[Wien] Multicore on i7 iMac

[pluto at physics.ucdavis.edu]

Dear Prof. Blaha,

Thank you for your comment, it helps.

My slab was indeed wrong not having the inversion symmetry. I have now
constructed the slab with the inversions symmetry (was found automatic by
nn and sgroup), and the calculation is running. Actually you have
explained that to me already in 2008 (see below)... so I'm a bit
embarrassed :-)

The calculation is for Fe slab. UG says, that it is sufficient to do SCF
without spin-orbit, and then initialize spin-orbit and do one single SCF
iteration with spin-orbit. Or should I converge SCF again after including
spin-orbit? In any case, once I get the slab running properly, I am
planning to test and compare the results.

Regards,
Lukasz




-------- Original Message --------
Subject: 	Re: [Wien] Fe slab
Date: 	Fri, 04 Jul 2008 14:56:26 +0200
From: 	Peter Blaha <pblaha at theochem.tuwien.ac.at>
Reply-To: 	A Mailing list for WIEN2k users <wien at zeus.theochem.tuwien.ac.at>
To: 	A Mailing list for WIEN2k users <wien at zeus.theochem.tuwien.ac.at>


Why would you use such a struct file ?

a) With limited experience, start with small models, 5 or 7 layers only.
b) I don't know how this struct file was created, but for sure a 15
layer Fe(001) slab can have inversion symmetry and I'm pretty sure that
WIEN should be able to find the proper symmetry (sgroup) when you allow
for it.    Remove ALL numbering for atomes (Fe1,2,3,...) and run the
initialization. sgroup (or nn in most cases) should always group 2 atoms
together (make them equivalent, except the center). sgroup should also
shift the atoms along z.
Your calculations will be 4 times faster when you have inversion symmetry!
c) When going for thicker slabs, you should also improve your vacuum. It
does not make sense to go to a thick slab, but have surface-surface
interactions through the vacuum.
d) k-mesh: why would one use a "2" fold k-mesh in z-direction. With
recent WIEN2k versions you can also specify a 21x21x1 mesh for kgen.
e) Some cases may need more than 40 iterations. As long as it does not
diverge, just continue.
f) Eventually TEMP with some broadening (0.005) may help convergence.
However, in particular with magnetic systems, make sure that the
broadening does not influence your magnetism and recheck with smaller
broadening.