Issues with tight binding methods

[Pierre-André Cazade]

Dear CP2K users,

I am trying to perform a cell_opt on a quercetin crystal. It is a simple organic crystal with 130 atoms. I want to test how well tight binding methods perform compared with a DFT calculation and a classical forcefield one. However, neither DFTB nor XTB work. The only way to perform the calculation is to disable the SCF cycle, otherwise the energy becomes positive and never converges. I get a relatively decent structure with DFTB in that case but it feels rather unsatisfactory. Also, it seems that the various options available for DFTB have little or no impact when the cycle is disabled, be it DO_EWALD, DISPERSION, or DIAGONAL_DFTB3. In particular, the D3 dispersion seems to have no effect whereas the default uff one leads to a contraction of the unit cell.

I tried to vary the GMAX in the Ewald sum, I played with multi_cell and kpoints, but nothing seems to make it possible to converge the scf cycle. Worse, increasing the number of kpoints seems to be detrimental to the crystal structure when optimised with the scf cycle off.

I am surprised that neither DFTB nor XTB work for this simple system, despite tight binding methods being used more and more to study organic co-crystals. I am looking forward to your insight.

Regards,

Pierre

[Maxime Van den Bossche]

Dear Pierre,

I don't know about the effect of dispersion, but this "3ob" parameter set is definitely supposed to be used with charge self-consistency. To get the SCC cycle to converge, usually only the DIRECT_P_MIXING method works here, and with this setting I indeed get good convergence (see attached input file, slightly modified from what you sent).

Also the DFTB regtests only feature this mixing method, so I presume the other methods like BROYDEN_MIXING don't work well with the current implementation of DFTB in cp2k. I don't think this is explicitly documented somewhere.

best,

Maxime

[Pierre-André Cazade]

Dear Maxime,

Thank you very much. That was so simple... I have tested the dispersions both d3 and uff. UFF works but leads to a contraction of the crystal on the c axis. D3 does not work, I obtain the same structure with it as when I disable the dispersion. This structure is particularly bad with the c axis doubling in value.

Regards,

Pierre

[Maxime Van den Bossche]

Hello again,

I had a closer look at the 3ob parameter set and it seems like not all the

'bells and whistles' have been implemented in the DFTB module in CP2K

to make proper use of this parametrization:

- It is supposed to be used with 'full' third-order terms, not just the diagonal ones

(note: there are also the LMAX_DFTB and DFTB3_PARAM keywords which

need to be set under CP2K_INPUT/FORCE_EVAL/SUBSYS/KIND).

- Becke-Johnson damping of the D3 correction doesn't seem available in the

DFTB module (though it is available for e.g. DFT calculations).

The simplest way to deal with this, I think, is to use the DFTB+ code instead of

CP2K. As an example, I have attached a ZIP file with DFTB+ input files and the

final geometry I got out of it. There is a +20% expansion in the c-direction.

I have no real experience with DFTB calculations of molecular crystals, so I don't

know if this is a typical deviation or not. Where did the initial geometry come from?

/Maxime

[Pierre Cazade]

Hello Maxime,

Thank you for the detailed answer. I will have a look at DFTB+ software.

Regarding cp2k, if I understand well, it is better so far to only use the scc and non-scc parameters which are provided within the data section of the code and not the more advanced parameters found on DFTB website.

Regards,
Pierre

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/4fed78db-4d46-4d57-9dec-b261fd096a2e%40googlegroups.com.

-- 
Dr Pierre Cazade, PhD
AD3-023, Bernal Institute,
University of Limerick,
Plassey Park Road,
Castletroy, co. Limerick,
Ireland