restarting from an already optimized structure

Florian Schiffmann flosch... at gmail.com
Sat Mar 8 07:04:21 UTC 2014


Dear Chaitanya,

usually if you have a properly converged structure and restart with the 
identical coordinates CP2K is very happy and will tell you it is converged. 
Without your input file with the exact details of your calculation it is 
fairly hard to tell what is going on. Nevertheless I will put some random 
guesses on.

a) your last structure wasn't perfectly optimized but reached the default 
step liimit of 200 steps
b) you use the default cutoff of 280 with elements like N/O/F. For these 
elements higher cutoffs (or smoothing) are needed to avoid grideffects 
(ripples) on your potential energy surface. If that's the case geometry 
optimizations will always have prolems to converge.
c) you use a nonperiodic system and have the CENTER_COORDINATES option on. 
In combination with the cutoff problem this leaves you at a different 
position w.r.t. the grid points and thus with a different energy which can 
be further optimized.

I think b) and c) are most likely. Try a cutoff of 600 and see whether the 
problem still exists. 
Of course higher cutoffs are expensive but it will give you a smooth PES 
and geometry optimizations will need much fewer steps and your ersults are 
converged wrt the auxillary basis.

One more comment to  your a)
I think the bit about *"non-variational"-ness of the GPW method *you got 
from a prevous post from Jürg. He was talking about the convergence of the 
plane wave cutoff (better plane wave basis != lower energy). With fixed 
settings (basis,...) the GPW energy for a system is variational (lowest 
energy = best solution). Therefore forget about that point unless you try 
to converge your plane wave basis (which might be an idea, see b) :-))  

Cheers
Flo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140307/733c1334/attachment.htm>


More information about the CP2K-user mailing list