The ABINIT Users Mailing List ( CLOSED )

[Matthieu Verstraete <mjv500@york.ac.uk>]

> papers I think thats enough).  The calculations are at a high presssure of 
> about 100GPa.
As mentioned in a recent mail, there can be corrections if you are not at 
equilibrium. Check the doc (I think it's Artem Oganov's paper in 
~abinit/doc somewhere).

> Then I used "anaddb" program to compute the full band-structure 
> by interpolating the IFC's in the fourier space.  I get an accoustic mode to 
> soften and go to negative frequency at a high symmetry point (N which is 
> along cubic [110] with reduced coordinates (0,0,0.5) ).  But if I check the 
> frequencies at the q-points in the grid I found that they are all positive. 
> So I don't understand why after interpolation I can get a negative value when 
> all points on the grid have positive frequencies? So I did a separate run at 
> that exact 'q' point (N) and found that it is not negative. In fact it is 
> highly positive.  Has anyone found similar problems using "anaddb" ?
This is annoying but normal, and has to do with the Fourier interpolation: 
if the phonon bands are not sufficiently "sinusoidal" or cosinusoidal, you 
get spurious contributions at high distances in the real space interatomic 
force constants. FT-ing back to points outside your qpoint grid can give a 
kind of FT aliasing, and in particular you can get negative modes near 
Gamma (or elsewhere for pathological cases, like yours).

* If you increase your qpoint grid, the negative bits will either go away 
or become very localized, and perhaps you can ignore them.

* Or, you can use the anaddb variable rifcsph, which cuts off 
contributions beyond rifcsph (in a.u.) from the ifc. Decrease rifcsph 
progressively and the negative modes will dissappear (magic!), but of 
course you lose detail in the phonon bands.

It's not the same as using a sparser qpoint grid, note! Because all of the 
qpoints you have contribute to all of the ifc, and so to all of the 
interpolations (obviously less so if the interp qpoint is far from the 
original one).


> Any 
> suggestions will be very helpful.  Is there a way to printout the dynamical 
> matrix at an arbtrary q-pt while using the "anaddb" program?
Using the qph1l (as you must already be doing) you can specify any qpoint 
and interpolate to it... You want the actual dynamical matrix itself? 
You'll have to either hack the code or reconstruct it from the frequencies 
and the eigenvectors gotten with the "eivec 1" option

Matthieu


--
================================================================
Dr. Matthieu Verstraete                 mailto:mjv500@york.ac.uk
Dept. of Physics, University of York,     tel: +44 1904 43 22 08
Heslington, YO10 5DD York, United Kingdom fax: +44 1904 43 22 14