The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

To choose a pseudopotential is really much more complicated and depends a lot on your project. There is a lot of criterions.

- Look at the ecut you need to get a certain value for absolute
convergency.  This is a nice criterion if you want the lowest
calculation cost. It's very interesting if you are computer limited.
- Look at the publication related to the pseudo. This might be very
important if you want to publish your results. It's always safer to
know on what kind of ground you are going to build.
- Look at the transferability of the pseudo to your particular problem.
If you want to do something very much related to the elastic properties
of your materials choose the one that will reproduce best elastic known
properties. If you want to do some calculations under pressure choose
the potential which seems best fitted to properties under pressure ...

As some  rules of thumbs: 
-usualy potentials with large ecut are more transferable than their counterparts with  lower ecut.  
-I you want to do many theoretical computations some kinds of pseudo are simpler to handles (like HGH)

On 4/24/06, Lu Jianmin <jmlv@imr.ac.cn> wrote:

Dear Mr. Anglade Pierre-Matthieu,

Thanks for your suggestions.

When I changed nband to a much larger value 13/14, the job ran 
successfully, without any ERROR message.

My next question is, among 3 types of pseudopotentials available from the 
ABINIT website: FHI, HGH and TM, which type would you prefer using for 
structural calculation?

The experimental value for acell of B2 TiNi unit cell is  3.010 Angstrom. 
The calculated values for it based on 3 pseudopotentials mentioned above are,
psp. acell(Angstrom) error
FHI 3.0690 2.0%
HGH 3.0721 2.1%
Tm 3.1781 5.6%

It seems like the FHI pseudopotential being the most accurate one among 
three of them. Do you agree? Any comment?

Best regards.

Jianmin Lu,
imr,cas, China

> Since you are dealing with metals at finite electronic temperature the
> band occupancy is not an integer. That is you may have occupancies
> like the following:
> 
> band 0 : 2 e-
> band 1 : 2 e-
> band 2 : 2 e-
> band 3 : 2 e-
> band 4 : 2 e-
> band 5 : 2 e-
> band 6 : 1.98 e-
> band 7 : 0.015 e-
> band 8 : 0.005 e-
> 
> That's the why the formula I've sent adds a few more bands than just
> (nb val. e-)/2.
> I suppose that your problem arises from the lake of empty bands.
> Anyway you'll need to have at least one band with zero occupancy at
> the end of your calculation if you aim at a converged result.

> Pierre-Matthieu Anglade
> 
> 

-- 
Pierre-Matthieu Anglade