[DFTB-Plus-User] need help about convergence of SCC-calculation

[Bálint Aradi]

Dear Lin,

>  I was trying to optimize some charged systems containing several hundreds
> of atoms. But the calculations are very hard to converge. I tried to use 2nd
> order MP scheme with higher electronic temperature ( 1000 Kelvin). The SCC
> calculation converged well at first but failed quickly after several CG or
> Steepest Descenct steps. I am not sure the relaxed configuration is really
> the correct one with such a high temperature. I also tried to using other
> mixing schemes such as Anderson and DIIS. Both of them doesn't work. It is
> appreciable for your help.

  Running your calculation myself, it indeed results in a geometry, for
which the SCC with your settings fails. Please note, that due to the
huge amount of charges you put onto your cluster (ca. 0.3/atom) the
outer shell of it formally explodes. This results in geometry with
covalently bonded atoms (core of the cluster) and isolated atoms around.
This is usually in every SCC/SCF method not trivial to converge.

 Starting from the failed geometry, and changing some settings, I could
let it converge further (I stoped it after 4 geometry steps). I changed
following things:

1) SCCTolerance set to default value. Your value 1e-6 is very tight,
that would result in an energy convergence of about 1e-6 -- 1e-7 eV,
which is just too tight, numerically unstable, and physically not
justified anyway.

2) Using the Broyden mixer with MixingParameter = 0.3. To my experience,
for almost all problems the Broyden mixer is the most reliable one.

3) Using Fermi filling instead of Methessel-Paxton and using moderate
tempereture of 300K.

 With those settings, things converged apart in less than 30 SCF-steps
(with excepction of the 1st geometry step, needing 74 steps). Please
note, that settings MaxSCCIterations to 2000 does not make much sense.
Usually even difficult systems converge with the right settings within
30 SCC steps in DFTB. What you in most cases end up after that, are just
charge oscillations between subsequent SCC-steps, and from that point
the system will never converge. It is usually advisable, to tune the
parameters, if your system do not converge within 200 iterations for the
1st step, and let's say, 50 iterations in the subsequent ones.

 I hope this helps.

  Best regards

  Bálint


-- 
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
http://www.bccms.uni-bremen.de/en/~b_aradi


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 252 bytes
Desc: OpenPGP digital signature
Url : http://www.dftb-plus.info/pipermail/dftb-plus-user/attachments/20100527/90299305/attachment.pgp