[CP2K-user] [CP2K:21087] Re: GW BSE calculation test - CPASSERT mp2 error ?

[Maximilian Graml]

Hi Maxime,

thank you for bringing that up. I tested your input for ethene as geometry, 
where I was able to reproduce the error.
In your input example, no explicit RI basis is given and apparently the 
automatic switch to a default RI basis doesn't work somehow.
I will have a look into that in the next couple of days.

Anyway, you can fix that by stating an explicit RI basis in your 
calculations, e.g. the ones in BASIS_RI_MOLOPT 
<https://github.com/cp2k/cp2k/blob/master/data/BASIS_RI_MOLOPT>would match 
your basis sets.
However, I did not yet test the BSE for the GTH basis sets or pseudos in 
general. Therefore, I always employed the aug-cc-pV..Z basis sets in the 
examples online.
Since these diffuse basis functions can be a bit hard to converge in the 
SCF (at least without tricks like OT or smearing etc), we are currently 
working on new basis sets, as you can find them here 
<https://github.com/JWilhelm/BASIS_AUG_MOLOPT/blob/main/BASIS_ALL_MOLOPT_AUG_for_excited_states>
.
They are not yet thoroughly tested for heavy atoms, but the atoms included 
in Thiel's set (H, C, N, O) should be fine. For now, we recommend 
<https://github.com/JWilhelm/BASIS_AUG_MOLOPT/blob/main/BASIS_ALL_MOLOPT_AUG_for_excited_states#L23>to 
employ the cc-RI basis sets for the RI basis.

In your provided input, there is also a PERIODIC XYZ remaining in the 
POISSON section, which should be replaced by a NONE to match the 
periodicity from the geometry.


I hope this helps to start your calculations.
Best,
Max
Maxime Hodée schrieb am Dienstag, 28. Januar 2025 um 11:03:56 UTC+1:

> After some testing, I still have trouble when using a supercell of my 
> system rather than a fully periodic system.
> After a normal startup and scf cycle, the calculation suddenly stops 
> without any error, right after stating the amount of memory needed.
> I first thought it was because not enough memory was available on the 
> machines I am using, so I reduced the size of the supercell and the basis 
> set, however, the same issue happened (and I am also pretty sure that I 
> have enough computational capacities for that).
> I am attaching my inputs and outputs here, I am not sure what is going 
> wrong here.
>
> Thanks in advance for your help,
> best regards,
> Maxime
> Le vendredi 24 janvier 2025 à 10:47:10 UTC+1, Maxime Hodée a écrit :
>
>> Alright thanks ! Yes I'll try using a cluster rather than a fully 
>> periodic calculation, should be sufficient for what I intend to do.
>> I'll try all of those in the following weeks thank you !
>>
>> Maxime
>>
>> Le jeudi 23 janvier 2025 à 11:18:32 UTC+1, Maximilian Graml a écrit :
>>
>>> Dear Maxime,
>>>
>>> thank you for your continued interest and my apologies for the 
>>> inconveniences!
>>> It seems that there are several separate issues here.
>>> Most of them occur, because the BSE module is under active development 
>>> and therefore a lot has changed since v2024.3.
>>> The unrecognised keywords have been introduced after 2024.3, e.g. 
>>> ETA_LIST in 
>>> https://github.com/cp2k/cp2k/pull/3628/files#diff-1e4cc8ae95bc574eac18fca486f3292754fedf7c6664c401be22af91e2bf9554R1430 
>>> and NUM_PRINT_EXC_DESCR in 
>>> https://github.com/cp2k/cp2k/pull/3781/files#diff-1e4cc8ae95bc574eac18fca486f3292754fedf7c6664c401be22af91e2bf9554R1390
>>> .
>>>
>>> In case you dont need these features, you can just comment them out in 
>>> the Input, as you did. Otherwise, v2025.1 is released by now and includes 
>>> these features of the exciton descriptors and Natural Transition orbitals.
>>>
>>> CPASSERT is quite a general routine and unfortunately the CPASSERT, 
>>> which causes your calculation to abort does not yet have a meaningful error 
>>> message in v2024.3. I fixed that in Commit 
>>> https://github.com/cp2k/cp2k/blob/master/src/mp2_integrals.F#L339, i.e. 
>>> in v2025.1, the error message would be included.
>>> The issue here is that you are invoking a periodic calculation, which is 
>>> not yet supported by the BSE module (but we plan to implement that). 
>>>
>>> Is it maybe possible to use a workaround here, e.g. by using a larger 
>>> supercell of your system similar to the long nanographenes in 
>>> https://github.com/cp2k/cp2k-examples/blob/master/bethe-salpeter/Nanographene/BSE.inp 
>>> ?
>>>
>>> I hope that helps. :) Please let me know if you have further questions!
>>> Best,
>>> Max
>>> Maxime Hodée schrieb am Dienstag, 21. Januar 2025 um 16:28:35 UTC+1:
>>>
>>>> Dear CP2K community,
>>>> I am currently trying to perform GW BSE calculations using CP2K.
>>>>
>>>> I am using the 2024.3 version of CP2K (as suggested in a previous post 
>>>> I made), however, I am running into some issues I cannot manage to solve by 
>>>> myself.
>>>>
>>>> I based my input on the one from there:
>>>>
>>>> https://github.com/cp2k/cp2k-examples/blob/master/bethe-salpeter/H2/cp2k.out#L28
>>>> However, some keywords were weirdly unrecognised and are commented in 
>>>> the provided input (NUM_PRINT_EXC_DESCR, ETA_LIST and the NTO_ANALYSIS 
>>>> keywords namely).
>>>>
>>>> Then, when running the calculation without those keywords I obtain this 
>>>> "CPASSERT failed" error which seem to be fairly common to users. So I tried 
>>>> several solutions that I found, none of them worked (change of basis set, 
>>>> change of the EPS_DEFAULT values, and other).
>>>> You'll find the input and output files attached, and I would be glad to 
>>>> know if someone has an idea (I'm also eager to understand how this code 
>>>> works so thanks in advance for your explanations !)
>>>>
>>>> Best,
>>>> Maxime
>>>>
>>>

-- 
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 visit https://groups.google.com/d/msgid/cp2k/1bb3a7ac-7c3f-45af-9d77-499c4f0f875cn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20250128/8704b3c5/attachment.htm>