ABINIT Discussion Forums

Dear all,

When we followed the instructions of GW lesson1 studying rutile and wurtzite ZnO, everything seemed fine. The problem is with nband number convergence (section 5 and 7): the results do not (energy) change with the band number variation, which happen to both cases. However, all other convergence studies worked perfectly. We thought it might relate to the suing of PAW, so we did exactly same to the Si case as shown in the tutorial, using Troullier-Martins pseudo-potential, and same things happened.
We are wondering if it has anything to do with the way Abinit installed: we are using 6.6.1 version with Intel 11.1/073 compiler, aggressive optimization, mkl, and DFT/TRIO fallbacks, no double-precision, on Ubuntu 10.04.
Any insight will be greatly appreciated!

Regards,
Rundong

Dear all,

It seems that the nbandkss variable is not working, since we have tried different scenarios like putting it =-1/positive number, but the nband was used instead. Could anyone shed light on this? Thanks in advance.

Regards,
Rundong

Dear Rundon
Are you using kssform =3? If this is the case, you may need to set nband for the KSS file preparation to the same value as nbandkss. This will give the number of bands stored in the KSS file to be equal to the nbandkss(so that the convergence study of GW with respect to the nband is possible). The reason to this is as the following message I got from the forum a long while ago.

Hope this help.
Thanusit

--------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> Dear all Abinit members
>
> I'm trying to prepare the KSS file for GW calculation of Si-band, with the parallelized abinit version 5.4.3. The
input for KSS generation is as below:
>
> --------------------------
> # Dataset2: calculation of kss file
> nkpt2 19
> nshiftk2 4
> shiftk2 0.0 0.0 0.0
> 0.0 0.5 0.5
> 0.5 0.0 0.5
> 0.5 0.5 0.0
> istwfk2 19*1
> iscf2 -2
> getden2 -1
> nband2 10
> nbandkss2 100
> kssform2 3
> tolwfr2 1.0d-10
> --------------------------------
>
> The calculation was completed. I found that the number of bands stored in the kss file was 10 instead of 100 as
specified by nbandkss2. The part of the output looks like this:
>
> ======================================================================
> Calculating and writing out Kohn-Sham electronic Structure file
> Using conjugate gradient wavefunctions and energies (kssform=3)
> number of Gamma centered plane waves 507
> number of Gamma centered shells 41
> number of bands 10
> maximum angular momentum components 3
> Conjugate gradient eigenvalues
> k eigenvalues [Hartree]
> 1 -0.2183 0.1626 0.1824 0.1824 0.2841 0.3474 0.3474 0.3664 0.4523
> 0.5137
> ......
>
> However, as I set the kssform2 to 1, the number of bands stored in the kss file was 100 as expected. My points are:
>
> 1. Is this the usual result when we set the kssform = 3?
Yes. When kssform is 3, the code uses the conjugate gradient to converge the states
specified by nband, not nbandkss. When the wavefunctions are converged below tolwfr,
the SCF cycle exits and nbandkss states are written on the KSS file.

In your case you are asking for 100 states for the KSS, but only 10 bands are available.
Abinit is smart but it's not yet able to create converged wavefunctions from nothing,
we are still working on it

To obtain 100 bands in the KSS file you should use nband2=100 and nbandkss2=100.
Moreover it's worth to add a buffer during the SCF run; you might try:

nband2=105
nbandkss2=100
nbdbuf 5 #see http://www.abinit.org/Infos_v5.5/input_ ... tml#nbdbuf


> 2. In the convergence study of GW corrections with respected to nband, we need to vary the nband to some value beyond
10. Suppose the number of bands stored in the KSS file is 10, can we still do so (set nband above 10) without getting
incorrect results?

If the number of bands found in the KSS file is less than the number
of states specified by nband, the code prints a WARNING and proceeds with the available states.
In your case the results are not wrong, only not converged and completely unreliable as few bands are employed.

Hope this helps,
Best Regards
Matteo Giantomassi

> Besr regards,
> Thanusit Burinprakhon

Dear Thanusit,

Thank you very much for the information. yes, we are using kssform=3, and we have found out that if nband is set high, we can avoid this problem. However, your information lift a the heavy stone from our hearts, since we know what we are doing is correct. Do you mind send me your email address so we can discuss further related issues? Mine is rundongwan@yahoo.com.cn.
Thanks again.

Best regards,

Rundong

Dear Rundong
Great to hear that you have found the way to get round the problem and my information help to confirm your confidence. It's no problem at all, if you would like to discuss further related issues personally. Here is my e-mail: thanusit@kku.ac.th.

Best regards,
Thanusit