[WIEN] How to speed up slab calculation!

[Peter Blaha]

> 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 would estimate for a 13 atom slab you need a metrix size of about
1500-2000. Thus one k-point takes on a modern P4 30-40 seconds.
(case.output1). Your PIII
might be 3-4 times slower, makes approximately 2 minutes /k-point.
For your slab you may need 5 k-points ? makes 10 minutes and lets
give it another 10 minutes for lapw0 and lapw2.

Thus I would estimate 20 minutes /iteration; 15 iterations ==> 5 h for a
scf calculation. When you need 5 days, you must do something strange.

What are your sphere sizes ? How large is the nn-distance ?
RKMax ? number of k-points ?

PS: You should not expect that a all electron calculation is faster than a
PP, but it should not be that bad.

PPS: The compiler is not so much of a problem, but the proper BLAS library
can speed up things with factors 2-3. DON'T use -lblas_lapw !
Use an ATLAS library or change to the ifc+mkl.

Regards and good luck.

                                      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/
--------------------------------------------------------------------------