DFTB charge equilibration discrepancy

[Nir]

Hi Juerg,

> there might be several reasons for this problem.
> To get more information could you do the following tests:
>
> - run both programs without EWALD (and PERIODIC NONE in CP2K)
>    in order to verify that we agree on the basics

Thank you for your speedy reply.  Unfortunately turning off the Ewald
sums in both codes (and using PERIODIC NONE in CP2K) did not produce
agreement.  I also tried doing the same by using the "cluster" option
with DFTB+ without success.  Is there anything else you can suggest to
try?  Again, DFTB/DFTB+ and CP2K agree very well for systems such as
CO, H2O and CH4, but disagree for CO and NO.  I'm using the scc input
files provided by CP2K for all three codes.

Thanks again,

Nir

>
> - Converge all EWALD parameters in both codes in order to
>    exclude problems with different implementations
>
> When we tested the implementation we reached agreements
> with the reference code (DMON) far better than 1%
>
> best regards
>
> Juerg
>
> ----------------------------------------------------------
> Juerg Hutter                   Phone : ++41 44 635 4491
> Physical Chemistry Institute   FAX   : ++41 44 635 6838
> University of Zurich           E-mail: hut... at pci.uzh.ch
> Winterthurerstrasse 190
> CH-8057 Zurich, Switzerland
> ----------------------------------------------------------
>
> On Thu, 8 Jan 2009, Nir wrote:
>
> > Hi All,
>
> > I have been benchmarking the DFTB implementation in CP2K against a
> > version of the DFTB code from Prof. Frauenheim's group (www.dftb.org)
> > and have run into some problems.  FYI, the DFTB code I used force
> > matches to DFTB+ to within 1%.  Specifically, for this problem we are
> > interested in systems containing C, N, O, and H.  For the benchmarking
> > tests I'm talking about, both codes were run using charge
> > equilibration (scc) and Ewald sums, without dispersion, and using
> > identical parameter files.  Similar very low SCF convergence criteria
> > were used for both codes.
>
> > For most systems I tested (e. g., CH4, H2O, CN, diamond), the two
> > codes compared very well and the forces and computed energies matched
> > to within ~1%.  However, the forces and energies were considerably off
> > for systems containing C-O, N-H, and N-O bonds (e. g., CO, NH3 and
> > NO2, respectively).  I hacked into both codes and saw that the
> > repulsive forces are virtually identical for both CP2K and the DFTB
> > code, but that for some reason they equilibrate to different charges
> > for the above systems.   Any help or insight into solving this problem
> > would be greatly appreciated.  I uploaded a tar file called
> > nir_dftb.tar of sample input and output files for a single NH3
> > molecule containing the following files (with hopefully self-
> > explanatory names):
>
> > cp2k.nh3_single.inp
> > cp2k.nh3_singe.out
> > cp2k.charges.dat
> > DFTB.nh3_single.out
> > DFTB.FRC_nh3_single.DAT
> > DFTB.CHR_nh3_single.dat
>
> > In case it matters, I ran cp2k.popt on a single node on a Linux-x86-64
> > machine compiled with the intel fortran compiler.
>
> > Thanks for your help!
>
> > Nir