[CP2K-user] [CP2K:21487] Re: frequencies

Frederick Stein f.stein at hzdr.de
Tue May 20 09:44:22 UTC 2025


Dear Sihem,
Apparently, your input file arrived after I started writing my message. I 
suggest you to optimize the cell seperately as stress tensor calculations 
are very costly and not too accurate as with a modulating cell sizes, 
integration grids will also change. If you know that your cell has a fixed 
shape with a single parameter (cubic, hexagonal cell with 2D peridocity, 
...), you should rather run several calculations while varying the cell 
parameter and fitting a Burch-Murnaghan equation of states. If a single 
SCF+force calculation still takes too much, you should switch to another 
computer if possible or wait for a very long time. Generally, you should 
optimize all accuracy and performance relevant parameters with this 
SCF+force calculation.
HTH,
Frederick

Frederick Stein schrieb am Dienstag, 20. Mai 2025 um 11:33:15 UTC+2:

> Dear Sihem,
> Without any further information (input and output file), it is hard to 
> give any recommendations. Optimize the force calculation by making the 
> calculation as accurate as necessary and as 
> cheap as possible (CUTOFF, EPS_DEFAULT, EPS_SCF, play around with mixing 
> schemes, ...).
> Regards,
> Frederick
>
> Sihem AZIZI schrieb am Dienstag, 20. Mai 2025 um 11:30:12 UTC+2:
>
>> Dear frederick,
>> Thanks a lot for your response.
>> I have another question for you. On another machine I ran an 
>> optimization calculation for a Ru slab (3 layers) with an H2 molecule. And 
>> the calculation has been running for a month. What should I do to make it 
>> converge faster?
>> Thank you again for your help.
>> Sihem
>>
>> On Monday, May 19, 2025 at 9:05:04 PM UTC+1 Frederick Stein wrote:
>>
>>> Dear Sihem,
>>> How much time do you need for a force calculation on the same machine 
>>> with 8 processes? Your total runtime will be 6*<number of atoms> in your 
>>> system as the elements of the Hessian will be determined from symmetric 
>>> finite difference quotients of the analytical gradients. Please be aware 
>>> that a vibrational analysis requires a very accurate setup especially with 
>>> your required displacement. Your EPS_SCF is definitely too low for that, 
>>> your CUTOFF and the EPS_DEFAULT in the QS section could be also improved. 
>>> The general recommendation is a value 1.0E-7 to 1.0E-8 for EPS_SCF. You 
>>> need analytical gradients with an accuracy of several digits as too noisy 
>>> gradients will render the frequencies meaningless. The expected differences 
>>> in gradients should be at least one or two orders of magnitude higher than 
>>> the accuracy of the gradients.
>>> Best,
>>> Frederick
>>>
>>> Sihem AZIZI schrieb am Montag, 19. Mai 2025 um 20:46:26 UTC+2:
>>>
>>>> Dear all,
>>>>
>>>> I've been running a frequency calculation for over a month and it's 
>>>> still running. I don't know what's wrong. Can someone help me?
>>>> Best regards,
>>>> Sihem
>>>>
>>>

-- 
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/f943dd80-fbe5-424f-a523-9e84ffaa853fn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20250520/d0a13a5d/attachment.htm>


More information about the CP2K-user mailing list