How to be sure the Bader charge analysis is correct for my calculation?

[oranus]

Hello,

I have done Bader charge analysis for my system (Li surface + one C3O3H4 molecule on it), the default grids are NGXF= 216 NGYF= 216 NGZF= 384,

I increased them to NGX =270 NGY =270 NGZ =384, NGX =300 NGY =300 NGZ =384, NGX =350 NGY =350 NGZ =384,NGX =450 NGY =450 NGZ =384. My question is, should we just increase the grids and check the number of electrons reported at the end of ACF.dat (which it is the total number of valence electrons)?

I see that the number of electrons didn't change in any of these calculations and for all the result is 214. Besides, the charges on e.g. C, O and H did't change that much.
I am not sure if this is the correct way to do these calculations and when I can trust the results and the grids I chose.

I'd really appreciate if you can help me with this problem.

[graeme]

The total number of electrons should always be correct, independent of the grid density.

You want to make sure that the atomic charges are converged to within your desired uncertainty as you increase the grid density. And it sounds like they are.

[oranus]

Thank you for your quick reply. Therefore, based on these results I can just chose the default grid for my calculation, am I right?

I thought according to this note on your webpage ( One finally note is that you need a fine fft grid to accurately reproduce the correct total core charge. It is essential to do a few calculations, increasing NG(X,Y,Z)F until the total charge is correct.) we always should increase the grids and the default grid can not be correct for this calculation.

Just to be sure that I understood you correct, in these analysis we need to increase the grids and check the partial charge on each atoms in the system to see If they change significantly or not.

[graeme]

If you are trying to integrate the total charge density then you will need a fine grid. Instead, we partition the charge based upon the total charge density but integrate only on the valance charge. As with any numerical method, you need to check convergence with respect to the numerical step size; in this case it is the grid density.

[oranus]

Hi, thank you for the information.