[DFTB-Plus-User] Correctness of SCC convergence in DFT+NEGF

[Gabriele Penazzi]

Hi Anusha,

the Poisson box may indeed play a very important role in the convergence 
path.

When everything is properly setup, equilibrium calculation are very 
often similar to a periodic dftb+ calculation, an SCC cicle of 14 steps 
is fine (a bit fast, even).
Hundreds of steps is definitely long and I would be suspicious as well. 
Also the SCC error gets very large.

My experience is that a small Poisson box (along any direction) can be 
responsible for it. The mixer is quite robust, therefore I observed that 
with a small Poisson box you can still reach convergence to the exact 
solution but with very long and weird SCC steps. I suggest you to check 
that, in case of non-periodic calculation, you have enough vacuum around 
the device (if you can reach it computationally, a couple of nm on each 
side for each direction). The minimal grid step default is usually fine, 
at worst try to reduce it slightly (0.3).

I rarely observed convergence to the trivial zero solution, i.e. all the 
electrons are shot away. I suggest you to check the detailed.out and 
verify that the charge population is reasonable.

Hope this helps
Gabriele


On 06/20/2014 03:58 PM, anz118466 at cse.iitd.ac.in wrote:
> Dear all,
>
> I have simulated two testcases on the DFTB+NEGF. Both consist of a
> biomolecule on a GNR. I observed two very different trends in SCC
> convergence while simulating for the transmission spectra, which I am
> sharing here.
>
> The first case consists of 1795 atoms in total and saw an extremely fast
> convergence in just 14 steps. The second one, however, had 604 atoms and
> saw convergence in around 600+ steps. Also, the SCC error in the second
> case increased gradually first and then decreased rapidly. For example,
> the SCC error reached a high of 0.21776430E+03 in 402 steps. It then
> gradually decreased to 0.31423479E+02 in another 200 steps, and fell
> sharply to 0.15680539E-05 in another 5 steps. The value of 'SCCTolerance'
> was kept 1e-4 in both the cases.
>
> Looking at these two trends, I am confused with the correctness of the
> final output. How do I ensure that no numerical noise has crept in any of
> my two testcases? Also, I observe that the final output changes by
> modifying the poisson box dimensions. Is there any rule of thumb regarding
> the choice of the poisson box? Please let me know if you need more data
> from me for further clarification.
>
> Thanks & Regards,
> Anusha Iyer
>
>
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at dftb-plus.info
> http://www.dftb-plus.info/mailman/listinfo/dftb-plus-user
>


-- 
Dr. Gabriele Penazzi
BCCMS - University of Bremen

http://www.bccms.uni-bremen.de/
http://sites.google.com/site/gabrielepenazzi/

phone: +49 (0) 421 218 62337
fax: +49 (0) 421 218 62770