The ABINIT Users Mailing List ( CLOSED )

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

Hello Olivier

> 3) when increasing the value of ecut from 17Ha to 26Ha (which seems more 
> appropriate for V based on GS energy convergence study), with the 
> default iscf=5 (and all other parameters identical to t11.in), the 
> response-function calculations converge ok, if I raise nstep to 50. But 
> the phonon dispersions display some anomalous behavior with transverse 
> acoustic branches showing begative slopes at Gamma, leading to negative 
> frequencies from Gamma to H and Gamma to N.
Ok, so you need at least 26Ha. Have you done a full convergence test on 
ecut with the pseudopotential in the GS? V is a hard case, and the cutoff 
for response function convergence is higher than for GS.

> 4) I have tried to increase the kpt mesh from 8x8x8 to 16x16x16 and 
> (independently) to decrease the value of tsmear from 15mHa to 5mHa, but 
> both lead to a failure of the convergence (using default iscf=5) as in 
> 1)
Hmmm, try increasing ecut more. You'll have to see if you really need all 
those kpoints (you may). When you say it doesn't converge, is that the GS 
or the phonons? Do you have preconditioning on (should if it worked for 
Nb)?

> -what parameters are likely to improve on the negative 
> slopes/frequencies I am getting (case 3 above)?
The negative frequencies come from the breaking of the acoustic sum rule - 
translational invariance - due to the potential in real space (several 
mails in the archive on this). It depends on the pseudopotential, and can 
be either:

a) eliminated by increasing the cutoff (sometimes by quite a lot, though)

b) corrected by imposing the acoustic sum rule

If there are no negative frequencies, the branches should stay positive. 
Sometimes for wierd phonon BS the interpolation scheme gives wiggles near 
gamma (branches start at 0 and then go down instead of up). Either:

1) add more qpoints, or

2) you can truncate the interatomic force constants in real space to a 
radius rifcsph (anaddb variable or nsphere) - large R IFCs correspond to 
small q phonons. This is kind of dirty since you are throwing away 
precious information for long wavelengths, but oh well. Plus you know what 
the bands are like at gamma (all acoustic). The value for rifcsph has to 
be tested (should be fast with anaddb): start with 100 or more and 
decrease. I'm not sure the maximum radius for the IFCs is printed out by 
default - see the analysis of the IFC and the anaddb variables page

http://www.abinit.org/Infos_v5.0/users/anaddb_help.html



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