The ABINIT Users Mailing List ( CLOSED )

[Paul Fons <paul-fons@aist.go.jp>]

I thought I would relay a simple rule that MSI uses in their Materials Studio.  They calculate the number of extra bands (beyond the valence electrons/2) as a proportion -- I think they add 10% more.  As far as I can see the calculation time difference is really minimal.  This is especially true in light of the fact that the SCF will converge more slowly if you get it wrong.  

Paul

On Apr 18, 2007, at 10:17 PM, Anglade Pierre-Matthieu wrote:

Hi,

Those two things are more or less unrelated :

1) occupation numbers over 2 come from things like "cold smearing"

(option occopt > 3)

2) Likely if you make a computation and the last occupation number of

any band is 0.6 you will have to increase the number of bands. Yet for

some system it is possible that there is no issue with such last

occupation number as the one you report.

Regards.

PMA

On 4/18/07, ha sa <ha_sadi@yahoo.com> wrote:

Hi,

I read the emails between you and Hongyi Zhao, and have a question about it.

 In the tutorial ouput files (e.g. t46.out, sequential version 4.6.2)

occupation numbers at different kpoints sometimes a little bit exceeds 2

like 2.11 or 2.004...etc. and the occupation of the last band is not exactly

zero, but less than one 0.6, 0.0043,...etc. does that mean there isn't

enough bands or what?

Regards

Hanan

Anglade Pierre-Matthieu <anglade@gmail.com> wrote:

Sorry,

I see that you may not know how to know that the number of band is correct:

Basically if there is too few, Abinit will warn you.

You will also be able to check by your own eyes:

Abinit writes the occupations numbers of the different bands at each

k-points. If the last band included in the computation at any k-point

is occupied (non zero occupation number) you need to increase the

number of bands.

Note that too much bands are harmless. They will just increase you

calculation time.

regards,

PMA

On 4/15/07, Anglade Pierre-Matthieu wrote:

> Hi,

> Setting the number of bands is usualy very easy.

> Basically you need one band per "independent" electrons plus a few empty

bands.

> If your system is insulating you can use a very small amount of empty

> bands. For metals with d electron you usualy need many more.

> Basically for most system you can use

> nband = nelectron / 2 + nion / 2 + 2

> or for spin polarized computations

> nband = nelectron + nion + 2

> If during calculations you see that you don't have enough bands you

> can increase nband.

> If you see that you have too much of them you can decrease nband.

> Just try.

>

> regards,

> PMA

>

> On 4/15/07, Hongyi Zhao wrote:

> > Dear all,

> >

> > I'm bewilder by the setting of the *nband*, I mean, when I

> > perform a specific computation for my model, how can I know the number

> > I used for *nband* is suitable for my computation?

> >

> > Thanks in advance.

> >

> > ---

> > Hongyi Zhao

> > GnuPG DSA: 0xD108493B

> >

> >

>

>

> --

> Pierre-Matthieu Anglade

>

--

Pierre-Matthieu Anglade

 ________________________________

Ahhh...imagining that irresistible "new car" smell?

 Check out new cars at Yahoo! Autos.

-- 

Pierre-Matthieu Anglade