[Wannier] wannier WF and Hamiltonian

[szaboa]

Dear Liping,

The total spread is just the sum of all the spreads of the WFs, so it's 
value depends on the number of WFs. Instead you should look at the 
individual WF spreads. It might depend on the system, but in my 
experience they should be in the order of a few (2-3) Angstrom^2 when 
everything goes fine.

Before calculating the Wannier functions, you should calculate the band 
structure of your system, and decide which bands you want to reproduce 
with WFs. You are probably interested in the ones close to Fermi energy. 
If you have a set of bands that are separated from the others, then you 
should use the exclude_bands [n1-n2] command in the wannier90.win file 
to exclude all the other bands.

If it's not the case, then you need to use the disentanglement method, 
where you have to specify an inner energy window, where the bands will 
be identical to the bloch states, and an outer, that's used to 
disentangle the bands. In this case increasing the size of the inner 
window will increase the spread of the WFs, so you should only cover the 
energy range with it that you want to reproduce very accurately.

Yes, the number of the projections must be the same as the Wannier 
functions. One initial projection corresponds to one WF. And the number 
of WF's must be the same as the number of input bands if those bands are 
not entangled.

To make good initial projections, you should look at the primary 
character of the considered bands. In Vasp, specifying LORBIT = 10 in 
the INCAR, it will project the DOS onto spherical harmonics, centered on 
the ions. It will create a file called PROCAR, where you can find your 
bands by looking at the energy and k points. Try to find the ion and the 
orbital type that contributes the most for a given band at various k 
points, and use that as the projection for that band.

Hope it helps,
Aron


On 2014-01-30 21:27, Liping Chen wrote:
> Dear Aron,
>
> Thanks for your quick reply!
>
> About the initial projections, I always have problems. Now I'm doing
> the calculation on SiO2. Actually, I tried different initial
> projections. They gave similar results of "Omega Total  =
> 579.766631183", which means that the WF is extended over 3-4 unit
> cells (the lattice constant lc in three directions is 7.12 Ang). It
> shouldn't be like this, right? My supervisor thinks the WF should be
> localized in one unit cell. I don't know it's correct or not, since
> I've looked at the WF of GaAs in example01 of wannier90. It seems 
> that
> the WF is not localized in one unit cell. If the WF is localized in
> one unit cell, does it mean that Omega Total should be close to the
> value of lc*lc? Is any principle to set the initial projections? It
> seems that num_wann should be consistent with the number of
> projections, since sometimes it says "too many projections" or "too
> less projections". Can you give some suggestions about the initial
> projections? Thanks!
>
> Liping
>
> ----- Original Message -----
> From: "szaboa" <szaboa at iis.ee.ethz.ch>
> To: wannier at quantum-espresso.org
> Sent: Thursday, January 30, 2014 3:02:18 PM
> Subject: Re: [Wannier] wannier WF and Hamiltonian
>
> Dear Liping Chen,
>
> You should add an LWRITE_UNK = .TRUE. line in the INCAR file to make
> vasp produce those UNK files.
>
> Regarding your second question: The matrix elements should be real
> indeed. As far as i know, having nonnegligible imaginary parts 
> usually
> means that the program converged to a fals minimum. You should 
> probably
> specify different initial projections.
>
> Best regards,
> Aron Szabo
>
> On 2014-01-30 19:37, Liping Chen wrote:
>> Dear wannier users,
>>
>> I want to show the wannier WF with xcrysden. I find that there is an
>> example (the first one) in wannier90 tutorial. It seems that the 
>> UNK*
>> files should to be prepared in advance. It's easy to get them by
>> PWSCF. But now I'm using VASP. I've tried to check the WAVECAR from
>> VASP. It seems that the WAVECAR for the coefficients is written like
>> this:
>> ##################################################
>> do ik=1,num_kpoints
>>   do iband=1,nband
>>     do iplane=1,nplane
>>       coeff(iplane,iband) ! write the coefficients
>>     enddo
>>   enddo
>> enddo
>> ##################################################
>> The number of plane waves (nplane) is not the same for all the
>> kpoints (ik). I don't know the relationship the number of plane 
>> waves
>> and the ngx, ngy, ngz written in the beginning of the UNK* files. 
>> How
>> to get the UNK* files from VASP so that I can get the WF from 
>> wannier
>> and then plot it?
>>
>> In the previous discussion of "how to calculate fermi surface with
>> VASP and WANNIER90", it seems that Jonathan said WF should be real.
>> Then should the Hamilton matrix <0n|H|Rm> and the position matrix
>> <0n|r|Rm> in WF basis be real accordingly? But when I print out
>> hamiltonian matrix with "hr_plot = true", it's not real but complex
>> with unignorable imaginary part. Is it wrong or not?
>>
>> Thanks in advance!
>>
>> Liping Chen
>> Department of Chemistry
>> University of Rochester
>> _______________________________________________
>> Wannier mailing list
>> Wannier at quantum-espresso.org
>> 
>> https://urldefense.proofpoint.com/v1/url?u=http://www.democritos.it/mailman/listinfo/wannier&k=p4Ly7qpEBiYPBVenR9G2iQ%3D%3D%0A&r=MhxMznMChIKt4IuapbjnvwFAOKg44L9%2FDsTFLScyNxw%3D%0A&m=dvh1GVVx08Fk0oEASvg2eLCYhLzlSQ47S4ouvCLQWgI%3D%0A&s=a0394c2f48f06646cee39dc908cfece9100e75fbcd14193291a49c29eb8638b9
> _______________________________________________
> Wannier mailing list
> Wannier at quantum-espresso.org
> 
> https://urldefense.proofpoint.com/v1/url?u=http://www.democritos.it/mailman/listinfo/wannier&k=p4Ly7qpEBiYPBVenR9G2iQ%3D%3D%0A&r=MhxMznMChIKt4IuapbjnvwFAOKg44L9%2FDsTFLScyNxw%3D%0A&m=dvh1GVVx08Fk0oEASvg2eLCYhLzlSQ47S4ouvCLQWgI%3D%0A&s=a0394c2f48f06646cee39dc908cfece9100e75fbcd14193291a49c29eb8638b9
> _______________________________________________
> Wannier mailing list
> Wannier at quantum-espresso.org
> http://www.democritos.it/mailman/listinfo/wannier