The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

Dear Linhui,

fdsa fda wrote:
> Dear all,
>
> I'm trying to apply phonon calculation for a difficult
> case, which are carbon nanotubes. Before I begin,
> there is a question on the accuracy of the expected
> result:
>
> To my understanding, when applying the linear response
> method to calculate purtabations like phonon, it is
> crucial that the EXACT atomic coordinates on
> unperturbed geometries are determined first by
> structural relaxation run. Am I right?

Not exactly. You will get an answer based on the
second derivative of the total energy around that
starting structure. It will differ from the one at the
exact equilibrium geometry, but the effect of the
discrepancy will depend on :
(1) the distance between your
structure and the real optimized one, as well as
(2) the physics of the problem (the rate of change of
second derivative with respect to atomic
displacement).

In some case, the effect will be minor, and
in other cases, especially when you have low-frequency
phonons, the influence will be large.

> But to my experiences, for systems like carbon
> nanotubes(several tens of atoms in unit cell), the
> accuracy of atomic coordinates can be determined no
> better than about 0.01 Angstrom by structural
> relaxation run. 

Seems quite large ! You might have to work harder ...
try different convergence options ...

> Is it justified to make such an
> estimation?
>
> So, suppose we already have an error bar of 0.01
> Angstrom on atomic positions in UNPERTURBED geometries
> of carbon nanotubes, then how large will the error be
> on the final phonon frequencies by ABINIT, or any
> other DFPT calculations, caused by such inaccuracy on
> atomic coordinates?

No idea : you have to do this testing by yourself.
Try to compute the phonons from slightly different geometries, and
examine whether the result is sensitive to the starting point ...

Good continuation,
Xavier Gonze