[CP2K-user] 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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200225/d9e82cf4/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: quer-opt.inp
Type: chemical/x-gamess-input
Size: 5911 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200225/d9e82cf4/attachment.inp>