[QE-users] SCF not converging in QE6.3 (but it does so in QE6.0)

Davide Ceresoli davide.ceresoli at cnr.it
Thu Jan 31 12:09:31 CET 2019


Actually, I'm always using 4 threads/CPU on the KNL CPUs of marconi-A2,
to get the maximum performance. However, I've experienced the same
problems as Guido with QE-6.3. The problems disappear if I use only
1 thread/CPU.

QE-6.1 with 4 threads/CPU is working perfectly on marconi-A2.

HTH

Davide


On 1/31/19 9:17 AM, Pietro Delugas wrote:
> Hello
> 
> I tried 6.3 in CINECA ( both skl and knl ).  The first scf  converges smoothly 
> in 10 steps as you reported for 6.0.
> 
> There could be an issue in your slurm script you are requesting 2*68 cpus, 
> asking for 2*68 MPI tasks and 2 threads per MPI task which means that you are 
> asking for a number of threads that is twice the number of allocated cpu's.
> 
> In SKL (Marconi A3) the submission is refused for this reason, in KNL (Marconi 
> A2 ) the submission is accepted but in my case the program was  choked and 
> didn't go beyond the input reading in 5 minutes.
> 
> hope it helps - Pietro
> 
> 
> On 30/01/19 16:10, Paolo Giannozzi wrote:
>> Hi Guido
>>
>> I have run your job on an old 16-cpu machine and today's version of QE: it 
>> converges without any problem. In addition, my results at the first iteration 
>> are visibly much closer to your "good old" results than to your "bad new" 
>> ones.  It looks like a weird CINECA-specific problem to me. I would try to run 
>> it on a different number of processors, e.g. 6 pools of 10 processors on a 
>> single node, or 3 pools of 18 processors.
>>
>> Paolo
>>
>> On Wed, Jan 30, 2019 at 11:23 AM Guido Fratesi <guido.fratesi at unimi.it 
>> <mailto:guido.fratesi at unimi.it>> wrote:
>>
>>     Dear all,
>>
>>     The problem I'm facing here is the missing SCF convergence with the 6.3
>>     and develop releases, while older 6.0 release convergences smoothly. I
>>     am computing a three-layer Au(111) slab, that I replicated from the
>>     (1x1) relaxation to a (4x3) supercell (and also other sizes). To my
>>     experience this system should not pose specific difficulties.
>>
>>     In detail:
>>
>>
>>     * I'm running (HPC system marconi at cineca, Italy) the precompiled version
>>     of QE6.3, and the SCF loop is not converging, regardless of changing
>>     smearing, mixing mode & parameter, pseudopotential. The input is given
>>     at the end of this message. E.g., "grep accuracy
>>     Au111_4x3_3L_K3x4.rlx.out" gives:
>>           estimated scf accuracy    <       1.89873147 Ry
>>           estimated scf accuracy    <       0.12553445 Ry
>>           estimated scf accuracy    <       0.01337648 Ry
>>           estimated scf accuracy    <       0.00966317 Ry
>>           estimated scf accuracy    <       0.01041836 Ry
>>           estimated scf accuracy    <       0.00627208 Ry
>>           estimated scf accuracy    <       0.01265435 Ry
>>           estimated scf accuracy    <       0.00686049 Ry
>>           estimated scf accuracy    <       0.02148796 Ry
>>           estimated scf accuracy    <       0.00703012 Ry
>>           estimated scf accuracy    <       0.00905125 Ry
>>           estimated scf accuracy    <       0.01597448 Ry
>>           estimated scf accuracy    <       0.00370159 Ry
>>           estimated scf accuracy    <       0.00418866 Ry
>>     and so on without reducing further.
>>
>>
>>     * As a comparison, I run older version of QE (QE6.0 also build by cineca
>>     staff on marconi) with the same input, and convergence is smooth and
>>     quick as expected:
>>
>>     grep accuracy qe6.0.rlx.out
>>           estimated scf accuracy    <       1.93398261 Ry
>>           estimated scf accuracy    <       0.15421538 Ry
>>           estimated scf accuracy    <       0.00644341 Ry
>>           estimated scf accuracy    <       0.00382849 Ry
>>           estimated scf accuracy    <       0.00265796 Ry
>>           estimated scf accuracy    <       0.00056037 Ry
>>           estimated scf accuracy    <       0.00022539 Ry
>>           estimated scf accuracy    <       0.00001359 Ry
>>           estimated scf accuracy    <       0.00000159 Ry
>>           estimated scf accuracy    <       0.00000009 Ry
>>           estimated scf accuracy    <       0.00000415 Ry
>>           estimated scf accuracy    <       0.00000089 Ry
>>           estimated scf accuracy    <       0.00000004 Ry
>>
>>
>>     * I downloaded today the develop version from
>>     https://github.com/QEF/q-e/archive/develop.zip recompiling on the same
>>     machine and still the SCF does not converge.
>>
>>
>>     * Important to say, if I place a molecule on one side of the slab, the
>>     SCF converges (also with QE6.3).
>>
>>
>>     Here is my input (all files are available in this online folder:
