[CP2K:1892] Re: Geometry optimization convergence problem

[Teodoro Laino]

> Well, I don't ask for an Ewald summation, since I consider an isolated
> system. I want only interactions between explicitly defined particles.
> Since only pair nonbonded interactions are considered - no need to
> solve Poisson eq.. So should I just omit PSOLVER?
>   
Considering an isolated system is absolutely fine but, please, keep in 
mind that the electrostatics in this way is only computed
in the non-bonded interaction loop. This means that you don't compute 
all N*(N-1)/2 interactions but only the one that fall
in the cutoff of your non-bonded interactions. This could create (and 
very probably creates) serious problems.

The only way to by-pass this problem would be to use a cutoff for the 
non-bonded interactions which is cell/2.
In your case (13K atoms), the whole evaluation of the N^2 terms would 
become quite expensive..

The ideal thing to do would be to go for EWALD (SPME or PME, which are N 
log(N) ) and then decouple the interactions.
I was looking into this possibility last year.. but there's still 
something which does not work properly.. so I would not
suggest you to use this smart approach yet! (unless you don't want to go 
and debug it :-) )

> P.S. Actually this calculation is a preparation for a QM/MM run.
>
>