[Pw_forum] "scf" calculation with "PBE0" hybrid functional

dario rocca roccad at gmail.com
Wed Aug 10 01:40:53 CEST 2016


Dear Narendra,
While a PBE calculation requires a single FFT for the density the PBE0
calculation requires N_bands*N_bands FFTs for the co-density. In your case
if nspin=1 you would have about 47000 FFTs for a single EXX evaluation.
Despite the number of processors, the number of bands might be the
bottleneck of your calculation.
I don't know if an efficient band parallelization is available for the EXX
in QE but it could help.
Best,
Dario


On Mon, Aug 8, 2016 at 1:39 PM, Narendranath Ghosh <ghosh.naren13 at gmail.com>
wrote:

> Dear Dario
>                 I could not complete my last PBE0 "scf" job containing 434
> atoms (CNT+Fullerene). The job could not stop but still running without
> writing any more in the output file. These lines are written in  the output
>  34 hours ago .I have tried it twice but I found same thing.
>
>      EXX: now go back to refine exchange calculation
>
>      total cpu time spent up to now is    42069.2 secs
>
>      Self-consistent Calculation
>
>      iteration #  1     ecut=    30.00 Ry     beta=0.45
>      Davidson diagonalization with overlap
>
> I could not understand what's happening.I have used 24 node with 384
> processors.I have alter ecut, cell length, beta values nothing happened.
>
> Please suggest any idea.......
>
>
>
> On Sun, Aug 7, 2016 at 10:53 AM, dario rocca <roccad at gmail.com> wrote:
>
>> I will let the MD experts to reply to this question :)
>>
>> On Sat, Aug 6, 2016 at 2:49 AM, Narendranath Ghosh <
>> ghosh.naren13 at gmail.com> wrote:
>>
>>> Dear Dario
>>>
>>> Thank you very much for your valuable suggestion. As my systems are
>>> quite large and require high computational resources , I decided to compare
>>> one of my result with that of PBE0.
>>>
>>> One problem that I am facing now "after successfully optimized the
>>> (CNT+Fullerene) systems now I am looking for  to get 5 picosecond "md"
>>> trajectory".
>>> In QE which flags  allowed one to get *MD trajectories * *with NVE
>>> ensemble. *I find some idea from http://qe-forge.org/piper
>>> mail/pw_forum/2009-June/087573.html. But that is not clear to me.
>>>
>>> Please suggest any idea.....
>>>
>>> Thank you again
>>>
>>> *Narendra Nath Ghosh*
>>>
>>> *Research Associate*
>>>
>>> *University of Gour Banga*
>>>
>>> *Department of Computational Physics *
>>>
>>> *Malda-732102*
>>>
>>> *India*
>>>
>>> On Sat, Aug 6, 2016 at 10:48 AM, dario rocca <roccad at gmail.com> wrote:
>>>
>>>> Dear Narendra,
>>>> If you do a PBE calculation with norm-conserving, ultrasoft, and PAW
>>>> pseudopotentials very likely you will get very similar results (but it
>>>> depends
>>>> also on the quality of the pseudopotential, the property under
>>>> consideration, etc.).
>>>> In the case of PBE0 the type of pseudopotential will have a stronger
>>>> effect on the final result.
>>>> For purely esthetic reasons, if your previous calculations didn't take
>>>> too long you could redo them with norm-conserving pseudopotentials and
>>>> confirm that the results are indeed the same.
>>>> Otherwise I don't see anything wrong in comparing, for example, a gap
>>>> obtained at PBE level with US pseudos and a gap obtained at PBE0
>>>> level with NC pseudos.
>>>> Best,
>>>> Dario
>>>>
>>>>
>>>>
>>>> On Fri, Aug 5, 2016 at 3:44 PM, Narendranath Ghosh <
>>>> ghosh.naren13 at gmail.com> wrote:
>>>>
>>>>> Dear Dario
>>>>>
>>>>> Thank you very much  for your support.It works well and some of the
>>>>> jobs have been finished successfully.
>>>>>
>>>>> But one thing in my mind. In my whole calculations , I used PBE with
>>>>> ultrasoft pseudopotentials. But to do "scf" calculations with  PBE0 "hybrid"
>>>>> functional I have to use "Norm conserving pseudopotential". So how
>>>>> can I compare these two results.
>>>>>
>>>>> Please suggest any idea.....
>>>>>
>>>>> Thank you again
>>>>>
>>>>> *Narendra Nath Ghosh*
>>>>>
>>>>> *Research Associate*
>>>>>
>>>>> *University of Gour Banga*
>>>>>
>>>>> *Department of Computational Physics *
>>>>>
>>>>> *Malda-732102*
>>>>>
>>>>> *India*
>>>>>
>>>>>
>>>>> On Thu, Aug 4, 2016 at 2:17 AM, dario rocca <roccad at gmail.com> wrote:
>>>>>
>>>>>> Dear Narendranath,
>>>>>> Your calculation has reached convergence if you read the message "EXX
>>>>>> self-consistency reached".
>>>>>> You can learn some details about the scf procedure with hybrid
>>>>>> functionals in PW/examples/EXX_example.
>>>>>> PBE0 and hybrid functional calculations are in general very
>>>>>> expensive. I would suggest you run an example to see what to expect.
>>>>>> You might also try to perform a calculation on your system with
>>>>>> minimal computational parameters (very small cut-off, few/one k-points)
>>>>>> just to see that your run can indeed terminate smoothly. Then I would look
>>>>>> for a set of parameters which are a good compromise between computational
>>>>>> time and the accuracy you need.
>>>>>> You might also try to modify the parameter ecutfock to gain some
>>>>>> speed.
>>>>>> For sure a hybrid functional calculation on a CNT+Fullerene system
>>>>>> might be rather challenging.
>>>>>> Best,
>>>>>> Dario
>>>>>>
>>>>>> On Wed, Aug 3, 2016 at 11:16 AM, Narendranath Ghosh <
>>>>>> ghosh.naren13 at gmail.com> wrote:
>>>>>>
>>>>>>> Dear all
>>>>>>>
>>>>>>>                    After optimizing a system "CNT+Fullerene" with
>>>>>>> "PBE" I am trying to calculate "scf" calculation with "PBE0" hybrid
>>>>>>> functional using  NC Pseudopotential.
>>>>>>>
>>>>>>> In output file I found "*convergence has been achieved in  15
>>>>>>> iterations" *But the job was not finished even after more than one
>>>>>>> week. I could not find any mistake in my input.
>>>>>>>
>>>>>>>
>>>>>>> Please suggest any idea.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Pw_forum mailing list
>>>>>>> Pw_forum at pwscf.org
>>>>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Pw_forum mailing list
>>>>>> Pw_forum at pwscf.org
>>>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> Pw_forum mailing list
>>>>> Pw_forum at pwscf.org
>>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Pw_forum mailing list
>>>> Pw_forum at pwscf.org
>>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>>
>>>
>>>
>>> _______________________________________________
>>> Pw_forum mailing list
>>> Pw_forum at pwscf.org
>>> http://pwscf.org/mailman/listinfo/pw_forum
>>>
>>
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://pwscf.org/mailman/listinfo/pw_forum
>>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20160809/967f9534/attachment.html>


More information about the users mailing list