[Wien] RKMAX and k-mesh

[Stefaan Cottenier]

With some delay, my contribution to this question:

Do not use the bcc-Fe example as given in the LAPW tutorial as a regular
procedure, this was just a didactical example. You can NEVER converge total
energies with respect to RKmax, as the variational principle tells you that
the total energy will always go down if you increase RKmax. If you search
for the point where the total energy goes up again (which is against the
variational principle) then you determine the basis set size for which you
start to run in overcompleteness problems (and that's usually starting from
a value of 10 onwards).

If you want to determine the best RKmax for your case, do not monitor the
total energy but monitor an observable property (:MMI, :EFG, :RTO, :HFF,
...). Observable properties will converge with respect to RKmax (they do not
follow a variational principle). Usually, you will find something in the 6-8
range, much lower than your 10.5.

With respect to the k-mesh question: both RKmax and k-mesh determine the
accuracy. In principle, you should scan both simultaneously, as you never
know which of them is really responsible for not reaching convergence yet.
In practice, take a reasonable k-mesh, determine the best RKmax, and then
vary at fixed RKmax the k-mesh in order to find the best value. Or the other
way around.

Stefaan

----- Original Message ----- 
From: "Sergio L. Palacios Diaz" <sergio at pinon.ccu.uniovi.es>
To: <wien at zeus.theochem.tuwien.ac.at>
Sent: Tuesday, February 22, 2005 4:26 PM
Subject: [Wien] RKMAX and k-mesh


> Hello Wien community !
>
> Just a simple question (I think):
>
> I am running Wien2k for fcc-Ni and performing an analysis similar to
> that done by S. Cottenier for bcc-Fe in his tutorial about LAPW methods.
> The values of the total energy are decreasing for increasing values of
> the RKMAX parameter and from RKMAX=11 the total energy starts to
> increase. So, the optimal value (after S. Cottenier) is RKMAX=10.5
> approx.
> The k-mesh employed consists of 5000 k-points (165 in the irreducible
> part).
> I am wondering the following questions:
> Is there something wrong in using such an unusual RKMAX value (the
> recommended values are from 5 to 9)?
> Is the effect of decreasing RKMAX similar to that of increasing the
> k-mesh?
>
> Any hint will be welcome.
>
> Regards,
> -- 
> Dr. Sergio L. Palacios
> Profesor Titular de Física Aplicada
> Dpto. de Física--Facultad de Ciencias
> Universidad de Oviedo
> C/ Calvo Sotelo s/n
> Oviedo 33007
> Spain
> Phone: +34 98 5 10 28 48
> Fax: +34 98 5 10 33 24
> e-mail: sergio at pinon.ccu.uniovi.es; slpalacios at uniovi.es
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
>