[Wien] -in1new N

[Stefaan Cottenier]

>   I missed some points in UG about -in1new.  The -in1new N is used   
> for automatic linearization of energies. During optimization of the   
> parameter such as RKmax and k point, is it sufficient to use -in1new  
>  N for only first calculation?

There is certainly no objection against using in1new for all steps in  
your parameter optimization. The better e.g. the basis set, the more  
accurate the best linearization energies will be.

The question is therefore: does it harm to use in1new only for the  
first calculation in such a series? A few wien2k version ago, this was  
impossible, as the in1 file produced by in1new was incompatible with  
the old scheme. I'm not sure, but I think that now they are  
compatible. If that is true, then not using in1new throughout will  
most often not harm, although you lose a tiny little bit of accuracy  
and you don't gain anything.

>   For example, i fixed RKmax is equal 6 and k point varies from 1000  
>  to 10000.
>   For first calculation RKmax=6 and k point=1000, i used -in1new 2.
>   Next calculation RKmax=6 and k point=2000, i didn't use -in1new 2.
>   Because once -in1new is used no energy searching will be done anymore.
>   am i right?

Yes. But you continue to use the linearization energies that were  
determined for the small k-mesh, and if that k-mesh was really small,  
they might not be optimal.

>   In -in1new N, N means new .in1 file after N iteration.  What are   
> the deciding factors of N?  or how to choose the N value?

If your starting density is very crude (the one produced by dstart,  
for instance), then this would lead to very bad linearization  
energies, which in turn will lead to a very bad density in the second  
iteration, etc. -- such a calculation risks to diverge. Therefore, it  
is best to take N rather large if you start from dstart (say N=5).  
Once the density is reasonably converged, there is no objection  
against taking N=1. What I usually do is to go to rough convergence by  
the default case.in1 (e.g. -cc 0.01), and switch on -in1new 1 on at  
that point:

run -cc 0.01
save_lapw -a -f first-iterations
run -cc 0.00001 -in1new 1

Stefaan

Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm