Using cNNP + D3 for geometry optimization in CP2K

[haphysix]

An example for running CP2K-MD with METHOD NNP is included in examples/from-trajectroy-to-model/04-C-NNP-MD.
Would you be able to include a geometry optimization input example with the following requirements?

• MOTION is geometry optimization
• System is periodic
• METHOD is NNP
• D3 dispersion correction is included in energy and force calculations.

Thanks very much!

[cschran]

Hi haphysix,

I break down this question into two answers, first for the geometry optimization:

Geometry optimizations work via the RUN_TYPE GEO_OPT

...
RUN_TYPE GEO_OPT
...

Replace the MD section for suitable input e.g. from here: https://www.cp2k.org/howto:geometry_optimisation

&GEO_OPT
    TYPE MINIMIZATION
    MAX_DR    1.0E-03
    MAX_FORCE 1.0E-03
    RMS_DR    1.0E-03
    RMS_FORCE 1.0E-03
    MAX_ITER 200
    OPTIMIZER CG
    &CG
      MAX_STEEP_STEPS  0
      RESTART_LIMIT 9.0E-01
    &END CG
&END GEO_OPT

[cschran]

For the addition of different types of interactions, cp2k allows to add various FORCE_EVAL sections.

&MULTIPLE_FORCE_EVALS
  FORCE_EVAL_ORDER  2 3
  MULTIPLE_SUBSYS  F
&END MULTIPLE_FORCE_EVALS

Use the MIXED section to define the mixing rules, e.g. simple addition:

&FORCE_EVAL
  METHOD  MIXED
  &MIXED
    MIXING_TYPE  GENMIX
    GROUP_PARTITION  1023 1
    &GENERIC
      MIXING_FUNCTION a+b
      VARIABLES a b
    &END GENERIC
  &END MIXED
...
&END FORCE_EVAL

and define afterwards what FORCE_EVAL you want to mix.

&FORCE_EVAL
  METHOD  NNP
...
&END FORCE_EVAL

&FORCE_EVAL
  METHOD  QS
...
&END FORCE_EVAL

[haphysix]

Thanks for your reply, Christoph.
At the moment, I am compiling CP2K. Once I have CP2K ready, I'll give it a try.

[haphysix]

Thanks for your message, Christoph. The suggested CP2K inputs work just fine. After playing with CP2K inputs for a few days, I know that D3 can only be included in the DFT/XC_FUNCTIONAL block, which means that some unwanted energy and force contributions will be added to NNP values. This will disrupt the geometry optimization process. Could FORCE_EVAL/NNP be configured to include the DFTD3 correction?

[cschran]

This is actually more a quickstep CP2K related question and best asked in the CP2K google group, where it might even have been answered.

However, you can try adding a XC_FUNCTIONAL dummy section via

&XC_FUNCTIONAL NONE
&END XC_FUNCTIONAL