and no rough approximation of the core charge will yield accurate Bader charges. Using a spherical region around each atomic core is one such approximation.

The VASP developers were able to write out the atomic charge so that it could be added to the valance charge density. This has been an excellent solution. It would be perfect if a similar thing could be done with the ECP basis sets. If the core charge could be obtained from the generator of the ECP, we could perhaps add this in manually to correct the charge density and fix the analysis. Even if the missing core charge could be modeled with a functional form, it could be used to fix the charge in the region of the dividing surfaces between atoms.

By the way, I don't understand the course grid comment at all. I don't see how a course grid would improve the problem with missing core charges.

graeme wrote: By the way, I don't understand the course grid comment at all. I don't see how a course grid would improve the problem with missing core charges.

It's not really "improve" - if the grid is too sparse, zero-density volume around the nuclei may be simply "missed" by the integrating program. But overall accuracy in such a case is not very good. At the same time, I have noticed that the difference values converge much faster. Say, if I am interested in the charge of La in some molecule, I can perform independent calculation with sole La(3+) cation using the same grid as in the molecule calculation. Even if the grid is too sparse for evaluation of individual charges, the difference between La in the molecule and La(3+) can be quite accurate. And then, knowing that the actual charge of L(+3) is +3, the charge of La in the molecule can be calculated.

By the way, is it possible to limit print-out of the "-p all_bader" command to only some selected atoms? I am working with molecules of 80-100 atoms, and full print-out with good grid is really VERY heavy.