[DFTB-Plus-User] Issue with the stress tensor calculation in DFTB+ v.1.1

[Ben Hourahine]

Hello Nir,

The code was tested numerically for mostly condensed matter systems
(diamond, silicon, nanotubes and some oxides), with most of the primary
tests being done for the derivatives of the energy with respect to
components of the lattice vectors. These derivatives are then converted
to the stress tensor using the formula from section IIA of

http://link.aip.org/link/JCPSA6/v122/i12/p124508/s1

I've attached the script I use for the stress tensor calculation (not
very well commented unfortunately) and an example for a distorted
diamond unit cell. You will need to edit the path to the dftb+ binary in
the script, which assumes that the dftb_in.hsd file is reading in a
geometry stored in geo.gen.template.

Could you re-send me the input so I can test it (the attachment for the
original message didn't get through to me for some reason)? Typically
I'd expect the numerical and analytical results to agree to better than
10^-4 a.u.

Regards

Ben

Goldman, Nir wrote:
> Hi All,
> 
> I've been testing the analytical stress tensor calculation in DFTB+
> v.1.1, and I can't seem to get the analytical results to match my
> calculations of the numerical values. Attached as an example is a
> configuration of 32 H2O molecules at 1 g/cc equilibrated at 298 K for
> several ps. I’ve been running my tests by doing static geometry
> calculations (Driver = {}) with the SCC convergence set to 1E-12 in
> order to converge the energies as tightly as possible. I’ve been
> calculating two point numerical derivatives by changing the
> computational cell anywhere from +/-0.001 to +/-0.01 AA in the xx, yy,
> and zz directions.  For the attached configuration, the numerically
> computed stress in the xx directions appears to differ from the
> analytical result by almost two orders of magnitude, and is off by ~100%
> in the yy and zz directions.
> 
> Are there any known issues with the stress tensor calculation?  What
> systems were used to test the stress tensor calculations?  Perhaps I can
> run some
> tests with the same input geometries in order to help figure things out.
> 
> Thanks,
> 
> Nir
> 
> -- 
> Nir Goldman
> Chemical Sciences Division, PLS
> Lawrence Livermore National Laboratory
> P.O. Box 808, L-282
> Livermore, CA 94551-0808
> Phone: (925) 422-3994
> Fax: (925) 424-3281
> https://www-pls.llnl.gov/?url=about_pls-scientific_staff-goldman_n
> 

-- 
     Dr. B. Hourahine, SUPA, Department of Physics,
   University of Strathclyde, John Anderson Building,
          107 Rottenrow, Glasgow G4 0NG, UK.
    +44 141 548 2325, benjamin.hourahine at strath.ac.uk
   The University of Strathclyde is a charitable body,
registered in Scotland, with registration number SC015263

-------------- next part --------------
A non-text attachment was scrubbed...
Name: calcStress.py
Type: text/x-python
Size: 4213 bytes
Desc: not available
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20101011/ab3b0a70/attachment.py>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: geo.gen.template
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20101011/ab3b0a70/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: dftb_in.hsd
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20101011/ab3b0a70/attachment-0001.ksh>