[Wien] Right Track

Peter Blaha pblaha at zeus.theochem.tuwien.ac.at
Tue Apr 20 15:26:23 CEST 2004


> So should I kill the job or continue since is started 16 days ago.

Kill your job.

Restart from scratch with the default NPT,...

> RMT of H is 0.5

Sphere sizes are VERY crucial for organic molecules. Choose H about 1/2 of
the C sphere (e.g. 1. and 0.5, when nn is 1.5-1.55; when the distance is
larger adapt the spheres. 10% larger spheres (1.1 and 0.55 if nn-distance is
1.65) will give a LARGE reduction in cpu-time. Make them almost as large as
possible!!!!

> I have got Gmin > Gmax I changed Gmin to 2 and Gmax to about 16.

How did you change Gmin to 2  ??? It works only indirectly via RKMAX in
case.in1. What is your RKMAX ? For Your H-atoms RKMAX=3-4 is enough.

GMAX should be larger! use GMAX=20 for H-containing systems.


> k point: I put 1500 in the box. But when I started the job I have got
> job started in parallel mode
> hostname1 [85]
> hostname2 [85]
> hostname3 [85]

Think about the physics behind: If you have a metal (and small unit cell)
you will have a large k-dispersion in your bandstructure (i.e. the
eigenvalues will differ a lot at the different k-points) and you need many
k-points. For an insulator with LARGE unit cell, the k-dispersion will be
VERY SMALL (maybe even completely negligible). Thus you need only very few
k-points (maybe just one, but if you have 3 PCs, try to get a final mesh
of about 3 k-points). If you use just 1 k-point per processor and not 85
you will gain a speedup of 85!!!!




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




More information about the Wien mailing list