DFTB charge equilibration discrepancy

[Nir]

Hi Juerg,

> Could it be an odd number of electrons? What happens if
> you are using "LSD" for NO?

I don't think an odd number of electrons is causing the problem.  For
example, if I add a -1 charge to NO, the codes still disagree.
However, contrary to what I wrote earlier, DFTB+  and CP2K do agree
for CO and NH3.  They still disagree for CO2, NO2, and for systems
such as nitromethane and TATB.  As a wrote earlier, I get excellent
comparisons for H2O, diamond, methane, and LN2.  It's just for CO2 and
systems with NO2 that I observe a discrepancy.  The energies for these
systems tend to be off by ~0.6 Ha, and the forces are off by anywhere
from 10-100%.  I also tried turning off the scc charge equilibration
for both codes.  The agreement improved slightly, but was still off
overall.  Another thing is that CP2K tends to produce lower energies
than DFTB+.  I haven't quantified this at all, but that seemed to be
what was happening when I eyeballed the results.

> Could it be that your DFTB+ run is using a l-dependent
> hardness by default? CP2K has only a single hardness
> value per atom implemented.
> Or is DFTB+ taking another hardness value from the
> input file, CP2K is using the s-value.

DFTB+ has an input flag where the code will only use the Hubbard U for
the s-shell.  I assume this is the same thing as you are talking
about, above.  I used that flag for the comparisons I'm talking about.

Any other thoughts on other things I could look into?

Thanks for the help,

Nir