Re: Temp control problem, OT convergence problem, Langevin dynamics SGCPMD

[Alex]

Hi,

While I too have questions about the annealing procedure in CP2K, if you have a small system, Nose-Hoover thermostatics is a generally bad idea regardless of its time constant. 

Try CSVR or adaptive Langevin. For the initial relaxation, try to go with a very tight temperature control (a time constant of a few femtoseconds for CSVR, for example).

Best,

Alex

On Wednesday, November 5, 2014 2:50:30 AM UTC-7, Bouzid Assil wrote:

Dear CP2k users,

I am using cp2k to perform MD simulations based on the second generation Car-Parrinello MD (Kühne scheme). I have started by a small binary chalcogenide system that I have already generated using the standard CPMD.
As a first step I want to relax my system at 0K using the ANNEALING procedure, this is supposed to reduce the forces acting between the "atoms" at each MD step by a given factor (say 0.9). To control the situation I have used at first gamma_D = 0.0000. My problem is that after more than 30ps the temperature stabilize around 1000K and won't go down. I have also tried an NVT run at 300K using the NOSE HOOVER thermostat but no way, the thermostat is not able to control the temperature (I have tried different TIMECON values, always the same behaviour!).
When analysing the output files, there is a common strange thing between all these runs (and also an other run on water molecules), when performing the SCF WAVEFUNCTION OPTIMIZATION typically the convergence is reached after  2 iterations of the outer SCF loop BUT after the first ~10 MD steps the convergence is not reached at all and only 2 OT steps are performed! is this normal (for sure no!) but why it happens then?!

 here is an example of output:

   ----------------------------------- OT ---------------------------------------

  Allowing for rotations:  F
  Optimizing orbital energies:  F
  Minimizer      : DIIS                : direct inversion
                                         in the iterative subspace
                            using      : -   7 DIIS vectors
                                         - safer DIIS on
  Preconditioner : FULL_KINETIC        : cholesky inversion of T + eS
  Precond_solver : DEFAULT
  stepsize       :    0.08750000
  energy_gap     :    0.00200000

  eps_taylor     :   0.10000E-15
  max_taylor     :             4

  mixed_precision    : F

  ----------------------------------- OT ---------------------------------------

  Step     Update method      Time    Convergence         Total energy    Change
  ------------------------------------------------------------------------------
     1 OT DIIS     0.87E-01    2.1     0.00000982      -906.5667899239 -9.07E+02
     2 OT DIIS     0.87E-01    2.5     0.00000932      -906.5667939977 -4.07E-06

  *** SCF run NOT converged ***


  Electronic density on regular grids:       -640.0000000000       -0.0000000000
  Core density on regular grids:              639.9999999985       -0.0000000015
  Total charge density on r-space grids:       -0.0000000015
  Total charge density g-space grids:          -0.0000000015

  Overlap energy of the core charge distribution:               0.00000056633200
  Self energy of the core charge distribution:              -1927.34045098667070
  Core Hamiltonian energy:                                    434.65748894621680
  Hartree energy:                                             790.75025229030484
  Exchange-correlation energy:                               -204.63408481387373

  Total energy:                                              -906.56679399769098

 ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):             -906.566801465010712


 *******************************************************************************
 ENSEMBLE TYPE                =                                         LANGEVIN
 STEP NUMBER                  =                                            34788
 TIME [fs]                    =                                     34788.000000
 CONSERVED QUANTITY [hartree] =                              -0.906018037069E+03

                                              INSTANTANEOUS             AVERAGES
 CPU TIME [s]                 =                        9.85                 9.56
 ENERGY DRIFT PER ATOM [K]    =         -0.352815792553E+06  -0.346958228435E+06
 POTENTIAL ENERGY[hartree]    =         -0.906566801465E+03  -0.906008788692E+03
 KINETIC ENERGY [hartree]     =          0.548764395757E+00   0.221663346691E+01
 TEMPERATURE [K]              =                     970.789             3921.326
 *******************************************************************************



Here is a part of my input file:


&MOTION
  &MD
    ENSEMBLE  LANGEVIN    
    STEPS 15000
    ANNEALING 0.8
    TIMESTEP 1.0
    &LANGEVIN
      GAMMA       0.0001  
      NOISY_GAMMA 0.000
    &END LANGEVIN
  &END MD
  &PRINT
    &RESTART
      &EACH
        MD 1
      &END EACH
      BACKUP_COPIES 3
    &END RESTART
  &END PRINT
&END MOTION

&FORCE_EVAL
  METHOD QUICKSTEP
  &DFT
    BASIS_SET_FILE_NAME BASIS_MOLOPT
    POTENTIAL_FILE_NAME GTH_POTENTIALS
    WFN_RESTART_FILE_NAME meta-RESTART.wfn
    &MGRID
      &RS_GRID
        DISTRIBUTION_TYPE REPLICATED
      &END RS_GRID
      CUTOFF 300
    &END MGRID
    &PRINT
      &DFT_CONTROL_PARAMETERS
      &END DFT_CONTROL_PARAMETERS
    &END PRINT
    &QS
      EPS_DEFAULT 1.0e-12
      EXTRAPOLATION ASPC    
      EXTRAPOLATION_ORDER 3 
      MAP_CONSISTENT TRUE
    &END QS
    &SCF
      EPS_SCF 1.0e-7
      MAX_SCF 50
      MAX_SCF_HIST 2        
      &OT ON
        MINIMIZER DIIS
        PRECONDITIONER FULL_KINETIC
        ENERGY_GAP 0.002
        STEPSIZE 0.0875    
      &END OT
      &OUTER_SCF ON
        MAX_SCF 100
        EPS_SCF 1.0e-7
      &END OUTER_SCF



Any input on the possible origin of this problem is highly appreciated! Many thanks.
All the best,

Assil Bouzid

[Lila Bouëssel Du Bourg]

Hi, I have exactly the same problem problem. But since you're last post is a bit old, I hope you've solved the problem so you could show me the way O:-)