[Pw_forum] Wild phonons

[Nicola Marzari]

Dear Miguel,

is the q=0.1 0.1 0 the result of a direct phonon calculation,
or the result of an interpolation ? If it's a matdyn interpolation,
I wouldn't be surprised at all - small changes at q=0 can really affect
your interpolation.

If it's a direct calculation, my only comment would be that a change 
from 0.02 to 0.15 is not that small (with smearing, you should think
at the inverse - i.e. it's much easier to converge with respect to
k-points when you have 0.02, that when you have 0.01).

This doesn't really solve your problems, of course, and the only thing
you need to figure out is if 26x26x26 and 0.02 is good, or not - and
the only way is to increase the k-point, and making sure htat things do
not change, and then decreasing the smearing, and retesting (painful).

I've seen samplings as high as 40x40x40 to get the elastic constants
right in Nb, from the sound velocities.

My experience is that it is less expensive to get elastic constants
from the energy than form the phonons, but once you really ramp up
the k-points (as said, even 40x40x40, and you are converged with the
smearing - not too small, not too large) the results are the same.

Note that at finite pressure one might need to be careful about
energy and enthalpy and what is exactly that you are calculating -
my guess is that you need the second derivatives of the enthalpy, and I
suspect that is what the phonon slopes would be, but it would be better
to check carefully the literature. Anyone care to comment ?

			nicola



Miguel Martínez wrote:
> Dear colleagues,
> 
> I've been working on high pressure phonon branches of alkali metals
> lately. Having all stuff calculated with PBE, I obviously wanted an Na
> PBE pseudopotential. My choice this time was Axel Kohlmeyer's
> Na.pbe-nsp-van.UPF.
> 
> The thing is, at 20 GPa, even with ecutwfc=60 Ry and ecutrho=600 Ry, on
> a 26 26 26 monkhorst pack grid, a change of degauss from 0.02 to 0.015
> makes the low frequency mode in the [110] branch unstable. I really
> can't imagine a 26 26 26 grid being too coarse for a notable jump on a
> subtle smearing change. At q= 0.1 0.1 0, the results are: 
> 
> degauss = 0.02 Ry
>      omega( 1) =       0.187580 [THz] =       6.257034 [cm-1]
>      omega( 2) =       1.178767 [THz] =      39.319706 [cm-1]
>      omega( 3) =       2.672894 [THz] =      89.158739 [cm-1]
> 
> degauss = 0.015 Ry
>      omega( 1) =      -0.165534 [THz] =      -5.521640 [cm-1]
>      omega( 2) =       1.183868 [THz] =      39.489864 [cm-1]
>      omega( 3) =       2.673752 [THz] =      89.187356 [cm-1]
> 
> This being Na bcc at 20 GPa, having checked that its enthalpy is lower
> than fcc Na with that pseudopotential, I'm worried to see such a jump in
> the mode related to C'. I know that I'm close to gamma, making
> convergence hard. But on the other hand, both the energy cutoff and the
> k-point mesh are pretty large. 
> 
> If anybody has any comments, I'd gladly recieve them. 
> 
> By the way, is it too crazy of an idea to calculate the elastic
> constants from the phonon spectrum instead of directly deformating the cell?

-- 
---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu