Error in charges increases every OUTER_SCF cycle.

[Florian Schiffmann]

Hi John,

I doubt that it has to do with the smoothing (as in the posted previously, 
I am anyway not a big fan of that option). . I have not tested it yet but 
the usual reason for such a behaviour is a badly conditioned overlap 
matrix. Means, the inversion of the matrix goes wrong in the preconditioner 
and that causes a bad P. 
There are various possibilities why that might happen. In your case it 
might be your box. 4A is very small. This causes some basis functions to go 
over the boundaries of your box. Furthermore you will use many periodic 
images (see total number of particle pairs). This might lead to a near 
linear dependence in S and so cause problems in the inversion.
In general, as Cp2K does not use k-points it is not a good idea to use such 
small unit cells. Using a 2x2x2 box won't be much more expensive and will 
give more reasonable results and probably fix your electron problem. 
Furthermore you have a very tight convergence criterium here you might even 
need a higher cutoff (just tried with 2x2x2, 800 works) to be able to reach 
that convergence level.

Flo 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20130517/0b819520/attachment.htm>