[Wien] convergence of Bader (aim) calculations

[Jorissen Kevin]

Dear users,
 
I'd like to consult you about the convergence of Bader calculations.  I have done a few calculations of a KC60 system while varying the basic parameters, and performed aim each time.  It's the Bader charge I'm mainly interested in.
 
It seems the k-mesh is not important (not surprising for this very big system), more surprisingly, better charge expansions (gmax up to 20, lmax up to 10) hardly mattered either, and also increasing the nth nphi and npr of case.inaim from 30 to 40 seems irrelevant.  Rkmax is really important - at rkmax 7 (for RMT(C)=1.3 and RMT(K)=1.45, approximately) I am starting to see (I hope :)) signs of convergence.  Unfortunately I doubt I'll be able to go significantly beyond 7.
I should add that the fullerene contains a lot of vacuum, and that the Bader volumes are not expected to be spherical at all - while the C-C bondlength limits them in some directions, they may extend very far in other directions.
 
My question is : 
1/ which values of the named parameters would you expect to be reasonable?  I'm especially interested in hearing about nth,nphi and npr - perhaps an increase of merely 10 points is ridiculously small?  (I'm sampling the full interval, by the way - I'm not using symmetry there).
And secondly : 
2/ What accuracy (I'm talking strictly calculational here - not about comparison to experimental data or other calculation techniques) would you expect to achieve?  0.1?  0.01? 0.001?  I recall from a paper by P. Blaha and others on some oxide system (Cu oxides, I believe?) that the numbers considered there were quite small.
 
Also, in contrast to simpler systems I studied before, in this case the sum of the Bader charges is not equal to the core charge.  This is not inconceivable, of course.  I'm just wondering :
3/ Is this unusual, even for complex systems?  Would this effect be expected to continuously decrease as a function of basis set size?
4/  Obviously this somehow limits the accuracy of the results, especially when Bader surfaces seem to be overlapping slightly.  Which brings me back to question 2 ...
 
 
I will welcome any comments from any of you.
Thank you very much,
 
 
Kevin Jorissen
EMAT
Universiteit Antwerpen (University of Antwerp)
Groenenborgerlaan 171
B-2020 Antwerpen
België