[DFTB-Plus-User] Failed geometry optimisation for Au atom cluster

[Tiem Leong Yoon]

Dear colleagues of DFTB+,

We have posted this problem before and received some feedback. Since then,
we make some progress and now we re-post the problem and request for some
input...

We have constructed a Slater-Koster file for gold (Au) nanocluster and run
on dftb+. The Slater- Koster information is the courtesy of Prof. Koskinen,
author of the paper "Density-functional based tigh-binding study of small
gold cluster", P. Koskinen et al 2006, New J. Phys. 8, 9.

First we try with 4 and then 5 atoms. The initial structures are obtained
from molecular dynamics simulation. All the structures are optimized through
optimizer like Conjugate Gradient, in the DFTB+. It is found that the
geometry will be optimized in less than 100 cycles.

But when we try to do with cluster of 10 atoms, the geometry optimization
failed as after 1000 cycles, the geometry still did not optimized. Of
course, each cycle of geometry optimization is done with self-consistency, i
believed.

The above optimization is done with "threshold value of force component" of
1e-4. However, if we raise threshold value to 5e-4, all the geometry(10
atoms of Au) optimization are achieved.

We also try with 38 atoms of gold cluster. The geometry optimization is
successful if we use threshold value of force component of 6e-4. For your
info., for 38 atoms of Cu cluster, the geometry optimization is successful
for  threshold value of force component of 1e-4. 1e-4 is the standard value
used in DFTB+ .

We try with threshold value of force component of 1e-4, and try the below
mentioned, and the geometry optimization failed after 1000 cycles:

1) we implement LDA+U corrections by inserting function "OrbitalPotential"
available in DFTB+.

2) we raise the fermi temperature from 0 to even 500.

3) No spinpolarization, i.e spinpolarization = {}.  "

So, our question is, is this problem DFTB+ -specific?

thanks

regards
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20110531/5a77262b/attachment.htm>