Abinit stops with no error during geometric optimization

Dear all,

I am experiencing the following issue in all the abinit versions 9.x.y . During a geometric optimization, after a certain number of geometric steps, abinit exits after the SCF converged and before doing the geometric move. The number of geometric steps after which this happens is random to me; for the same system, such number is different for every run. This happens to me regardless the kind of system and setup (ecut, k-mesh etc). I never experienced this with version 8.10.3 and lower. If more info is needed to find the source of the problem, please, let me know.

Thanks a lot in advance

Antonio

Hi @antonio, thanks for sharing your problem. Of course it would be helpful for us to help you if you could send us your input file, log file and any stderr file if relevant. Your pseudos could also be helpful.

Cheers,

Félix

Hi,
I am attaching a tar.gz archive with the relevant files; I changed the extension “gz” into “in” because the web page didn’t allow me to upload it.
The files are:
ab.in - abinit input
ab.abo - abinit output
out - standard output
*psp8 - pseudopotentials
__ ABI_MPIABORTFILE__ - abinit mpi abort file

ab.tar.in (424.9 KB)

Best
Antonio

Hi @antonio ,
For some reason I can’t open the file you attached. I think I might be able to help if you provide me with the input and the output files. You can just copy and paste in the chat.

Best.

Hi @Carl_D ,
the web page says that I exceeded the number of characters, I can’t copy here the whole output. If you use a graphical interface, it is enough to rename the archive as ab.tar.gz, or if you use linux, the command “tar xzvf ab.tar.in” in a bash shell should work. Here I copy the input in the case this solution doesn’t work, I hope it is enough to understand what is the source of the problem. Or I can send you the files via email. Thanks a lot for your help.
Best
Antonio


ab.in

#Definition of lattice parameters
#********************************
  acell      3*1.0 Angstrom
  rprim 
7.15071608882484 -1.0447381503306E-06 -2.02806185102241E-07
-1.04474053828105E-06 7.15070965106656 -6.00802461319068E-07
-2.02800949942834E-07 -6.00804956678975E-07 7.15069836435166
 
#Definition of atomic positions and types
#*************************************
  chksymbreak  0
  chkprim 0
  natom       40
  ntypat      3
#             Ca Si O
  znucl       20 14 8
  typat       8*1 8*2 24*3
  xred        
  0.2878173690030232  0.2743907855214384  0.2785533885156544
  0.7880481340769769  0.2755514776085617  0.2793743840743456
  0.2880481340769767  0.7755514776085618  0.2793743840743456
  0.7878173690030236  0.7743907855214384  0.2785533885156544
  0.2880481340769767  0.2755514776085617  0.7793743840743454
  0.7878173690030235  0.2743907855214384  0.7785533885156543
  0.2878173690030233  0.7743907855214383  0.7785533885156545
  0.7880481340769768  0.7755514776085618  0.7793743840743454
  0.0379863362666620  0.0249969452036381  0.0289950476230066
  0.5379872231943380  0.0249950295623619  0.0289942880769934
  0.0379872231943380  0.5249950295623620  0.0289942880769934
  0.5379863362666620  0.5249969452036382  0.0289950476230066
  0.0379872231943380  0.0249950295623619  0.5289942880769934
  0.5379863362666620  0.0249969452036381  0.5289950476230066
  0.0379863362666620  0.5249969452036382  0.5289950476230066
  0.5379872231943380  0.5249950295623619  0.5289942880769934
  0.0803089518628395  0.2750060876583085  0.9878390608187473
  0.4957583531691606  0.2750104012916914  0.0701919503982527
  0.9957583531691605  0.7750104012916913  0.0701919503982527
  0.5803089518628396  0.7750060876583086  0.9878390608187473
  0.9957583531691605  0.2750104012916914  0.5701919503982527
  0.5803089518628396  0.2750060876583086  0.4878390608187473
  0.0803089518628395  0.7750060876583084  0.4878390608187472
  0.4957583531691606  0.7750104012916913  0.5701919503982527
  0.0813146710270418  0.0667401055813446  0.2790126253915051
  0.4947507888039583  0.9832845014396554  0.2790079976984949
  0.9947507888039582  0.4832845014396553  0.2790079976984949
  0.5813146710270418  0.5667401055813447  0.2790126253915051
  0.9947507888039582  0.9832845014396554  0.7790079976984949
  0.5813146710270419  0.0667401055813446  0.7790126253915050
  0.0813146710270418  0.5667401055813446  0.7790126253915052
  0.4947507888039583  0.4832845014396553  0.7790079976984949
  0.2880144681011212  0.9853913697377806  0.9890094740437325
  0.7880137044888790  0.0646332963922193  0.0690217833532675
  0.2880137044888789  0.5646332963922193  0.0690217833532675
  0.7880144681011212  0.4853913697377806  0.9890094740437325
  0.2880137044888789  0.0646332963922193  0.5690217833532676
  0.7880144681011213  0.9853913697377806  0.4890094740437325
  0.2880144681011213  0.4853913697377806  0.4890094740437325
  0.7880137044888790  0.5646332963922194  0.5690217833532676
 
  ixc         2 # LDA PZ, CA
  pseudos     "ca.psp8, si.psp8, o.psp8"

#Parameters of the SCF cycles
#****************************
   iscf        17
   nstep       100
   toldfe      1.0d-11

#Plane wave basis and k-point grid
#*********************************
   ecut         500 eV
   ecutsm       0.5
   dilatmx      1.1
                
   ngkpt        3 3 3
   nshiftk      1
   shiftk       0.0 0.0 0.0

#   autoparal 1
#   max_ncpus 128
   paral_kgb 1

   npkpt     14
   npfft     9
   npband    1
   bandpp    1 

   kptopt      1
   prtposcar   1
   prtwf       0

ndtset 1 jdtset 2 

#DATASET2: full geometric optimization
#********************************************
  optcell2     2
  ionmov2      22
  ntime2       300
  tolmxf2      1.d-8

The error message reported in __ABI_MPIABORTFILE__ reads:

--- !Ionmov22Error
src_file: m_pred_bfgs.F90
src_line: 884
mpi_rank: 120
message: |

Very likely the MPI process was killed by the resource manager and it didn’t have enough “time” to output the entire message to file.

The full message reads:

  write (ionmov22_errmsg, '(a,i0,3a)') &
    'Lbfgs routine failed. Returned value: ', info,ch10, &
    'Restart your calculation from last step or try a different ionmov'

In other words, the LBFGS algorithm corresponding to ionmov 22 failed to predict
the new configuration and Abinit stops.
This is not the first time I see this kind of error message but we still don’t know if this error is triggered by some kind of numerical instability or by unphysical input values.

If I look at your output files, it seems that both forces and stresses are decreasing as expected so it’s rather disappointing that ionmov 22 suddenly fails.
For the time being, I would suggest to use ionmov 2.

BTW:
ecut = 18 Ha seems too low to me as the pseudojo Oxygen should require ~ 46 Hartree

I tested ionmov 2 and it seems that works in this case; I had to stop the job before the convergence was reached. However, in other cases I tried the same (ionmov 2), and it didn’t work; most of the time, I obtained the geometric convergence by switching to the FIRE algorithm which does not suffer of this problem.
So, from what I understand, it is not clear why abinit stops in these cases. OK, I’ll try to get the convergence case by case by switching among the different algorithms or by distorting a bit the last geometry configuration and restart the calculation.

As a final comment, I used that ecut to speed up the run for this example. Thanks for the remark.

Thanks a lot for your prompt response!

Best
Antonio