The
structure is that of Mg2Si (3 atoms/cell, Fm -3 m #225). As with AlAs
(F-4
3 m #216) in the tutorial, it is a FCC structure so I expected the
input files
to be the same. Is this right?
should be allright, mainly because the high symmetry points
should be the same, and the details (atoms, acell...) are in the ddb
files, not the anaddb input.
Now I'd like to compute the dispersion curve. I used trf2_5 files as
such for
Mg2Si. In this input file, there are several things I don't understand:
1- If I'm right, at this point abinit interpolates (using the DDB file)
the
dispersion curve between the k-points provided in the trf2_5 file. But
how is
it possible that none of the 31 k-points used to build the DDB are
listed among
those (71) of the trf2_5.in file? In other words, in this case, how
is it
possible to make an interpolation?
The interpolation is an fft one: the qpoints are completed on a
regular grid, then the dynamical matrices are FTed to real space
(interatomic force constants), and then FTed back to arbitrary qpoints
(interpolation). In this way, the final qpoints interpolated to are
arbitrary, and need not contain the original points. Usually they do,
since you start with qpoint grids which are regular and contain gamma,
and other high symmetry points, which are also in your band structure
2-
How are these 71 k-points generated, and in particular, those between
the
special k-points (G, X,L,W,...) ? Is there a rule? a program package?
Any
information (textbook or web link) would be appreciated.
They should be along a path joining high symmetry points in the
BZ (the special points are written in the .in file!). The interpolating
qpoints are are just listed in the anaddb input file. The simplest way
to generate the list is to do an _electronic_ band structure, as in the
tutorial (
t35.in),
and to copy the list of kpoints (eg from the eigenvalue _EIG file, or
the log or output of abinis) to the list of qpoints for interpolation (
http://www.abinit.org/Infos_v5.6/users/anaddb_help.html#qph1l).
This is what was done to make the input file in
trf2_5.in
3- If I compare with the phonon dispersion curve published in the
literature,
the path to draw the curve is different to that proposed in trf2_5.in:
W-L-G-X-W-K vs G-K-X-G-L-X-W-L, respectively. Again, is there a rule to
follow
a particular path dictated by the symmetry or are we free to choose any
path?
The special points are fixed by symmetry, and some paths between
them are more natural or obvious (don't cross the whole BZ, go to the
nearest special point of a given type, try to sample all the paths
between special points of different nature...) Usually the simplest is
to look at the litterature and to reuse the paths they have, for ease
of comparison. eg:
0 0 0 -> 0.5 0 0 -> 0.5 0.5 0 makes sense, whereas
0 0 0 -> 0.5 0 0 -> 0. 0.5 0.5 does not because you are
going from an X point on one side of the BZ to an M point on the
opposite side, instead of staying along high symmetry lines.
Look at the spatial representation of the BZ to get an idea of which
special points are close to each other and make sense for the path. Eg
on
http://cst-www.nrl.navy.mil/bind/kpts/fcc/index.html
for fcc - please check carefully that the kpoint coordinates you use
are the correct reduced ones, and not some type of cartesian ones. They
look ok.
You may also have specific points and directions where interesting
features appear (Kohn anomalies, softenings, kinks), but you'll need to
explore or know in advance where to look.
Now
to create the eps file with band2eps, I also tried to use trf2_6 files
as
such. But it's not working at all. The frequencies are just wrong (not
relevant
to those in the output file of abinit, obviously downshifted so there
are
negative frequencies).
You should read the input file, and adjust it according to your
system (number of atoms, special points, etc...)
You may get negative frequencies because of the interpolation (not
enough qpoints in your grid for interpolation) or because of a breaking
of the acoustic sum rule - search the forum archives for these keywords
for more information.
Could also be that the path given in the band2eps input does not
correspond to the output from anaddb.
Now,
I have 2 questions about trf2_6.in:
1- on the first line, is it the number of atoms/cell or the number of
type of
atoms? (for Mg2Si: setting "2" works whereas "3" fails).
As stated in the file, it should be the number of atoms in the
cell. If it crashes we need more information about the files you are
using etc... and you should make sure all the other inputs in
trf2_6.in
are coherent, as well as the other files you are reading in from.
2-
How do you set the scale factor for each line?
You should look at the cartesian distance between each of the
vertices in your BZ path, and normalize the segments accordingly. If
you just want a qualitative look or the values of the frequencies (as
opposed to the slopes of the phonon bands), you can set them all to 1.
Matthieu