[CP2K:9944] Problematic DFTB geometry optimization

[Maxime Van den Bossche]

OK, I see -- using a Gamma-point-only k-point grid indeed
gave the expected repulsive energy for the TiO2 molecule
as well as for the rutile structure (i.e. same as in DFTB+).
Using a Gamma-point-only k-point grid also solved the 
original rutile geometry optimization issue, so the repulsive
interactions were indeed the culprit here. Would it be possible 
to document this behaviour more explicitly, e.g. in the manual?

Both for the isolated TiO2 molecule as for the rutile TiO2,
I needed to reduce the ALPHA value in the MIXING section 
to 0.11 or lower (with the DIRECT_P_MIXING scheme) to 
get the SCF to converge.

I did some additional one-to-one comparison for an isolated 
TiO2 molecule. The electronic energy contributions (0th order 
Hamiltonian energy and the charge fluctuation energy),
as well as the charges and forces,  both with OT and (gamma-point) 
diagonalization, fully agree with DFTB+. This is the case,
provided that "orbital-resolved SCC" is turned off in DFTB+ 
(i.e. only one Hubbard value per element, that of the s-orbital in 
the SKF file, as mentioned previously 
<https://groups.google.com/d/msg/cp2k/1ZG1uIVCH2s/FSk6JjDe4q0J>). So the 
electronics
seem to be in order.

Regards,
Maxime
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180202/5f72bce0/attachment.htm>