[CP2K:4782] some strange phenomena on running CP2K with PBE functional

[Huan Wang]

Dear Marcella,

Thank you for your reply. I know what your mean that the increasing number
of SCF cycles leads to the longer time to converge the wave function.

But it does not make sense, because I also checked the BLYP results. For
the same benzene-water cluster with the same input settings, when I used *
32-CUP*, the CPU-time increased from about 40 second per step to about 150
second per step after 1200 steps and *the SCF cycles also increased from 8
to 37*.

However, when I stopped this calculation and restarted it with *8-CPU*, it
only takes about 40 second per step, *the SCF cycles was about 7*. And most
importantly, the speed and SCF cycles were very stable until finished the
whole 10000 steps.

When I restarted this calculation again with 32-CPU, the CPU-time was
around 150 second per step and SCF cycles was about 41.  *These tests were
performed in the same node.*

It so strange. Why the same calculation takes different number of
iterations when using different number of CPU?

And back to the PBE functional, since I tried 8-CPU, 16-CPU, 24-CPU and
32-CPU for PBE funcitional, the calculation speed always slowed down after
several steps. Why PBE functional do not have an "optimized performance" as
BLYP functional with 8-CPU?   Is it common that the speed of calculation
with PBE functional slower than that with BLYP functional?

Best regards,
Huan


On Sun, Oct 27, 2013 at 9:04 PM, Marcella Iannuzzi <marc... at pci.uzh.ch>wrote:

> Dear Huan,
>
> the reason for the slowing down of the MD simulation is
>  the increasing number of iterations needed  to converge the wave function.
>
> regards
> marcella
>
>  --------------------------------------------------------------
> Marcella Iannuzzi              Phone : ++41 44 635 4479
> Physical Chemistry Institute   FAX   : ++41 44 635 6838
> University of Zurich               E-mail:  *marc... at pci.uzh.ch<hut... at pci.uzh.ch>
> *
> Winterthurerstrasse 190
> CH-8057 Zurich, Switzerland
> ---------------------------------------------------------------
>
>
>
>
> On Oct 27, 2013, at 4:42 PM, huan... at mail.huji.ac.il wrote:
>
> Dear All,
>
> I am a newbie on CP2K and I have some questions on running CP2K with PBE
> functional.
>
> Last week I did some tests on my benzene-water cluster project using both
> BLYP and PBE functional. I found that for BLYP functional, 8 CPU was the
> best choice for my research system, while using more CPUs for MD
> simulation, such as 16, 24, 32, the speed for every step slowed down. It
> can be explained that the more CPUs I used, the more time spent on
> communication.
>
> But when it comes to PBE functional, no matter how many CPUs I used, the
> calculation speed always slowed down after about 40 steps. (The first 40
> steps, the calculation just take about 50 seconds per step, but after 44
> steps, it takes about 300 second for each step). Please see my *
> nve_PBE_50steps.ener* file in attachment. Actually, the structure of the
> benzene-water cluster is the same for both BLYP and PBE input files. And I
> just changed the functional into PBE, choosing the potential as GTH-PBE
> type, and using the same basis set as BLYP functional used. The rest of
> setting are the same with the input file of BLYP functional. I also
> attached the output file for detail information. (*nve_PBE_50steps.out*)
>
> These decreasing of speed phenomena also happened when I used more than 8
> CPU with BLYP functional. But if I use 8 CPU for calculation with BLYP
> functional, the speed was very stable, which was about 40 seconds per step.
>
> My question is why the calculation speed slow down after 40 steps when I
> use PBE functional? Why the speed could not stable as using BLYP functioanl?
>
> Best regards,
> Huan
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to cp2k+uns... at googlegroups.com.
>
> To post to this group, send email to cp... at googlegroups.com.
> Visit this group at http://groups.google.com/group/cp2k.
> For more options, visit https://groups.google.com/groups/opt_out.
> <nve_PBE_50steps.ener><nve_PBE_50steps.out>
>
>
>  --
> You received this message because you are subscribed to a topic in the
> Google Groups "cp2k" group.
> To unsubscribe from this topic, visit
> https://groups.google.com/d/topic/cp2k/hJnTBoZM08I/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to
> cp2k+uns... at googlegroups.com.
> To post to this group, send email to cp... at googlegroups.com.
> Visit this group at http://groups.google.com/group/cp2k.
> For more options, visit https://groups.google.com/groups/opt_out.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20131028/2663228f/attachment.htm>