[CP2K-user] [CP2K:22426] Large time difference between NEB images

[Sofia Scozziero]

Hi everyone, 

I've recently started working with CP2K and I'm running into the following 
issue. When doing NEB calculations some images take significantly more time 
per OT step than others, I'm not talking about number of SCF steps to 
converge but wall time per OT step, which is up to 15x slower for some of 
the images (tipically I've found issues with maybe 2-3 images out of 12 per 
run). Now a 2x-3x difference is manageable but when it gets past the 5x its 
problematic. This has been happening running in a cluster on 3-4 nodes of 
64 cpus each, and I'm running purely on MPI with this

mpirun -np $SLURM_NTASKS --bind-to core cp2k.psmp -i neb.inp -o neb.out

and open mp threads set explicitly to 1.

I've brought this up to my system admins and they haven't been able to 
detect any issues like over subscription, load balancing or any of the 
common culprits. I'm clueless on what could be happening. On the physics 
side everything seems ok... I'm not getting any unreasonable energies or 
gaps, the total charge and spin also seem ok.

I'm attaching my input (.inp file) and a shortened txt file with the lines 
I'm more concerned about. I tested the convergence parameters vs a plane 
wave calculation and with successive cp2k calculations so I believe they 
should be Ok.

Extra note: I've tried on 1 and 2 nodes with a smaller system and this 
problem didn't come up, granted being a different system it is not the same 
problem.

Any help is welcome, and thanks in advance,

Sofía

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/6e3b0ae1-efb1-42f2-a89e-27b55d5730a7n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260915/43c302a8/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: neb.inp
Type: chemical/x-gamess-input
Size: 2622 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260915/43c302a8/attachment.inp>
-------------- next part --------------
&MOTION
  &BAND
    NPROC_REP  16           
    BAND_TYPE  CI-NEB      
    NUMBER_OF_REPLICA  10    
    K_SPRING  0.1             ! Ha/Bohr^2 
    ALIGN_FRAMES  T           

    &OPTIMIZE_BAND
      OPT_TYPE  DIIS
      OPTIMIZE_END_POINTS T  
      &DIIS
        MAX_STEPS  500
      &END DIIS
    &END OPTIMIZE_BAND

    &REPLICA
      COORD_FILE_NAME  start.xyz
    &END REPLICA
    &REPLICA
      COORD_FILE_NAME  end.xyz
    &END REPLICA

  &END BAND
&END MOTION

&FORCE_EVAL
  METHOD QS
  &DFT

    BASIS_SET_FILE_NAME ../../../BASIS_MOLOPT
    POTENTIAL_FILE_NAME ../../../GTH_POTENTIALS

   #----------------
     CHARGE -2
     MULTIPLICITY 0
     UKS
   #----------------

    &MGRID
      REL_CUTOFF 80.0
      NGRIDS     5
      CUTOFF  800.0
    &END MGRID
    &QS
      EPS_DEFAULT 1.0E-10
      EPS_PGF_ORB 1.0E-6
    &END QS

    &SCF
      SCF_GUESS ATOMIC
      MAX_SCF 10
      EPS_SCF 1.0E-5
      &OUTER_SCF
        EPS_SCF 1.0E-5
        MAX_SCF 300
      &END OUTER_SCF

      &OT T
        MINIMIZER DIIS
        PRECONDITIONER FULL_SINGLE_INVERSE  !  FULL_ALL
        SAFE_DIIS  T
        ENERGY_GAP 0.05
        STEPSIZE 0.2
        ROTATION
      &END OT

    &END SCF

    &XC
      &XC_FUNCTIONAL
        &MGGA_X_R2SCAN
        &END MGGA_X_R2SCAN
        &MGGA_C_R2SCAN
        &END MGGA_C_R2SCAN     
      &END XC_FUNCTIONAL
      &XC_GRID
        XC_SMOOTH_RHO NN50
        XC_DERIV NN50_SMOOTH
      &END XC_GRID
    &END XC

  &END DFT