convergence issues with rCAM-B3LYP

Rad rad.... at arl.army.mil
Thu Feb 21 15:12:08 CET 2008


Dear All,

I managed to compile cp2k (Last CVS entry               qs_scf_post.F/
1.149/Wed Feb 13 23:07:05 2008) with libint (thaks to Teo's
suggestions) and tested the sample input that uses rCAM-B3LYP. This
input file was set up to do a single point energy calculation. I
adopted the same input file for my system with just 56 atoms and tried
a simple geometry optimization. The wave function is not converging at
all (at most I got two convergences of the outer loop). Here is the
input section I used:
&FORCE_EVAL
  METHOD Quickstep
  &DFT
    BASIS_SET_FILE_NAME EMSL_BASIS_SETS
    POTENTIAL_FILE_NAME POTENTIAL
    &MGRID
      CUTOFF 1600
      NGRIDS  5
      REL_CUTOFF 50
    &END MGRID
    &QS
      METHOD GAPW
      MAP_CONSISTENT
       EXTRAPOLATION PS
       EXTRAPOLATION_ORDER 2
    &END QS
    &SCF
      &OT on
       MINIMIZER  DIIS
       PRECONDITIONER FULL_SINGLE_INVERSE
       STEPSIZE 0.05
      &END OT
      &OUTER_SCF
         EPS_SCF 1.0E-6
         MAX_SCF 40
      &END OUTER_SCF
      EPS_SCF 1.0E-6
      SCF_GUESS RESTART
      MAX_SCF 500
    &END SCF
    &XC
      &XC_FUNCTIONAL
        &BECKE88
          SCALE_X 0.95238
        &END
        &BECKE88_LR
          OMEGA 0.33
          SCALE_X -0.94979
        &END
        &LYP
          SCALE_C 1.0
        &END
        &XALPHA
          SCALE_X -0.13590
        &END
      &END XC_FUNCTIONAL
      &HF
       &SCREENING
         EPS_SCHWARZ 1.0E-10
         EPS_BOX 1.0E-10
       &END
       &MEMORY
         MAX_MEMORY 100
       &END
       &INTERACTION_POTENTIAL
         POTENTIAL_TYPE MIX_CL
         OMEGA 0.33
         SCALE_LONGRANGE 0.94979
         SCALE_COULOMB 0.18352
       &END
      &END
    &END XC
  &END DFT

Could you please take a look at it and offer me suggestions to tweak
the parametrs for wavefunction convegence.
I also tried CG with different step sizes in the OT section and also a
run without OT section but still no convergence at all. I will
appreciate any suggestions/pointers you may offer in this regard.

Thanks
Rad


More information about the CP2K-user mailing list