[CP2K-user] Seeking for recommendations of running AIMD with transition metal surface

[Nam Tran]

Dear Marcella,

Thank you a lot for your quick reply

You can try to speed up the convergence by changing the smearing 
> temperature and the initial guess.

I did test with different initial guess. I will try change the smearing 
temperature. However, I wonder if there is any typical range of the 
electronic temperature? I addition, is there any way to compare the 
electronic temperature with the smearing width in VASP or QE ? 

The performance can be improved by optimising cholesky method and the 
> distribution over processors.

I have checked the cholesky methods and come up with the default one.
Please tell me which keyword I can control this processors distribution and 
is there any general procedure of testing this ?

Please consider that Ni has magnetic properties. 

I am scare of that too. I did turn on the spin-polarized calculation and it 
significantly increased the simulation time.  
I only want to investigate the absorption energy of some molecule. Is it 
acceptable to check if the absorption energy 
does not change much and only use the non-polarized calculation ?

Sorry for putting a lot of questions and thank you so much for your time

Regards
Nam



On Saturday, 26 October 2019 01:59:10 UTC+11, Marcella Iannuzzi wrote:
>
> Dear Nam,
>
> I would say that the behaviour you observe is perfectly normal for a 
> metal. 
> You can try to speed up the convergence by changing the smearing 
> temperature and the initial guess.
> The performance can be improved by optimising the cholesky method and the 
> distribution over processors.
> Please consider that Ni has magnetic properties. 
>
> Regards
> Marcella
>
> On Friday, October 25, 2019 at 1:34:34 AM UTC+2, Nam Tran wrote:
>>
>> Dear CP2K users and experts,
>>
>> I am trying to perform some optimizations and AIMD simulations with a 
>> metal transition system (including a graphene layer on top of a Ni (111) 
>> surface). However, the SCF convergence is very poor, I have to reduce the 
>> mixing alpha to 0.02 in order to make it converges (my input and output 
>> files are attached below.). Because of this small mixing value, my MD step 
>> requires ~38 SCF circles. So I am looking for possible solutions to make 
>> the SCF converges better and reduce the simulation time.
>>
>> I appreciate any advice and recommendation.
>>
>> Best Regards
>> Nam Tran
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20191025/cc47d0c9/attachment.htm>