[CP2K:9757] MGRID vs KPOINTS for small systems

Sebastian Hütter sebastia... at ovgu.de
Sat Dec 2 18:06:38 UTC 2017


thank you. The way I understood it is that to get better results, I can 
either increase the volume of real-space the integration happens over 
(MULTIPLE_UNIT_CELLS), or can move to a model in the reciprocal lattice and 
sample the 1st BZ at high(er) resolution? That means that for such a small 
system, what I effectively did in my test was the same as if I used a too 
coarse k-point grid? The thing that confused me was that most energies were 
very similar, just that one term wildly different.

For my test system, I need 27 supercells to get the kpt energies, but even 
more for correct stresses - but at about 1/10th the  computation time and a 
little less memory/proc that could be worth it. But now I have all 
quantities output for the replicated atoms as well, making things a bit 
harder to use... guess I'll stick with k-point sampling. Is there a 
break-even point regarding volume or atom/electron count where one method 
is generally faster than the other?

As a related question, is there a way to specify an M-P grid using point 
spacing instead of point counts?

Best regards,
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20171202/8f009f17/attachment.htm>

More information about the CP2K-user mailing list