>>     https://drive.google.com/drive/folders/1072PbthHTvbqNTCXqHKeAJUyPOwi_2Ez)
>>
>>       &control
>>          calculation = 'relax' ,
>>          restart_mode = 'from_scratch' ,
>>          prefix='Au111_5x4_3L' ,
>>          pseudo_dir = '../0-pseudo.d/' ,
>>          outdir='./tmp/' ,
>>          max_seconds   = 1200
>>          iprint = 1
>>          verbosity = 'high'
>>       /
>>       &system
>>          ibrav=  12 ,
>>          a = 11.8022910005134168d0 ,
>>          b =  8.8517182503850626d0 ,
>>          c = 20.0d0 ,
>>          cosab = -.5 ,
>>          nat =  36 ,
>>          ntyp = 1 ,
>>          ecutwfc = 36 ,
>>          ecutrho = 220 ,
>>          occupations = 'smearing' ,
>>          degauss = 0.02 ,
>>          smearing = 'mv' ,
>>       /
>>       &electrons
>>          mixing_mode = 'local-TF'
>>          conv_thr = 1e-7
>>       /
>>       &ions
>>       /
>>     ATOMIC_SPECIES
>>       Au  0.0 Au.pbe-dn-rrkjus_psl.0.1.UPF
>>     ATOMIC_POSITIONS Angstrom
>>     Au      1.475286375063      0.851756985775     -4.818265124511
>>     Au      4.425859125193      0.851756985775     -4.818265124511
>>     Au      7.376431875324      0.851756985775     -4.818265124511
>>     Au     10.327004625454      0.851756985775     -4.818265124511
>>     Au      0.000000000000      3.407027943102     -4.818265124511
>>     Au      2.950572750130      3.407027943102     -4.818265124511
>>     Au      5.901145500260      3.407027943102     -4.818265124511
>>     Au      8.851718250391      3.407027943102     -4.818265124511
>>     Au     -1.475286375063      5.962298900428     -4.818265124511
>>     Au      1.475286375067      5.962298900428     -4.818265124511
>>     Au      4.425859125197      5.962298900428     -4.818265124511
>>     Au      7.376431875328      5.962298900428     -4.818265124511
>>     Au      0.000000000000      1.703513971551     -2.409132562256
>>     Au      2.950572750130      1.703513971551     -2.409132562256
>>     Au      5.901145500260      1.703513971551     -2.409132562256
>>     Au      8.851718250391      1.703513971551     -2.409132562256
>>     Au     -1.475286375063      4.258784928877     -2.409132562256
>>     Au      1.475286375067      4.258784928877     -2.409132562256
>>     Au      4.425859125197      4.258784928877     -2.409132562256
>>     Au      7.376431875328      4.258784928877     -2.409132562256
>>     Au     -2.950572750126      6.814055886203     -2.409132562256
>>     Au      0.000000000004      6.814055886203     -2.409132562256
>>     Au      2.950572750134      6.814055886203     -2.409132562256
>>     Au      5.901145500265      6.814055886203     -2.409132562256
>>     Au      0.000000000000      0.000000000000      0.013283696000
>>     Au      2.950572750130      0.000000000000      0.013283696000
>>     Au      5.901145500260      0.000000000000      0.013283696000
>>     Au      8.851718250391      0.000000000000      0.013283696000
>>     Au     -1.475286375063      2.555270957326      0.013283696000
>>     Au      1.475286375067      2.555270957326      0.013283696000
>>     Au      4.425859125197      2.555270957326      0.013283696000
>>     Au      7.376431875328      2.555270957326      0.013283696000
>>     Au     -2.950572750126      5.110541914652      0.013283696000
>>     Au      0.000000000004      5.110541914652      0.013283696000
>>     Au      2.950572750134      5.110541914652      0.013283696000
>>     Au      5.901145500265      5.110541914652      0.013283696000
>>     K_POINTS automatic
>>       3 4 1  1 1 0
>>
>>     Thank you for your kind attention,
>>
>>     Guido Fratesi
>>
>>
>>
>>     -- 
>>     Guido Fratesi
>>
>>     Dipartimento di Fisica
>>     Universita` degli Studi di Milano
>>     Via Celoria 16, 20133 Milano, Italy
>>
>>
>>
>>
>>     _______________________________________________
>>     users mailing list
>>     users at lists.quantum-espresso.org <mailto:users at lists.quantum-espresso.org>
>>     https://lists.quantum-espresso.org/mailman/listinfo/users
>>
>>
>>
>> -- 
>> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
>> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
>> Phone +39-0432-558216, fax +39-0432-558222
>>
>>
>> _______________________________________________
>> users mailing list
>> users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users

-- 
+--------------------------------------------------------------+
   Davide Ceresoli
   CNR Institute of Molecular Science and Technology (CNR-ISTM)
   c/o University of Milan, via Golgi 19, 20133 Milan, Italy
   Email: davide.ceresoli at cnr.it
   Phone: +39-02-50314276, +39-347-1001570 (mobile)
   Skype: dceresoli
   Website: http://sites.google.com/site/dceresoli/
+--------------------------------------------------------------+


More information about the users mailing list