Problem during SCF convergence in AIMD (NVT) run

[Rohan Gupta]

Dear CP2K community,

I recently started working on cp2k. I am performing DFT-MD calculations and facing problem regarding SCF convergence during the NVT run.

Initially for around 1.8 ps each step (NVT run) SCF was converging smoothly (within one loop) but there after some of the steps are not converging during the run. For such steps maximum limit of the inner and outer loops were reached and the SCF was not converged. The cp2k still dumps the position, velocity and the energy for the system for such steps (with a warning SCF NOT CONVERGED). The system then runs further and convergence is achieved after few fs. But later on in around 4.3 ps same problem arises and all the maximum no of loops (inner and outer) were completed and yet no convergence was achieved.

Kindly help me in finding the cause of the problem and if it is a regular problem faced in most of the cases then suggest the reason for the same.

Enclosed within are parameters of the .inp file for the DFT-MD simulation and also the .ener file of the run carried out so far.

Thanks

Best Regards

Rohan Gupta

PARAMETERS (INPUT) FILE:

PBC: ON

Basis Set:

TZVP-MOLOPT-GTH : For N/S/O/C/F/H atoms

DZVP-MOLOPT-GTH: For Li atom

Mass for H atom: 2

Potential:

GTH-PBE-qx : Where x are the no of valence electrons for each atom used

&DFT
    BASIS_SET_FILE_NAME BASIS_MOLOPT
    POTENTIAL_FILE_NAME GTH_POTENTIALS
    &QS
      EPS_DEFAULT 1.0E-10
    &END QS
     &MGRID
      NGRIDS 4
      CUTOFF 500
      REL_CUTOFF 50
    &END MGRID
    &SCF
      SCF_GUESS ATOMIC
      EPS_SCF 1.0E-07
      MAX_SCF 200
      &OT
        PRECONDITIONER FULL_KINETIC
        MINIMIZER DIIS
      &END OT
      &OUTER_SCF
        MAX_SCF 200
        EPS_SCF 1.0E-07
      &END OUTER_SCF
      &MIXING T
        METHOD BROYDEN_MIXING
        ALPHA 0.4
        NBROYDEN 8
      &END MIXING
      &PRINT
        &RESTART OFF
        &END RESTART
      &END PRINT
    &END SCF
    &XC
      &XC_FUNCTIONAL PBE
      &END XC_FUNCTIONAL
      &VDW_POTENTIAL
        POTENTIAL_TYPE PAIR_POTENTIAL
        &PAIR_POTENTIAL
          PARAMETER_FILE_NAME dftd3.dat
          TYPE DFTD3
          REFERENCE_FUNCTIONAL PBE
          R_CUTOFF 16
        &END
      &END VDW_POTENTIAL
    &END XC
    &PRINT
      &MULLIKEN ON
        FILENAME charges
        #PRINT_ALL
        &EACH
          MD 1
        &END EACH
      &END MULLIKEN
    &END PRINT
 &END DFT
 &MOTION

  &MD
    ENSEMBLE NVT
    STEPS 500000
    TIMESTEP 1
    TEMPERATURE 400
    &THERMOSTAT
      &NOSE
        TIMECON 100
      &END NOSE
    &END THERMOSTAT
  &END MD
&END MOTION

[Thomas Kühne]

Dear Rohan Gupta, 

200 OUTER_SCF iterations are mot effective, i.e. you either converge much earlier, or never. 

I suspect that you are experiencing an oscillation between DIIS and SD, which must be prevented 

except that you fallback to SD only very few times. This happens when the solution can not be 

obtained by a linear combination of previous guess vectors, leading to an increase an energy 

and thus activating SAFE_DIIS. Pleas consult http://vergil.chemistry.gatech.edu/notes/diis/diis.pdf 

for details. Depending on when this problems happens during OT minimization steps, one way is 

to reduce STEPSIZE and/or use another PRECONDITIONER such as FULL_SINGLE_INVERSE, 

or FULL_ALL. IF this does not help, you may switch from DIIS to CG … 

Good luck, 

Thomas Kühne

--
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+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/55d97975-11cd-49f9-8873-c638f38fadaan%40googlegroups.com.
<LiFSI-system-md-1.ener>

==============================

Thomas D. Kühne

Dynamics of Condensed Matter

Chair of Theoretical Chemistry

University of Paderborn

Warburger Str. 100

D-33098 Paderborn

Germany

thomas...@upb.de

+49/(0)5251/60-5726

[Rohan Gupta]

Dear Thomas Kühne,

Thanks a lot for looking into the problem and for your suggestions and explanation. I will make changes in my input file as per you suggestions and get back to you as soon as possible with the results.

Thanks

Best Regards

Rohan Gupta