[QE-users] query on number of k points in nscf

[Giovanni Cantele]

Dear Lorenzo,
thanks you for your prompt response. This is actually the case:
     Parallel version (MPI & OpenMP), running on    2400 processor cores
     Number of MPI processes:              2400
     Threads/MPI process:                     1

     MPI processes distributed on   100 nodes
     K-points division:     npool     =       2
     R & G space division:  proc/nbgrp/npool/nimage =    1200
     wavefunctions fft division:  Y-proc x Z-proc =       2    600
     wavefunctions fft division:  task group distribution
                                  #TG    x Z-proc =       2    600


So, it is explained also why the time used to compute 10 k-points coincides with the whole calculation
computational time.

Maybe, for the sake of clarity, it could be really explanatory if in the output in would be explicitly 
mentioned that only the output from pool #1 is give.

Thank you again,

    Giovanni



> On 8 Apr 2019, at 12:37, Lorenzo Paulatto <paulatz at gmail.com> wrote:
> 
> Dear Giovanni,
> if you are using pools, only a fraction of the k-points will be enumerated in the NSCF calculation.
> kind regards
> 
> On 08/04/2019 12:07, Giovanni Cantele wrote:
>> Dear all,
>> I’ve experienced a “strange” (?) behaviour in an NSCF calculation. I’m using Quantum-ESPRESSO v. 6.4 on CINECA-Marconi.
>> The calculation is with no symmetry ("No symmetry found”) and lists 19 k-points at the beginning of the calculation.
>> In the input the k-point specification is obtained through
>> K_POINTS { automatic }
>> 36  1  1   0 0 0
>> If I
>> grep "k =.*bands " NSCF.out | wc
>> 19 lines are correctly found in the output.
>> However
>> grep "Computing" NSCF.out
>> returns
>>      Computing kpt #:     1
>>      Computing kpt #:     2
>>      Computing kpt #:     3
>>      Computing kpt #:     4
>>      Computing kpt #:     5
>>      Computing kpt #:     6
>>      Computing kpt #:     7
>>      Computing kpt #:     8
>>      Computing kpt #:
>>      Computing kpt #:    10
>> as if only 10 k-points were considered in the calculation. Even more strangely, by analizing the cpu-time information, I get
>>      Computing kpt #:     1
>>      c_bands: 56 eigenvalues not converged
>>      total cpu time spent up to now is    13728.9 secs
>> [….]
>>      Computing kpt #:    10
>>      c_bands: 48 eigenvalues not converged
>>      total cpu time spent up to now is   133710.2 secs
>> that is, the time for 10 k-points, is about 10 times that for a single k-point. However, the full calculation takes
>>      PWSCF        :   1d13h59m CPU   1d15h18m WALL
>> about 1d and 13 hours, that is 37 hours that correspond to about 133000 seconds. So it seems that the whole calculation
>> takes the same time as taken for the first 10 k-points.
>> So, the questions are:
>> 1) why only 10 “Computing ….” lines are printed in output?
>> 2) How can I be sure that the code effectively computed the k-points from 11 to 19 (and if it did not do it, what are
>> the eigenvalues printed at the and of the output for those k-points)?
>> Maybe I’m missing something very trivial (in the case I apologise for that!), but I cannot figure out what!
>> I also checked PW/src/c_bands.f90 and it seems that the line
>>  IF ( iverbosity > 0 ) WRITE( stdout, 9001 ) ik
>> is within the loop
>> k_loop: DO ik = ik_+1, nks
>> so the printing of the message with format
>> 9001 FORMAT(/'     Computing kpt #: ',I5 )
>> should occur for either all k-points or none of them.
>> I thank you in advance for any suggestion.
>> Giovanni
>> -- 
>> Giovanni Cantele, PhD
>> CNR-SPIN
>> c/o Dipartimento di Fisica
>> Universita' di Napoli "Federico II"
>> Complesso Universitario M. S. Angelo - Ed. 6
>> Via Cintia, I-80126, Napoli, Italy
>> e-mail: giovanni.cantele at spin.cnr.it <mailto:giovanni.cantele at spin.cnr.it> <mailto:giovanni.cantele at spin.cnr.it <mailto:giovanni.cantele at spin.cnr.it>>
>> gcantele at gmail.com <mailto:gcantele at gmail.com> <mailto:gcantele at gmail.com <mailto:gcantele at gmail.com>>
>> Phone: +39 081 676910
>> Skype contact: giocan74
>> Web page:https://sites.google.com/view/giovanni-cantele <https://sites.google.com/view/giovanni-cantele>
>> _______________________________________________
>> users mailing list
>> users at lists.quantum-espresso.org <mailto:users at lists.quantum-espresso.org>
>> https://lists.quantum-espresso.org/mailman/listinfo/users <https://lists.quantum-espresso.org/mailman/listinfo/users>
> 
> -- 
> Lorenzo Paulatto - Paris
> _______________________________________________
> users mailing list
> users at lists.quantum-espresso.org <mailto:users at lists.quantum-espresso.org>
> https://lists.quantum-espresso.org/mailman/listinfo/users <https://lists.quantum-espresso.org/mailman/listinfo/users>
-- 

Giovanni Cantele, PhD

CNR-SPIN
c/o Dipartimento di Fisica
Universita' di Napoli "Federico II"
Complesso Universitario M. S. Angelo - Ed. 6
Via Cintia, I-80126, Napoli, Italy

e-mail: giovanni.cantele at spin.cnr.it
            gcantele at gmail.com
Phone: +39 081 676910
Skype contact: giocan74
Web page: https://sites.google.com/view/giovanni-cantele

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190408/64d892ea/attachment.html>