[Wien] non-rectangular cell

[Peter Blaha]

> digits in storing variables ?  For example, does a charge distance
> :DIS=0.00000001 make sense ?

NO, absolutele nonsense !!!!! You must have lots of computer power and
just waist it.

a) You should converge your "quantity of interest" to a meaningfull
accuracy.
b) Quantity of interst: maybe just a bandstructure or DOS (usually
does not need very strict convergence), maybe a total energy (when you do
structural optimizations,..) or a force (for minimizations F of 1-2
mRy/bohr is more than enough, since the atoms do not move significantly
anymore with such small forces), or a magnetic moment (usually X.yy is
fine, unless for a difficult case, where the moment changes only very
little from one iteration to the next...

c) Don't forget: it does not make sense to "overconverge", because it will
still be an arbitrary quantity. Just think what happens when you change
the k-mesh or increase RKmax. It will give you a jump in :DIS and all
resulting quantities will change again. So when you converge a moment for
one set of parameters to X.yyyyy (:DIS=0.00001), by changing k-mesh or
RKmax you may get a moment which differs already in the first or second
digit after the comma.

d) It is MUCH more IMPORTANT to check the results with respect to RKMAX
and K-mesh, than with respect to scf-convergence.


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