Minimization bug

roberto roberto.... at gmail.com
Mon Sep 15 14:39:26 CEST 2008


Hi to all,
I use cp2k to minimize a lot of structures of a big system (305atoms)
I start with only five minimization steps, about 10% of the structure
minimization doesn't reach the last stepm, whit smaller system
(100atoms) the value is 2-3%.
After one or two minimization steps the algorithm doesn't reach
convergence during autoconsistent procedure.
The visible anoumalousness is relative to the energy, these 10%
structures, in less than 5 steps, reach an energy of about 2000kJ/mol
(yes 2MJ/mol) less than the others.
I used the follow template.inp

Same suggestion?
thanks

## INPUT ##
&FORCE_EVAL
  METHOD Quickstep
  &DFT
    CHARGE +2
    &MGRID
      CUTOFF 280
    &END MGRID
    &QS
      EPS_DEFAULT 1.0E-8
    &END QS
    &SCF
      &OT
        PRECONDITIONER FULL_SINGLE_INVERSE
      &END OT
      MAX_SCF 40
      &OUTER_SCF
        MAX_SCF 5
      &END OUTER_SCF
      SCF_GUESS ATOMIC
    &END SCF
    &XC
      &XC_FUNCTIONAL BLYP
      &END XC_FUNCTIONAL
    &END XC
  &END DFT
  &SUBSYS
    &CELL
      ABC 35.0 35.0 35.0
      UNIT ANGSTROM
    &END CELL

    &COORD
     XXXCOORDXXX
    &END COORD
    &KIND H
      BASIS_SET TZV2P-GTH
      POTENTIAL GTH-BLYP-q1
    &END KIND
    &KIND C
      BASIS_SET TZV2P-GTH
      POTENTIAL GTH-BLYP-q4
    &END KIND
    &KIND N
      BASIS_SET TZV2P-GTH
      POTENTIAL GTH-BLYP-q5
    &END KIND
    &KIND O
      BASIS_SET TZV2P-GTH
      POTENTIAL GTH-BLYP-q6
    &END KIND
  &END SUBSYS
&END FORCE_EVAL

&MOTION
  &GEO_OPT
    OPTIMIZER LBFGS
    MAX_ITER    5
    &CG
       MAX_STEEP_STEPS 5
       &LINE_SEARCH
         TYPE 2PNT
       &END LINE_SEARCH
    &END CG
  &END GEO_OPT

  &MD
    ENSEMBLE NVT
    STEPS 50000
    TIMESTEP 1.0
    TEMPERATURE 50.0
  &END MD

  &PRINT
    &FORCES
       FORMAT XMOL
       &EACH
          MD 1
       &END EACH
    &END FORCES
  &END PRINT
&END MOTION

&GLOBAL
  PROJECT project_name
  RUN_TYPE GEO_OPT
  PRINT_LEVEL LOW
&END GLOBAL



More information about the CP2K-user mailing list