On 06/27/2011 03:59 AM, Jan Sommer wrote:
Am 25.06.2011, 11:20 Uhr, schrieb Hongyi Zhao <[email protected]>:
For a bandstructure calculation, you should always fully relaxation
your system to its equilibrium geometry firstly. When you double the
length of your cnt along its axis direction, though this is the same
with your orignal cnt according to periodic boundary conditions, I
think the geometry optimization should be done carefully before the
bandstructure calculation, in addition, the vacuum thickness in other
two directions perpendicular to the axis should also be retested, I
think. As for the others things, the k-points samping density should
be the same, and fro the other convergence criteria shoulbe also be
tested carefully.
Strictly speaking, cnt is a quasi-one-dimensional periodic structure,
not a perfect three-dimensional periodic structure. So when change the
length of it, I think we must firstly ensure the image interaction
between itself should be fully eliminated.
These thinks are clear. The mentioned thing was just a quick and dirty
example as a calculation only runs a few minutes.
The normal cnts I use have mostly more than 100 atoms in their unit cell.
Just because the real system is so huge, the corresponding pre-test for
bandstructure calclations is more important, though it is a
time-consuming and cpu-expensive thing ;-(
Otherwise, the bandstructure you obtained should make no sense.
Why do this? By saying the same cnt with different length, do you mean
you have applied a strain along its axis direction?
I mean how much CNT-Unit-Cells are explicitliy in the siesta input file.
In the example 36 or 72 (corresponding to 1 or 2 unit cells of the
(9,0)-cnt)
I've tried some thought on this issue sometime before. Here I post the
steps I think we should do for this purpose here, just for your information:
------------
For your above issue, i.e., the optimal number of CNT-Unit-Cells used in
the calculation can be determined by the following process:
1. Do a series of convergence test w.r.t. MP grid, E_cutt and so on,
based on the supercell which includes one repeated unit along the
periodical direction. By this way, we can find the convergence
parameters used for follow-up calculations.
2. Based on the convergence parameters obtained from step 1., do a
geometry optimization calculation for the supercell included one
repeated unit along the periodical direction and found the stable
equilibrium structure for this supercell.
3. Based on all of the calculation parameters to obtain the stable
equilibrium structure, we change the numbers of repeated units along
the periodical direction and do a series of single single point energy
calculations for these supercells with different repeated units along
the periodical direction.
4. Finally, we calculate the total energy per repeated unit, i.e.,
E_total/[repeated unit] , and plot this energy with the number of
repeated units and find the minimum repeated units which can ensure the
E_total/[repeated unit] has a relative stable value.
This is just my own point of view. I don't know whether it's
appropriate or not. Any hints/improvements for my above description
will be highly appreciated. Thanks in advance.
------------
If you keep the same cnt under the same external conditions ( strain,
pressure, temperature, and so on ), I think this no need to do a
series of such calculations you mentioned here, instead, just
calculate the bandstructure of cnt including appropriate primitive
unit cells in its axis direction will give you the understanding of
the physical properties of it.
That's true if I would be interested in only the bare cnt.
In future I want to dope the cnts or alter them in general. So in some
of these alterings the primitive unit cell would correspond to a cnt of
"length 1" (36 atoms in the above example), some to one of "length 2"
(72 atoms) and a few maybe even to "length 3".
As I want to be able to compare them together in a joint plot, I could
always multiplicate the smaller systems to the size of the longest so
the brillouin zone always had the same size, but this would result in a
big overhead in computing time, especially as relaxation can take very
long. I just want to make sure beforehand that these things work. That's
why I also use a quite small cnt at the moment, so extra calculations to
test things don't last long.
If you take a look at the image of my last email, you see that the bands
are basically the same. The thing is, that the right and left side of
the red plot are folded inwards in the green plot and I dont really know
how to determine the corresponding points in the data file to unfold them.
Maybe I understand what you mean this time but I'm so sure ;-).
If you mean you have different dataset in the same data file which are
difficult for your to distinguish between them in some ranges, perhaps
you can prepare one datafile for each and then plot them within
different panel and them combine them. It looks like you use gnuplot
to prepare your plot. Gnuplot has many advanced functions which I don't
know, but I think you can find your solutions from its official site or
manual.
If you mean in some ranges the bandstructures have folded together, you
can use use different symbols to plotting them for different bandstructures.
Anyway, you issue sounds like a data post-processing problem instead of
a technical problem for using siesta itself. Not sure if I understand
correctly what you mean.
Regards.
I hope, I could make my problem clearer this time.
Maybe as it is a new and different question I should open a new thread...?
But thank you all very much for your efforts in helping me out.
Best Regards,
Jan
--
Hongyi Zhao <[email protected]>
Institute of Semiconductors, Chinese Academy of Sciences
GnuPG DSA: 0xD108493
