[CP2K:4967] Geometry optimisation converges to nonsense

[Jörg Saßmannshausen]

Hi Herbert,

I would guess that your convergence criteria are set too loose and hence it 
'converges' but the results are nonsense.

You  can check that from your output file. 
Just search for:
"Informations at step"

and have a look at the printed values here. 
You should see something like that:
  Optimization Method        =                 BFGS
  Total Energy               =      -160.5982073285
  Real energy change         =        -0.0000000071
  Predicted change in energy =        -0.0000000070
  Scaling factor             =         0.8736246364
  Step size                  =         0.0001454725
  Trust radius               =         0.1000000000
  Decrease in energy         =                  YES
  Used time                  =               39.292

  Convergence check :
  Max. step size             =         0.0000532434
  Conv. limit for step size  =         0.0003000000
  Convergence in step size   =                  YES
  RMS step size              =         0.0000171249
  Conv. limit for RMS step   =         0.0001500000
  Convergence in RMS step    =                  YES
  Max. gradient              =         0.0000112942
  Conv. limit for gradients  =         0.0000450000
  Conv. in gradients         =                  YES
  RMS gradient               =         0.0000036405
  Conv. limit for RMS grad.  =         0.0000300000
  Conv. in RMS gradients     =                  YES

Some parameters which might help here are:
In the force_eval section I got these settings:
  &DFT
    CHARGE ${CHARGE}
    &MGRID
      CUTOFF ${CUTOFF}
      NGRIDS 5
    &END MGRID
    &QS
      EPS_DEFAULT 1.0E-16
    &END QS
    &SCF
      SCF_GUESS random 
      EPS_SCF       ${SCF_CONV}
      MAX_SCF      30
      &PRINT
[ ... ]

with 
SCF_CONV   1.0E-8

I can send you the whole file off-list if you want to.

Let u know how you are getting on here. 

All the best from dull London!

Jörg

On Wednesday 19 Feb 2014 15:20:28 hfruchtl wrote:
> Folks,
> 
> I am trying to do my first baby steps with CP2K, after several aborted
> attempts, so please tell me if I do something very stupid.
> 
> One thing that worries me after first tests is that geometry optimisations
> look converged, when they clearly aren't. In the example below, I start
> something that should be a water dimer at a very wrong geometry. When I
> plot the energy changes, they become smaller as expected, and after less
> than 30 steps the program finishes. But the geometry in the xyz file looks
> nothing like I expect (H2O)2 to look. When I optimise this structure with
> the program that must not be named, using the same functional, I get a
> proper dimer. If I then plug this into CP2K, it remains close to this
> geometry (good!) with a much lower energy than the first attempt (also
> good). I understand that BFGS may not be good from a bad guess, but I would
> expect CG to get there in the end, or at least not look like it converged.
> I'd be very surprised if this was a local minimum. In short: how do I know
> if I can trust an optimised geometry?
> 
> Input below. Thanks in advance,
> 
>   Herbert
> 
> &GLOBAL
>   PROJECT water-dimer
>   RUN_TYPE GEO_OPT
>   PRINT_LEVEL high
> &END GLOBAL
> 
> &MOTION
>   &GEO_OPT
>     MAX_ITER 2000
>     MINIMIZER CG
>   &END GEO_OPT
>   &PRINT
>     &TRAJECTORY
>       LOG_PRINT_KEY T
>       FORMAT xyz
>       UNIT angstrom
>       &EACH
>         GEO_OPT 1
>       &END EACH
>       ADD_LAST NUMERIC
>     &END TRAJECTORY
>   &END PRINT
> &END MOTION
> 
> &FORCE_EVAL
>   METHOD Quickstep
>   &DFT
>     BASIS_SET_FILE_NAME /usr/local/CP2K/BASIS_MOLOPT
>     POTENTIAL_FILE_NAME /usr/local/CP2K/GTH_POTENTIALS
>     &MGRID
>       CUTOFF 100
>     &END MGRID
>     &QS
>     &END QS
>     &SCF
>       SCF_GUESS ATOMIC
>     &END SCF
>     &XC
>       &XC_FUNCTIONAL PBE
>       &END XC_FUNCTIONAL
>     &END XC
>   &END DFT
> 
>   &SUBSYS
>     &CELL
>       A    15.0 0.0 0.00
>       B    0.0 15.0 0.0
>       C    0.0 0.0 15.0
>       PERIODIC NONE
>     &END CELL
>     &COORD
>       O         0.0006535865        0.0348037292        6.5225572751
>       H        -0.0003839912       -0.0241840602        7.5017980482
>       H        -0.0019131905        1.0517623171        6.4318839506
>       O         0.0244126401        0.0102938023        4.0045222106
>       H        -0.0164725514       -0.0290451673        5.4795600635
>       H         1.0326311856        0.2923983937        4.3857121035
>     &END COORD
> 
>     &KIND O
>       BASIS_SET DZVP-MOLOPT-GTH-q6
>       POTENTIAL GTH-PBE-q6
>     &END KIND
>     &KIND H
>       BASIS_SET DZVP-MOLOPT-GTH-q1
>       POTENTIAL GTH-PBE-q1
>     &END KIND
>   &END SUBSYS
> &END FORCE_EVAL

-- 
*************************************************************
Jörg Saßmannshausen
University College London
Department of Chemistry
Gordon Street
London
WC1H 0AJ 

email: j.sassma... at ucl.ac.uk
web: http://sassy.formativ.net

Please avoid sending me Word or PowerPoint attachments.
See http://www.gnu.org/philosophy/no-word-attachments.html
-------------- next part --------------
Q29udGVudC1UeXBlOiBhcHBsaWNhdGlvbi9wZ3Atc2lnbmF0dXJlOyBuYW1lPSJzaWduYXR1cmUu
YXNjIg0KQ29udGVudC1EZXNjcmlwdGlvbjogVGhpcyBpcyBhIGRpZ2l0YWxseSBzaWduZWQgbWVz
c2FnZSBwYXJ0Lg0KDQotLS0tLUJFR0lOIFBHUCBTSUdOQVRVUkUtLS0tLQ0KVmVyc2lvbjogR251
UEcgdjEuNC4xMiAoR05VL0xpbnV4KQ0KDQppRVlFQUJFQ0FBWUZBbE1FMm9BQUNna1FFSTFhS1Br
OHZBc2oxZ0NlSXdkM21Mck1TS1pMWXVlYmlEWmtHc1g4DQpzdklBbjBhZkpUa1VnRkd4djE0N3Vu
Z1FLYWx3TkFpcQ0KPSt1a0gNCi0tLS0tRU5EIFBHUCBTSUdOQVRVUkUtLS0tLQ0K