Hi Andrey,
I believe you are correct. However, now the 'steady state solution' that
you come to is limited by the number of k points specified in the
Monkhorst Pack grid. i.e. If your optical mesh is infinitely dense, the
Monkhorst Pack grid limits the accuracy of the integral over the BZ.
In my experience the variations are due to insufficient mesh sampling up
to a large optical mesh, say 45^3, and then the solution starts to
oscillate and doesn't improve with increased sampling. However, I use
0.1eV broadening, which may in itself cause the oscillation and is much
higher detail than one could hope to see in the experimental
permittivity of a bulk material.
I shall have to go and read the code and get back to you for more info.
Cheers,
Marty
A.S. wrote:
Hi Martin,
Thanks for this information. The optical mesh in Siesta is
then simply a set of k-points at which the intraband transitions for
evaluating eps(w) are calculated, and its refinement allows
approximating the volume integral over the Brillouin zone with a higher
accuracy - am I correct?
Suppose, I calculate the properties of a solid using a 5x5x5 Optical.Mesh, then
10x10x10, then ... 60x60x100, and first
I find big variations in eps(w) for between 5^3 and 10^3. Then, as Optical.Mesh becomes denser, I find that eps(w) have converged to
some "steady state solution". Does this mean that the variations
in eps(w) at low k-point density are due to the insufficient sampling rate and as the sampling rate increases, I have a more accurate
value of the integral over the BZ?
Andrey
-----Original Message-----
From: Martin Blaber <[EMAIL PROTECTED]>
Sent: Jul 19, 2007 6:47 PM
To: [email protected]
Subject: Re: [SIESTA-L] Optical.Mesh
Hi Andrey,
For bulk materials a Monkhorst Pack grid of 21^3 is sufficient to stop
the eps(w) for a constant optical mesh (Depending on the system).
Whereas an optical mesh grid of 41^3 vs 45^3 still shifts eps(w) around
a bit for constant Monkhorst Pack grid. The broadening also affects what
can be considered converged. I generally keep the optical broadening
very low as you can always broaden more but never less. However, this
does mean an optical mesh of 45 isn't sufficient for convergence. Hope
this helped.
Cheers,
Marty
---------------------
Martin Blaber
Institute for Nanoscale Technology
Dept. of Physics and Advanced Materials
University of Technology, Sydney
[EMAIL PROTECTED] wrote:
Hi all,
I am trying to find some description of the Optical.Mesh variable
which is more detailed than the one given in the Siesta manual.
Some of my colleagues insist that the convergence of optical properties
eps(w) with respect to the number of k-points in the integral wrt
k over the BZ has to be shown not based on the vaues in the Optical.Mesh but rather based on the values of MonkhorstPack_grid. How can I prove the opposite, i.e. that BZ sampling grid refinement for optical property
calculations must be based on the Optical.Mesh variable?
Thanks,
Andrey Semichaevsky,
CEMOS, ECE Dept., UMass Lowell
________________________________________
PeoplePC Online
A better way to Internet
http://www.peoplepc.com
