[Pw_forum] hybrid calculation stopping after first scf convergence

Paolo Giannozzi p.giannozzi at gmail.com
Wed Jan 6 22:10:17 CET 2016


A hybrid calculation with k-points requires, in addition to all other terms
for ordinary DFT, the computation of the exact-exchange term costing
O(Nk*Nq*Nb*Nb*Nw) floating-point operations, plus the storage of Kohn-Sham
states, filling Nq*Nb*Nr double precision complex numbers. Legend:
Nk = number of k-points in the Irreducible Brillouin Zone (301 in your case)
Nq = number of k+q-points (1728)
Nb = number of Kohn-Sham states (53)
Nw = number of plane waves (20005)
Nr = number of points in the FFT grid = nr1*nr2*nr3 = 72*72*72
Arrays and computations are distributed so divide Nw and Nr by the number
of processors. If I made the math correctly, it's 2.65 Gb RAM per MPI
process on 192 processors. I am not surprised it crashes. Even if it
doesn't, it will take approximately forever. Are you sure you need that
many k- and q-points?

Paolo

On Wed, Jan 6, 2016 at 9:32 PM, Dumett Torres, Daniel <dumettt2 at illinois.edu
> wrote:

> I have been trying to run a hybrid PBE0 calculation in parallel and have
> run into a recurring problem in which the calculation stops after scf
> convergence is first acheived - the calculation stops and doesn't go on to
> refine the calculation with exx. I have tried to run the calculation with
> between 24 and 192 processors because we previously  thought memory
> requirements were to blame for the error, but this didn't seem to help.
>
> *''**TACC: Setting up parallel environment for MVAPICH2+mpispawn.*
> *TACC: Starting parallel tasks...*
>
> *     Program PWSCF v.5.2.0 *'
>
>
> *'Parallel version (MPI), running on   192 processors*
> * R & G space division:  proc/nbgrp/npool/nimage =     192*'
>
>
> *'highest occupied, lowest unoccupied level (ev):     6.3409    6.0501*
>
> *convergence has been achieved in  16 iterations*
>
> *TACC: MPI job exited with code: 1*
>
> *TACC: Shutdown complete. Exiting.'*
>
>
> this error produces a CRASH file that reports an error reading an xml-data
> file and upon inspection,the .save folder produced contains the
> charge-density.dat file but is missing the .UPF files, the data-file.xml
> file, and data-file.xml.eig file. Furthermore, while the .igk files have
> content, all the .wfc files produced by the job are empty (0 bytes). My
> impression is that this missing content is what is keeping the calculation
> from being able to proceed but I don't know how it might be addressed.
>
> attached is the input file for my calculation.  Attached are the input and
> output files resulting from one such run. I did notice an additional
> message in the output
>
> *'     Message from routine sym_rho_init:*
> *     likely internal error: no G-vectors found'*
>
> but the calculation seems to proceed despite it...
>
> I appreciate any insights you might have regarding these complications I
> am facing
>
> Thank you for your help,
> Daniel Dumett Torres
> Graduate Student at the University of Illinois at Urbana-Champaign
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>



-- 
Paolo Giannozzi, Dip. Matematica Informatica e Fisica,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20160106/2e62b902/attachment.html>


More information about the users mailing list