AIMD simulation: increase of the number of the SCF cycles per MD step after some MD steps

Max latevimax.... at gmail.com
Mon Sep 29 16:47:21 CEST 2014



Le lundi 29 septembre 2014 09:16:06 UTC+2, Florian Schiffmann a écrit :
>
> Hi Max,
>
> Hi Flo, 
 
Thanks a lot for your kind reply !   

I am currently looking a bit in the performance of the preconditioners. 
>

These are great news :) 
 

> Unfortunately, the better preconditioners are expensive as they require 
> diagonalization and I don't see a way around it at the moment. 
>

Ok. 
 

> There are some ideas around how to improve Full_single inverse, but that 
> will take a little while and nothing promised. 
>

Ok. This is fine  with me. :-) Please do no hesitate to tell me if you want 
I run some tests. 

Full Kinetic is the best choice in terms of balancing cost and efficiency 
> for large systems. The performance of this preconditioner strongly depends 
> on the ENERGY_GAP value in your input. Have you tried playing around with 
> it? Quite often a somewhat higher than default value seems to work best.
>
>
I haven't played with the value of energy_gap when using full_kinetic yet. 
I'll give it a try and 
will keep you informed. 

Thanks again for your reply and your help. 

Cheers,
Max

Cheers
> Flo
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140929/88e7eedd/attachment.html>


More information about the CP2K-user mailing list