[Pw_forum] inhomogeneous K-point sampling

[Aritz Leonardo Liceranzu]

Hi all,

I wrote a little program that computes the binding energies of  
excitons for different semiconductor systems like GaAs, GaN CdSe ...etc.

Starting from equilibrium KS DFT wave-functions and making use of  
linear response theory exciton binding energies maybe calculated.  
Basically one has to digonalize the momentum dependent  
exchange-correlation kernel F_kq(r,r',w).

To test the convergence of my results in terms of the k-point sampling  
I would need to increase the number of K,q points considered to build  
the above matrix. Considering that the code I wrote is serial and is  
far from being computationally optimized the number of k points  
considered makes a big deal in terms of computation time.

On the other hand I have the feeling that for the description of  
excitons it is important to consider "many" kpoints in the  
surroundings of the Gamma point where the valence band max. and the  
conduction bands minimum are. But no so many far from Gamma.

Of course, I could always use a huge k grid but that is  
computationally impossible for me.

So, the question would be, is it stupid to do the following?

Take a huge grid, let us say (14,14,14) and get rid of points that are  
at farther than a certain threshold from Gamma, normalizing the  
weights of the "new" grid so that they sum one.

Thank you very much for your help and feel free to say that it is a  
stupid idea.

Aritz