[CP2K-user] [CP2K:11930] Selecting cutoff from the Energy and NGRID data

[Patrick Gono]

Dear Ramanish,

In that particular case, the difference between 700 Ry and 750 Ry is on the
order of 1e-6: (725.7828004716 - 725.7827990986 = 0.000001373). That's 1e-5
in eV. I think it's safe to say we are converged.

The convergence might be easier to establish if you were to plot the
calculated values.

Yours sincerely,
Patrick Gono

On Sat, 29 Jun 2019 at 17:01, Ramanish Singh <sing... at umn.edu> wrote:

> Dear Patrick,
>
> Thank you very much for your suggestion.
> I still have a doubt. I think the peculiar behavior is also seen for
> REL_CUTOFF=60. The convergence is 1e-7 for CUTOFF=550 but after that it
> suddenly decreases to 1e-4 and 1e-3 at CUTOFF=750. So shall I go with 500
> or with 800 for CUTOFF.
>
> Thanks,
> Ramanish Singh
>
> On Saturday, June 29, 2019 at 8:53:20 AM UTC-5, Patrick Gono wrote:
>>
>> Dear Ramanish,
>>
>> Looking at the rest of your data, it appears that the problem was a value
>> of the relative cutoff that was too small. For example, if you consider the
>> convergence of energy with respect to the total cutoff for REL_CUTOFF = 60,
>> the suspicious behaviour vanishes and even small values of CUTOFF yield
>> converged energies.
>>
>> I would suggest you stick with REL_CUTOFF = 50 or 60. These are also the
>> values suggested in the input manual to get accurate results. For the
>> absolute cutoff, you are better off using higher values than the default
>> one as well. Some even suggest using values as high as 800 or 1000 Ry, if
>> accuracy is of the highest importance. The correct choice of the value will
>> also depend on the system (e.g. its size, or the elements present). In any
>> case, from your tests it appears that smaller values of CUTOFF are
>> sufficient as long as you use REL_CUTOFF = 50 or higher, though.
>>
>> What I personally do is start with a reasonably high value for CUTOFF
>> (e.g. 700 or 800 Ry). If I find that a few extra percent of performance are
>> critical for my calculations, I might test a smaller value for the CUTOFF
>> by looking at the quantity of my interest. For example, if I were to study
>> binding energies of small molecules on surfaces, I would study the effect
>> of a smaller value of CUTOFF on that exact quantity. If I find that the
>> effect is negligible, I might use a smaller cutoff (e.g. 500, or even 350
>> Ry).
>>
>> However, from my experience, other parameters, such as the exchange and
>> correlation functional used, or the basis sets, have a much stronger effect
>> on the computational resources required.
>>
>> Hope this helps.
>> Yours sincerely,
>> Patrick Gono
>>
>>
>>
>> On Thu, 27 Jun 2019 at 16:03, Ramanish Singh <si... at umn.edu> wrote:
>>
>>> Hi,
>>>
>>> I am simulating a system of 64 Iodine atoms. I performed the cutoff test
>>> to find the suitable cutoff and reliable_cutoff but the data I got is
>>> somewhat weird. I have attached the data with this post.
>>> For example, for REL_CUTOFF=30 I got the following data:
>>>
>>> # Number of Grids: 4
>>>
>>> # Relative Cutoff (Ry): 30
>>>
>>> # Cutoff (Ry) | Total Energy (Ha) | NG on grid 1 | NG on grid 2 | NG on
>>> grid 3 | NG on grid 4
>>>
>>>     200.00  -725.7833908007       0   15624   15391    4590
>>>
>>>     250.00  -725.7860435954       0   12298   13049   10258
>>>
>>>     300.00  -725.7849325650       0    7090   18177   10338
>>>
>>>     350.00  -725.7837028953       0    3652   19891   12062
>>>
>>>     400.00  -725.7839064559       0    2188   17023   16394
>>>
>>>     450.00  -725.7853932286       0    1619   17592   16394
>>>
>>>     500.00  -725.8084189293       0     527   16090   18988
>>>
>>>     550.00  -725.7834120461       0       0   16617   18988
>>>
>>>     600.00  -725.7833214512       0       0   15624   19981
>>>
>>>     650.00  -725.7826948381       0       0   15624   19981
>>>
>>>     700.00  -725.7826948381       0       0   15544   20061
>>>
>>>     750.00  -725.7844897119       0       0   12298   23307
>>>
>>>     800.00  -725.7829016340       0       0   12218   23387
>>>
>>> The energy is exactly the sane for CUTOFF = 650 and 700 Ry , but it
>>> changes for CUTOFF=750 and 800 ( with error more than 1e-3).
>>> Similar thing happens for other values of REL_CUTOFF as well. The error
>>> decreases to 1e-7 as CUTOFF increases but suddenly jumps to 1e-3 when
>>> CUTOFF is increased beyond 600 Ry.
>>> Do I need to increase my CUTOFF beyond 800 to see if it converges? And
>>> is it worth it to keep it that high keeping in mind the increase in the
>>> simulation time?
>>>
>>> Thanks,
>>> Ramanish
>>>
>>> --
>>> 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 c... at googlegroups.com.
>>> To post to this group, send email to c... at googlegroups.com.
>>> Visit this group at https://groups.google.com/group/cp2k.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/cp2k/f3462318-02c6-4f95-b97b-97a567821932%40googlegroups.com
>>> <https://groups.google.com/d/msgid/cp2k/f3462318-02c6-4f95-b97b-97a567821932%40googlegroups.com?utm_medium=email&utm_source=footer>
>>> .
>>> For more options, visit https://groups.google.com/d/optout.
>>>
>> --
> 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 cp... at googlegroups.com.
> To post to this group, send email to cp... at googlegroups.com.
> Visit this group at https://groups.google.com/group/cp2k.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/faa64132-3ff8-44a0-b72d-afd4fc71bdf3%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/faa64132-3ff8-44a0-b72d-afd4fc71bdf3%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
> For more options, visit https://groups.google.com/d/optout.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20190630/ef07d1c5/attachment.htm>