The ABINIT Users Mailing List ( CLOSED )

[Masayoshi Mikami <mmikami@rc.m-kagaku.co.jp>]

Hello,

In principle, Yes, you can follow test samples such as the 
Test_v2/t30-32
(Dataset 1 to Dataset 4). You can modify the samples for your cluster 
case.

Before running calculation, you should have fully geometry-optimized 
cluster...
As you said, you will need d/dk perturbation result,
then you will also need "rphon 1" as well as "rfelfd 1" for Born 
effective charge.
(Born effective charge is the mixed perturbation with respect to ionic 
position
as well as electric field)
(still, you will have negative frequency modes ... it will not be 
avoided,
as discussed in this forum ML.  annaddb can be practically used, just 
as t32
of Test_v2.)

About the k-point choice: it will depend on your Brillouin zone shape 
and size.
I have not yet tried RF calculations (with respect to just electric 
field )
for molecular clusters, but I tried for some molecular crystals.
(Our work will be published in Phys.Rev.B this month.)

My experience tells :
1) if a system has a large unit cell with low symmetry, single k-point 
sampling
may be all right. Dielectric tensors appears fairly converged.
2) if a system has a small unit cell with high symmetry (e.g. C60 fcc 
solid),
results will depend on the choice of single k-point will change.

In your case, I suppose that you use a large cubic supercell.
So, your case may be dependent on your choice of the sampling.
You might want to try with (1/4, 1/4, 1/4) as well as Gamma-point.
You can refer to Chadi-Cohen's paper as well:
Phys. Rev. B 8, 5747–5753 (1973)
http://prola.aps.org/abstract/PRB/v8/i12/p5747_1
You might want to try more k-point samplings to check the convergence;
Kindly note that you can take into account only the time-reversal 
symmetry
for k-point sampling in RF calculations at q=(0 0 0).
(The size of supercell might be also relevant... you could check this 
also !)

Good continuation,
Masayoshi

On 2004/04/08, at 0:49, andersen@physics.ucdavis.edu wrote:

> Can ABINIT calculate the Born effective charges for a molecule? I'm a 
> newbie, but it looks like the procedure is to first calculate d/dk. 
> Can this be done if there is only 1 k-point? Does anybody have any 
> experience (maybe literature references?) using the linear response 
> capabilities of ABINIT to study a cluster?