[WIEN] How to speed up slab calculation!

[Stefaan Cottenier]

> I am running ZnSe(001) slab of 13 atoms (7 Zn and 6 Se) for 4 days now.
> Convergence seems to be fine, but its a long time to wait. It would be
> even much longer when I want to relax the surface (minimize forces).
> 
> I must admit that using a pseudopotential code the calculation was much
> faster, 

This happens in many cases. Look at the slides of this talk :

http://www.ifw-dresden.de/agtheo/FPLO/lectures03/cottenier.ps.gz

where you will find a.o. a speed comparison between VASP and WIEN2k (for an 
interface slab, if I remember well). It shows the same dramatic speed 
difference between pseudopotentials and APW+lo (which is fortunately not 
always so unfavourable). I'm afraid there is not much you can do against this.

> I am pretty sure there are ways to speed up WIEN calculation by making
> it less accurate - I would like to ask for some advise which parameters
> could be changed to make it faster?

The -nohns option in the first iteration and the -it switch are things you can 
try first, without giving up accuracy.

Next, you can reduce k-mesh (x kgen) and basis set size (RKMAX, top of 
case.in1). But test first on a simple but similar slab how low you can go 
before going into nonsense.

> Maybe my problem is also hardware since I am using serial code with
> Linux on PIII 1Ghz, 2 GB RAM. But I was using the same machines with
> pseudopotential code.
> 
> I used pgf90 to compile WIEN.

Did somebody test if there is a compiler related speed gain between pgf90 and 
ifc ? For some codes, the compiler alone can give a speedup of a factor of 3.

Do you link with ATLAS or mkl ? There is another factor of 2 there.

Stefaan