[CP2K-user] Convergence problem in further steps of ab initio molecular dynamics

Mohammad Shakiba mshakiba.... at gmail.com
Sat Mar 30 11:19:32 UTC 2019


Dear Matt,

Your video and also the videos inside the playlist were very useful and I 
learned so much and thank you so much for that. Although I modified my 
input but I have to recompile my cp2k version with libint and elpa as well.

What I noticed in the videos was that you used SCF_GUESS RESTART and 
WFN_RESTART_FILE_NAME ${project}-RESTART.wfn. Now I have some other 
question in my mind:

** Does using a RESTART file make the calculations and convergence faster 
and better? How can I obtain such a file before starting AIMD because I 
found no such file but I have, for example, *.restart.bak.1 file (after 
running calculations) but not *.wfn file? and I don't know if I can use it 
in the EX_RESTART section and obtain good convergence.

** As I saw the "RESTART", I remember that in some papers they have 
equilibriated the system for some pico-seconds with time step of 10fs and 
then continued the calculations with smaller time steps like 2fs. To this 
end I guess that I have to make two inputs: one with 10fs time step and the 
other with a EXT_RESTART file from the previous and 2fs time step. But I 
don't know what to put for SCF_GUESS for the second input? ATOMIC?

On Friday, March 29, 2019 at 4:17:26 PM UTC+4:30, Matt W wrote:
> Hi,
> hybrid functional calculations using MOLOPT basis sets are extremely 
> expensive. You probably need to use the Auxiliary Density Matrix Method 
> approximation. See these execises 
> <https://www.cp2k.org/exercises:2016_summer_school:hfx> and some slides 
> <https://mattatlincoln.github.io/talks/GhentWorkshop/#/> maybe this talk 
> <https://www.youtube.com/watch?v=snG4fbpI0_g&list=PLrmNhuZo9sgYJqeTWUhdhtYu3Ol988prg&index=6&t=0s>
> .
> Matt
> On Thursday, March 28, 2019 at 8:15:26 PM UTC, Mohammad Shakiba wrote:
>> Dear CP2K users,
>> I have installed CP2K on our university cluster with intel 2019 and it 
>> works properly for tests. However, when I use it for ab initio molecular 
>> dynamics with my structure, the first 20 steps converge but the convergence 
>> in the next step takes too long and it does not converge. Actually, it 
>> becomes close to the convergence limit of 1.0E-6 (like 2.04E-6 which was 
>> the closest) but does not converge even in outer loops and oscillates 
>> between 2.0E-6 and 1.0E-4.
>> Can you help me find out why this happens and how can I fix it? is it 
>> because of my basis sets and potentials? or my structure? Is it practical 
>> to increase the MAX_SCF to a high number like 1000? What are the normal 
>> ranges for this limit?
>> By the way I tried both CG and DIIS but the problem still exists.
>> Thanks in advance.
>> My structure contains Pb ans S atoms -> (PbS)16. Here are some main parts 
>> of my attached inputs:
>>     BASIS_SET_FILE_NAME BASIS_MOLOPT   ! I used MOLOPT bases file
>>     POTENTIAL_FILE_NAME POTENTIAL          ! with GTH pseudopotentials 
>> and PBE functionals
>>     &MGRID
>>        CUTOFF [Ry] 400
>>     &END
>>     &QS
>>        METHOD GPW 
>>        EPS_DEFAULT 1.0E-10
>>     &END
>>     &POISSON      
>>       PSOLVER  MT
>>     &SCF                              
>>       MAX_SCF 400
>>       EPS_SCF 1.0E-6
>>       &OT
>>         MINIMIZER DIIS                                            ! I 
>> also tried CG
>>         LINESEARCH 3PNT
>> !        STEPSIZE 0.08
>>       &END OT
>>       &OUTER_SCF                                                 ! Outer 
>> SCF loop for 10 times
>>         MAX_SCF 10
>>         EPS_SCF 1.0E-6
>>       &END
>>     &XC
>>       &XC_GRID
>>         XC_DERIV SPLINE2
>>         XC_SMOOTH_RHO NN50
>>       &END XC_GRID
>>       &XC_FUNCTIONAL 
>>          &PBE
>>          &END
>>     &END XC
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190330/360c7f50/attachment.htm>

More information about the CP2K-user mailing list