[CP2K-user] Ehrenfest MD together with UKS and HFX (using ADMM) does not work

[Vladimir Rybkin]

Dear Daniel,

could you please sent the input?

Yours,

Vladimir

среда, 16 октября 2019 г., 19:18:13 UTC+2 пользователь Dan_M написал:
>
> Dear community,
>
> I am trying to perform an Ehrenfest MD run in a system that requires both 
> spin-polarization and a hybrid functional in order to be described 
> correctly (in case you are wondering, an ionized water dimer, i.e. 
> H2O...H2O^+). For the HFX, I am of course using the ADMM method.
>
> I have done many tests and it happens that all three methods together do 
> not work. The code crashes with a segfault, and the last lines in the 
> output file are these (no warning/error message, the initial T I set is 
> actually 0):  
>
>  ************************** Velocities initialization 
> **************************
>  Initial Temperature                                                      
> 0.00 K
>  COM velocity:            0.000000000000      0.000000000000      
> 0.000000000000
>
>  *******************************************************************************
>
> Given the system size, I assume it is not an issue of memory or anything 
> similar, 
> Now the curious thing is: when trying to narrow down the problem, all the 
> following combinations of methods DO work fine:
>
> - RT_PROPAGATION (instead of Ehrenfest) with UKS and HFX is ok.
> - EHRENFEST_DYN with UKS and no HFX is ok.
> - EHRENFEST_DYN with RKS and HFX is ok.
> - EHRENFEST_DYN with RKS and no HFX is of course ok.
> - And as expected, regular MD with UKS and HFX is ok.
>
> I have tested more thoroughly version 5.1, and also a bit version 6.1 (not 
> that thoroughly, but in the few tests I did, the case EMD+UKS+HFX does not 
> work either). I have done quite some calculations using all these different 
> approaches separetely (EMD with RKS, regular MD with HFX/ADMM, etc), and I 
> am quite confident that the code is well compiled and produces correct 
> results. I did not yet try the trunk version since it seemed to me that 
> there have been no new additions in the EMD routines w.r.t. versions 
> 5.1/6.1, so I assumed that whatever is the problem probably it is still 
> present there.
>
> Since the RTP+UKS+HFX is fine, and given that the code crashes at the 
> stage of MD initialization, seems to me that there is something happening 
> when trying to compute the forces. But it is quite puzzling that the forces 
> for EMD+HFX+RKS are fine, EMD+UKS+regularGGA are fine, but together 
> EMD+HFX+UKS then it crashes.
>
> Now, before I post the input file/s: Can anyone that has a deep knowledge 
> of this part of the code tell me whether there is something that has not 
> yet been implemented, or should this in principle work? Has anybody 
> successfully done such a calculation?
>
> Thanks a lot,
> Daniel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200113/dba88528/attachment.htm>