Orbital Occupation and Restarting OT calculations

[Christopher O'Brien]

OT calculations cannot be restarted in CP2K.

The error listed below results from using an existing WFN file and the appropriate coordinates from another CP2K MD run to start a new METADYNAMICS run. In my DFT section I have:

&FORCE_EVAL

 METHOD QS
 &DFT

    BASIS_SET_FILE_NAME BASIS_MOLOPT
    POTENTIAL_FILE_NAME GTH_POTENTIALS
    WFN_RESTART_FILE_NAME XXX-RESTART.wfn
    UKS
    RELAX_MULTIPLICITY 0.15

  &SCF

    &OT

      PRECONDITIONER FULL_SINGLE_INVERSE
      MINIMIZER CG

    &END OT

    EPS_SCF 1.0E-6
    SCF_GUESS RESTART
    MAX_SCF 100

  &END SCF

 &END DFT

 &SUBSYS ...

&END FORCE_EVAL

However, after searching the manual I can't find an option to add MOs to OT calculations. I also don't understand why I need to considering that, in the output of the calculation I am starting from, all spins are already paired. 

 Spin 1


 Number of electrons: 1993

 Number of occupied orbitals: 1993

 Number of molecular orbitals: 1993


 Spin 2


 Number of electrons: 1993

 Number of occupied orbitals: 1993

 Number of molecular orbitals: 1993
 

 Number of orbital functions: 15770

 Number of independent orbital functions: 15770


 Extrapolation method: initial_guess


 *** 04:08:41 WARNING in qs_mo_types:set_mo_occupation_2 :: All alpha MOs ***

 *** are occupied. Add more alpha MOs to allow for a higher multiplicity ***

[Matt W]

Hi,

I think the problem is coming from the relax multiplicity keyword in your input - that only seems to make sense to me using extra MOs and not the OT optimizer as it requires having extra MOs it can put electrons into to vary the multiplicity.

Matt

[Christopher O'Brien]

Matt,

Thanks for you response.  Although I don't exactly understand how the OT minimizer works, there is no error generated by CP2K or any mention of a conflict between changing multiplicity in the documentation for the OT minimizer.  

However, I don't think there is a conflict with using OT and varying multiplicity because using the same input, without the WFN file used as an initial guess, runs without error.   

[Matt W]

I can confirm what you say, but I'm pretty sure it is incorrect behaviour of cp2k to run with OT and relax_multiplicity.  

The relax_multiplicity flag must be getting ignored when the code does an atomic guess.

Matt

[Christopher O'Brien]

So is it still possible to use OT with ADDED_MOS or must multiplicity be fixed. I am modeling a molecule interacting with a surface, I wanted to let the spin state be allowed to change during the interaction if needed. I don't want to give up on OT since it is so fast compared to diagonalization for my large system.

[Matt W]

I think you have to fix multiplicity.

I'd think that it would likely be more efficient to relax/perform MD in one spin state, then check whether other spin states were competitive in many cases. This is where it really is annoying that currently you can't restart easily from the orbitals of a different spin state.

Matt