Geometry optimisation converges to nonsense

hfruchtl hfru... at
Fri Feb 21 18:06:33 UTC 2014

Just to report back: The cutoff value was indeed the culprit. With 200 it 
converges to something that at least consists of two H2O molecules, and 300 
gives pretty much the result I expected. It takes a very long time though, 
and I was drawn to CP2K because of its famed speed.

Anyway, I will move on to systems closer to what I really intend to do 
(materials and surfaces). Using CASTEP or VASP at the moment, but some of 
the systems are sparse, so using a non-PW basis should improve things. Yes, 
I know, no K-points and therefore no metals...

Thanks for your help,


On Thursday, February 20, 2014 7:29:02 AM UTC, Florian Schiffmann wrote:
> Hi Herbert,
> this is clearly a cutoff problem (100 is way too low for whatever you want 
> to do). This breaks translational invariance with respect to the grid and 
> artificial minima arise simply from the atomic positions relative to the 
> grid which explains the strange "minimal energy" geometry you observe.
> Looks to me like you created your input from one of the regtest files. 
> Keep in mind there, the regtest are tuned for speed not for meaningful 
> results. They are simply consistency checks for which convergence doesn't 
> matter. For more meaningful inputs have a look at the benchmarks but even 
> there they are not meant for high quality results.
> Flo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list