[CP2K-user] [CP2K:20423] Re: Issues with SOC calculation using wB97XD

[Jan-Robert Vogt]

Dear M. Rullan,

changing the order of the wave functions changes the sign of the couplings, 
but the absolute values stay the same. The block structure is more for 
continents, since the diagonalization routine is not effected by the 
additional elements, if I remember correctly. The routines are based on the 
xas-Module within CP2K, 

https://doi.org/10.1039/D0CP06164F. 

Within the SI, additional information about the used terms can be found.
Regarding your main problem, I am not really sure if I can help. The soc 
routine performs a perturbational correction to the excited state energies, 
thus not interacting with the XC-Kernel itself, but only receives the 
excitation vectors and energies. The strange behavior within the SOC module 
seems to occur for negative eigenvalues, but I have not tested that more 
extensively.

I tried to change a regression test file for the SOC-correction in order to 
lock like your input files. By adding "COMMENSURATE" within the "MGRID" 
section while starting from the atomic guess to the input, I was able to 
reproduce this error. If starting from a better wave function (already 
converged without "COMMENSURATE" at "EPS_SCF 1.0E-7"), the same problem did 
not occur for this very small case (one neon atom within your unit cell). 
Maybe you could use a better "LINESEARCH" option within the "&OT" section 
to optimize the ground state, for example "3PNT" or "GOLD". If they do not 
change in energy but are not converged, the RESTART-file can be used as a 
good guess for the "DIIS" minimizer in some cases. You could also make 
sure, that the "CUTOFF" and "REL_CUTOFF" in the "MGRID"-section are 
converged, if you haven't already. 
Best regards,
Jan-Robert Vogt
Raphaël Rullan schrieb am Donnerstag, 4. Juli 2024 um 13:57:30 UTC+2:

> Dear M. Vogt,
> Thank you very much for your answer. But I wonder why isn't the matrix 
> block-triangular as we should observe the same interactions between two 
> same states ?
> Also do you have an idea about  the issues I observed while doing SOC 
> calculation using the wB97XD functional posted above? 
> I have not found any solution yet. 
> Thank you very much,
> Have a good day,
> Best regards.
>
> Le mercredi 3 juillet 2024 à 13:45:54 UTC+2, Jan-Robert Vogt a écrit :
>
>> Dear Raphaël,
>>
>> as default, the SOC-Matrix is written in eV. The matrix is not completely 
>> block-triangular, but would not be far away from it.
>>
>> The file contains the couplings between the different states, which were 
>> calculated. The first state is the ground state (numbered as 1), followed 
>> by the singlet states S (numbered as 2 to nstates+1) and the triplet 
>> states T-1 (numbered as nstates+2 to 2 x nstates +1), T0 (numbered as 2 
>> x nstates+2 to 3 x nstates +1)  and T+1 (numbered as 3 x nstates +2 to 4 
>> x nstates +1). Since a restricted closed shell reference is used, the 
>> same wave functions are used for T-1, T0 and T+1. 
>> The coupling states are represented by the first two columns, which 
>> should be integers. Then the complex coupling element is shown, the first 
>> number is the real part, while the second number represents the imaginary 
>> part. At the real part of the diagonal are the excitation energies.
>>
>> Best regards,
>> Jan-Robert Vogt
>> Raphaël Rullan schrieb am Mittwoch, 3. Juli 2024 um 09:32:37 UTC+2:
>>
>>> Dear CP2K users,
>>> I have another question concerning SOC calculations and especially about 
>>> how to read the output. The file *.socme looks like a triangular matrix, is 
>>> that normal ? Also, what is the unit used in this file ? 
>>> Thanks a lot for the answers,
>>> Best regards,
>>> Raphael Rullan
>>>
>>> Le mercredi 12 juin 2024 à 09:50:48 UTC+2, Raphaël Rullan a écrit :
>>>
>>>> Dear CP2K users,
>>>>
>>>> I am trying to do SOC calculations using the wB97XD functional and I am 
>>>> encountering how can I say ... strange behaviors . It is either crashing 
>>>> without apparent reasons in the out file or  not giving any electronic 
>>>> transitions (see ....out). It also seems like it is dependent on the number 
>>>> of excited states I ask for.
>>>> First, It is important to note that those calculations work if I'm 
>>>> using the PBE0 functional.  
>>>>
>>>> I send you the input and output files. We tried to change the 
>>>> diagonalisation libraries from ELPA to SCALAPACK and also changed COSMA for 
>>>> SCALAPACK without much change. 
>>>>
>>>> I have no idea what could be the problem.
>>>> Could you help me please,
>>>> Thanks a lot
>>>> Raphael Rullan
>>>>
>>>>

-- 
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+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/28c79477-1b39-4eb9-862a-f9801dafc520n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240705/2b70a779/attachment.htm>