[Wien] LDA-1/2 in Wien2k?

[Peter Blaha]

Hi,
I briefly checked your paper. It sounds interesting.

There's of course no copyright violation. The reason why we distribute the
source code is to encourage people to advance science and make changes in
in the code.

I'll try to answer your questions below:
> 1 - The entries of the Fourier expansion in the file case.vns are the 
> Fourier coefficients themselves, or their complex conjugate?
These are the Fourier coefficients.
> 2 - The expansion is such that its value inside the spheres are zero or 
> very small? That is, integrals of the potential outside the spheres can 
> be made by integration in the whole cell?
Exactly. The series contains a convolution with a step function, so that the
series is zero inside the spheres.
> 3 - Is there a constant multiplying the Fourier coefficients?
I don't think so.
> 4 - The expansion is in "stars", as defined in the book of Singh?
Yes.
> 5 - When constructing the potential, is it wholly displaced to fix the 
> zero of energy? Where is the energy zero?
The energy zero in an infinite solid is "arbitrary". In WIEN2k it is chosen such that
the average Coulombpotential in the interstitial is zero. (V_coul(K=0)=0)
>     Let us say that, in trying to understand how the potential is 
> written in the files,
> we made simple tests adding a constant inside the spheres and the same 
> constant outside. Of course we were hoping to obtain the same 
> eigenvalues, perhaps displaced by a constant. That was not what we got.
How did you "add" a constant in the interstitial (to case.vns). You cannot simply
add a value to V(K=0), because this would add it "everywhere", not just in the
interstitial region. This "constant" must also be convoluted with a step function.

> PS - We think we understood the file case.clmsum containing the charge 
> density. Thus we could take our potential, calculate its Laplacian and 
> add the resulting charge density to the file case.clmsum (our charge 
> density has zero total charge). But there is a problem: the 
> exchange-correlation of our effective density would also be added. So 
> this procedure would only work if you taught us to run lapw0 setting no 
> exchange-correlation. The potential generated would be the one to be 
> added to case.vsp and case.vns.
-- 
I would not do it that way. But I guess you need to work (and understand)
lapw0. In this program the potential is calculated and modifiying this program
ensures that you do things "right".

Unfortunately, our source code is often quite ugly and not well documented,
but in lapw0.F there are calls to xcpot1 and xcpot3, where the xc-potential
inside spheres and in the interstital is calculated. When you "add" your potential
there, it should be quite ok.
After those calls, the interstital potential is "reanalized" (subroutines rean*),
i.e. convoluted by the step function.

lapw0 has also options to plot the potential, and when you plot the difference
between the original potantial and the one with your modifications, you can easily
"verify" your implementation.

PS: Try to make your changes as "locally" as possible. We are often updating
the sources and it is much easier when you need to insert onyl a few lines into the
original code (i.e. some calls to a subroutine).

Best wishes
                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------