The ABINIT Users Mailing List ( CLOSED )

[PGanesh <pganesh@ciw.edu>]

Dear D. R. Hamann,

Thanks!

Ganesh

D. R. Hamann wrote:
> Dear PGanesh,
>
> Piezo constants calculated with the analytic ddk (rfelfd = 2) will 
> converge with k mesh much faster than with the finite-difference 
> ddk.   The only reason for using the finite-difference approach in 
> this context is to test the agreement of the piezo tensor calculated 
> with density functional perturbation theory with that obtained by 
> numerical strain derivatives of the polarization (which is calculated 
> from ground-state finite differences).  This is all discussed in the 
> "Elastic" tutorial.
>
> Don Hamann
>
> PGanesh wrote:
> > Dear All,
> >
> > There are two methods to compute the first-derivative wavefunctions 
> > (ddk), one using iscf = -3 and rfelfd = 2 and the other using 
> > berryopt = -2 in a non-scf calculation.  The second method uses the 
> > finite-difference formula introduced by Marzari and Vanderbilt and 
> > susceptibility has been show to converge much faster with kpt. grid.  
> > So will I be correct in saying that second-order derivatives as well 
> > (like the piezo) will converge much better using this method as 
> > opposed to method one?  Its probably a moot question, but still 
> > wanted to confirm.  This would probably mean that in the tutorial for 
> > getting non-linear properties, using "tffield_2.in" would give better 
> > piezo. constants than "tnlo_2.in + tnlo_3.in".
> >
> > Thanks a lot.
> >
> > PGanesh
> >
> >
> >
> > >